CN111324819B - A method, device, computer equipment and storage medium for searching media content - Google Patents

Abstract

The present disclosure provides a method, an apparatus, a computer device and a storage medium for media content search, wherein the method comprises: receiving a search instruction for target media content; determining a first set of candidate media content matching the scene intent of the target media content and a second set of candidate media content matching the entity intent of the target media content based on the search instruction; and sending a search result corresponding to the target media content to the user side based on the first candidate media content set and the second candidate media content set. By adopting the scheme, the media content can be directly searched based on the media content, the user does not need to input text information to search the media content, and the searching efficiency and the searching accuracy of the user can be improved to a certain extent.

Description

A method, device, computer equipment and storage medium for searching media content

technical field

The present disclosure relates to the technical field of media content processing, and in particular, to a method, apparatus, computer device, and storage medium for searching media content.

Background technique

With the development of the Internet, search engines have become one of the indispensable tools for people to surf the Internet. Traditional search engines are text-based, that is, search through text input. In addition, even search engines related to media content (eg, videos) search through textual information such as names, descriptions, introductions, tags, etc. of media content programs.

Usually, in the process of searching for media content, the text information input by the user can be directly identified, and then the identified results can be searched. For example, between the text information input by the user and the media content label information of the candidate media content When they are consistent, the information of the candidate media content can be fed back to the user.

It can be seen that the above-mentioned media content search method relies on the user's accurate text description and text input operation for the media content that is intended to be searched, which will lead to low search efficiency. Media content searches are less accurate.

SUMMARY OF THE INVENTION

The embodiments of the present disclosure provide at least one solution for searching for media content, which automatically searches for similar media content from both entities and scenes in the target media content, and does not need to search for media content through text input, which can improve search efficiency and search efficiency. accuracy, and improve the probability of getting search results that meet the user's intent.

It mainly includes the following aspects:

In a first aspect, the present disclosure provides a method for searching media content, the method comprising:

Receive a search instruction for target media content;

determining, based on the search instruction, a first candidate media content set that matches the scene intent of the target media content, and a second candidate media content set that matches the entity intent of the target media content;

Based on the first candidate media content set and the second candidate media content set, a search result corresponding to the target media content is sent to the user terminal.

In a possible implementation manner, the determining of the first candidate media content set that matches the scene intent of the target media content includes:

determining the target media content feature vector corresponding to the target media content based on the feature information of the target media content in multiple preset dimensions;

By matching the target media content feature vector with each candidate media content feature vector in the scene index library, at least one first candidate media content matching the scene intent of the target media content is determined, and the at least one first candidate media content is determined. A candidate media content constitutes the first candidate media content set.

In a possible implementation, the media content is a video, and the multiple preset dimensions include multiple ones of the following dimensions:

Visual dimension, textual information dimension, musical dimension.

In a possible implementation manner, the scene index library is generated according to the following steps:

Perform scene intention identification on each primary selection media content, and determine the first candidate media content with scene intention in each primary selection media content;

extracting feature information of the first candidate media content under multiple preset dimensions;

generating a candidate media content feature vector of the first candidate media content based on feature information of the first candidate media content in multiple preset dimensions;

The first media content identifier of the first candidate media content and the candidate media content feature vector of the first candidate media content are stored in the scene index library correspondingly.

In a possible implementation manner, by matching the target media content feature vector with each candidate media content feature vector in the scene index library, determine at least one first target media content that matches the scene intent of the target media content. Candidate media content, including:

Matching the target media content feature vector of the target media content with each candidate media content feature vector in the scene index library, and determining at least one candidate media content feature vector that matches the target media content feature vector;

The at least one first candidate media content is determined based on the first media content identifier corresponding to the determined feature vector of the candidate media content in the scene index library.

In a possible implementation manner, forming the at least one first candidate media content into the first candidate media content set includes:

acquiring user behavior information corresponding to the first candidate media content identified by the first media content identifier;

Based on the user behavior information, first candidate media content that meets a preset requirement is selected from the first candidate media content identified by the first media content identifier to form the first candidate media content set.

In a possible implementation manner, determining the second candidate media content set that matches the entity intent of the target media content includes:

Determine the target entity feature vector of the target entity corresponding to the entity intent in the target media content;

Match the target entity feature vector of the target entity with each candidate entity feature vector in the entity index library, and determine a second candidate media content set that matches the target entity; the second candidate media content set contains at least A second candidate media content.

In a possible implementation manner, the determining of the target entity feature vector of the target entity corresponding to the entity intent in the target media content includes:

detecting an entity in the target media content;

Perform intent recognition on the detected at least one entity, and determine at least one of the target entities;

For each target entity, the target entity feature vector corresponding to the target entity is generated based on the image information corresponding to the target entity in the target media content.

In a possible implementation manner, the entity index library is generated according to the following steps:

determine the entities included in each primary media content;

Perform intent identification on entities included in the primary selection media content, and determine a candidate entity and a second candidate media content where the candidate entity is located;

generating the candidate entity feature vector corresponding to the candidate entity based on the image information of the candidate entity in the corresponding second candidate media content;

The second media content identifier of the second candidate media content where the candidate entity is located and the candidate entity feature vector of the candidate entity are correspondingly stored in the entity index library.

In a possible implementation, the target entity feature vector of the target entity is matched with each candidate entity feature vector in the entity index library, and the second candidate media content set that matches the target entity is determined, including: :

Matching the target entity feature vector of the target entity in the target media content with each candidate entity feature vector in the entity index library, and determining at least one candidate entity feature vector that matches the target media content;

The second candidate media content set is determined based on the second media content identifier corresponding to the determined candidate entity feature vector in the entity index library.

In a possible implementation manner, determining the second candidate media content set based on the second media content identifier corresponding to the determined candidate entity feature vector in the entity index library includes:

acquiring user behavior information corresponding to the second candidate media content identified by the second media content identifier;

Based on the user behavior information, a second candidate media content that meets a preset requirement is selected from the second candidate media content identified by the second media content identifier to form the second candidate media content set.

In a possible implementation manner, the sending a search result corresponding to the target media content to the user terminal based on the first candidate media content set and the second candidate media content set includes:

generating first set identification information corresponding to the first candidate media content set, and generating second set identification information respectively corresponding to at least one second candidate media content subset in the second candidate media content set; wherein each The second candidate media content subset corresponds to a candidate entity that matches the entity intent;

Combining the first candidate media content set and the first set identification information corresponding to the first candidate media content set, at least one second candidate media content subset and each second candidate media content in the second candidate media content set The second set identification information respectively corresponding to the content subsets is sent to the user terminal as the search result.

In a possible implementation manner, the first set identification information includes a first thumbnail image and/or first text description information;

The second set identification information includes second thumbnail pictures and/or second text description information.

In a second aspect, the present disclosure also provides a method for searching media content, the method comprising:

Send a search instruction for the target media content to the server;

Receive a search result fed back by the server; the search result includes a first candidate media content set that matches the scene intent of the target media content, and/or a second candidate that matches the entity intent of the target media content collection of media content;

Based on the search results, a search result presentation page is displayed.

In a possible implementation manner, the sending a search instruction for the target media content to the server includes:

In response to the trigger operation of the search button on the target media content screen, send a search instruction for the target media content to the server; or,

In response to the trigger operation of the frame selection button acting on the target media content screen, a search instruction for the frame-selected media content is sent to the server.

In a possible implementation manner, the search result further includes first set identification information corresponding to the first candidate media content set, and at least one second candidate media content subset in the second candidate media content set. respectively corresponding second set identification information; wherein, each of the second candidate media content subsets corresponds to a candidate entity that matches the entity intent;

The displaying a search result display page based on the search result, including:

Based on the search result, displaying a search result display page containing the first set identification information and the second set identification information;

After the displaying the search result display page, the method further includes:

In response to a trigger operation for any set identification information, display candidate media content corresponding to the any set identification information; wherein, the any set identification information is the first set identification information or any of the second sets identification information.

In a possible implementation manner, based on the search result, displaying a search result display page containing the first set of identification information and at least one second set of identification information, including:

Based on the search result, displaying the search result display page including the first search result display area and the second search result display area;

Wherein, the first search result display area includes the first set identification information and the second set identification information; the second search result display area includes a media content list, and the media content list includes the each media content in the first candidate media content set and the second candidate media content set.

In a possible implementation, the method further includes:

After displaying candidate media content corresponding to any set identification information in response to a trigger operation for any set identification information, in response to a sliding trigger operation, switching to display other candidate media content corresponding to other set identification information.

In a third aspect, the present disclosure also provides an apparatus for searching for media content, the apparatus comprising:

an instruction receiving module for receiving a search instruction for the target media content;

a set determining module, configured to determine, based on the search instruction, a first candidate media content set that matches the scene intent of the target media content, and a second candidate media content set that matches the entity intent of the target media content;

A result search module, configured to send a search result corresponding to the target media content to the user terminal based on the first candidate media content set and the second candidate media content set.

In a fourth aspect, the present disclosure also provides an apparatus for searching for media content, the apparatus comprising:

an instruction sending module, configured to send a search instruction for the target media content to the server;

A result receiving module, configured to receive a search result fed back by the server; the search result includes a first candidate media content set that matches the scene intent of the target media content, and/or an entity that matches the target media content an intent-matched second set of candidate media content;

A page display module, configured to display a search result display page based on the search result.

In a fifth aspect, the present disclosure also provides a computer device, comprising: a processor, a memory, and a bus, the memory stores machine-readable instructions executable by the processor, and when the computer device runs, the processor In communication with the memory through a bus, the machine-readable instructions, when executed by the processor, perform the steps of the method for searching media content according to any one of the first aspect and its various embodiments.

In a sixth aspect, the present disclosure also provides a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor when any of the first aspect and its various embodiments are executed. Steps of a method for searching media content.

With the above solution for searching media content, after responding to the search instruction for the target media content, the server can, on the one hand, determine the first candidate media content set that matches the scene intent of the target media content; Then, the first candidate media content set and the second candidate media content set can be returned to the user terminal for viewing as search results in response to the search instruction.

It can be seen that the above solution not only realizes the automatic search of similar media content based on scene intention, but also realizes the automatic search of similar media content based on entity intention. Since the search can be directly based on the media content, the user does not need to input text information to search for the media content. Search can improve the user's search efficiency and search accuracy to a certain extent. At the same time, the above solution can search for media content based on multiple user intentions, which provides a guarantee for satisfying the user's search intention. For example, when the target media content is When it comes to the interactive media content between a cat and its owner, the above solution can not only search for a set of media content related to the target entity of the cat, but also search for a set of media content related to the interactive scene, which covers a variety of users. Possible search intent, improving the comprehensiveness of search results.

In order to make the above-mentioned objects, features and advantages of the present disclosure more obvious and easy to understand, the preferred embodiments are exemplified below, and are described in detail as follows in conjunction with the accompanying drawings.

Description of drawings

In order to explain the technical solutions of the embodiments of the present disclosure more clearly, the following briefly introduces the accompanying drawings required in the embodiments, which are incorporated into the specification and constitute a part of the specification. The drawings illustrate embodiments consistent with the present disclosure, and together with the description serve to explain the technical solutions of the present disclosure. It should be understood that the following drawings only show some embodiments of the present disclosure, and therefore should not be regarded as limiting the scope. Other related figures are obtained from these figures.

FIG. 1 shows a flowchart of a method for searching media content provided by Embodiment 1 of the present disclosure;

FIG. 2 shows a schematic diagram of a search application of the method for searching media content provided by Embodiment 1 of the present disclosure;

3 shows a flowchart of a specific method for determining a first candidate media content set in the media content search method provided by Embodiment 1 of the present disclosure;

4 shows a flowchart of a specific method for generating a scene index library in the method for searching media content provided by Embodiment 1 of the present disclosure;

5 shows a flowchart of a specific method for determining a second candidate media content set in the method for searching media content provided by Embodiment 1 of the present disclosure;

6 shows a flowchart of a specific method for generating a target entity feature vector in the method for searching media content provided by Embodiment 1 of the present disclosure;

7 shows a flowchart of a specific method for generating an entity index library in the method for searching media content provided by Embodiment 1 of the present disclosure;

FIG. 8 shows an application schematic diagram of the method for searching media content provided by Embodiment 1 of the present disclosure;

FIG. 9 shows a flowchart of a method for searching media content provided by Embodiment 2 of the present disclosure;

FIG. 10 shows a schematic diagram of an apparatus for searching media content according to Embodiment 3 of the present disclosure;

FIG. 11 shows a schematic diagram of another apparatus for media content search provided by Embodiment 3 of the present disclosure;

FIG. 12 shows a schematic diagram of a computer device according to Embodiment 4 of the present disclosure;

FIG. 13 shows a schematic diagram of another computer device provided by Embodiment 4 of the present disclosure.

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present disclosure more clear, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only These are some, but not all, embodiments of the present disclosure. The components of the disclosed embodiments generally described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the disclosure provided in the accompanying drawings is not intended to limit the scope of the disclosure as claimed, but is merely representative of selected embodiments of the disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present disclosure.

In the process of searching for media content, it generally relies on the text information entered by the user to describe the media content. This method relies on the user's accurate textual description of the media content to be searched for, which will lead to low search efficiency. , and at the same time, when the user cannot accurately describe the text, the accuracy of the media content search is often low.

Based on the above research, the embodiments of the present disclosure provide at least one solution for media content search, which automatically searches for similar media content from both entities and scenes in the target media content, and does not need to search for media content through text input, which can improve the The search efficiency and search accuracy are improved, and the comprehensiveness of the obtained search results is improved due to the consideration of multiple search intents.

The defects existing in the above solutions are all the results obtained by the inventor after practice and careful research. Therefore, the discovery process of the above problems and the solutions to the above problems proposed by the present disclosure hereinafter should be the inventors Contributions made to this disclosure during the course of this disclosure.

It should be noted that like numerals and letters refer to like items in the following figures, so once an item is defined in one figure, it does not require further definition and explanation in subsequent figures.

In order to facilitate the understanding of this embodiment, the method for searching media content disclosed by the embodiment of the present disclosure is first introduced in detail. The computer equipment can be: terminal equipment or server or other processing equipment, and the terminal equipment can be user equipment (User Equipment, UE), mobile equipment, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (Personal Digital Assistant, PDA), handheld devices, computing devices, in-vehicle devices, wearable devices, etc. In some possible implementations, the method for searching for media content may be implemented by a processor invoking computer-readable instructions stored in a memory.

First, a method for searching media content provided by an embodiment of the present disclosure will be described below by taking the execution subject as the server.

Example 1

Referring to FIG. 1, which is a flowchart of a method for searching media content according to Embodiment 1 of the present disclosure, the method includes steps S101-S103, wherein:

S101, receiving a search instruction for target media content;

S102, based on the search instruction, determine a first candidate media content set that matches the scene intent of the target media content, and a second candidate media content set that matches the entity intent of the target media content;

S103. Based on the first candidate media content set and the second candidate media content set, send a search result corresponding to the target media content to the client.

In order to facilitate understanding of the method for searching media content provided by the embodiments of the present disclosure, a possible application scenario of the method for searching media content is briefly introduced first. After the user initiates the search instruction for the target media content, the client sends the search instruction for the target media content to the server, and the server obtains the candidate media content matching the scene intent of the target media content based on the above method steps, and the target media content The candidate media content that matches the entity intent corresponding to the content, and then, in the search result returned to the client, the above-mentioned candidate media content that matches the scene intent of the target media content and the candidate media content that matches the entity intent corresponding to the target media content can be sent to the client.

In this way, the user terminal can mix and display the candidate media content that matches the scene intent and the candidate media content that matches the entity intent, and can also display the candidate media content that matches the scene intent and the candidate media content that matches the entity intent. .

Wherein, the above-mentioned search instruction may be generated based on the triggering of a relevant search button set on the user terminal, for example, a search button is set on the media content playback page of the user terminal, and the above-mentioned search instruction is generated after the search button is triggered; or It is generated by performing a corresponding search operation on the media content during the playback of the media content, for example, using the search box to select the media content area on the currently playing media content to initiate the above-mentioned search instruction; Triggered by default after entering a certain media content playback page, that is, the media content search is automatically performed after entering the media content playback page. Besides, the above-mentioned search instruction may also be triggered by other manners capable of initiating a media content search instruction, which is not specifically limited in this embodiment of the present disclosure.

The above-mentioned target media content may be one or more frames of pictures, or, it may be a video, or a certain picture frame in the currently playing video, or a certain picture frame in the currently playing video and before and after the picture frame. The picture group formed by several picture frames may also be other media content, which is not specifically limited in this embodiment of the present disclosure. Considering the wide application of video search, a specific example can be given below with video as the target multimedia content.

Based on the response to the search instruction, the server may determine a first set of candidate media content that matches the scene intent of the target media content and a second set of candidate media content that matches the entity intent of the target media content.

The above-mentioned first candidate media content set may be a set of each first candidate media content related to the scene intent in the target media content filtered from the preset media database. In this embodiment of the present disclosure, the target media content can be compared by comparing the target media content The first candidate media content is determined by the comparison result between the corresponding target media content feature vector and the media content feature vector corresponding to each primary selection media content in the preset media database. The above-mentioned second candidate media content set may be a set of each second candidate media content related to the entity intent in the target media content filtered from the preset media database. The second candidate media content is determined by comparing the target entity feature vector of the target entity with the candidate entity feature vector of candidate entities (entities matching the target entity) in each primary media content in the preset media database.

The scene intent in the embodiments of the present disclosure may correspond to classified scenes such as film and television scenes, game scenes, natural scenes, etc., or may be further classified scenes of the above-mentioned scenes, such as outdoor film and television scenes, home film and television scenes, etc., and may also be other scenes with For scenes with emotional attributes, for example, scene styles such as funny, serious, sad, cheerful, etc., the embodiments of the present disclosure may determine corresponding scene intentions based on different application requirements, which are not specifically limited in the embodiments of the present disclosure.

In the embodiment of the present disclosure, the target media content may have a single scene attribute, or may be composed of multiple scene attributes. In order to take into account various scene attributes, the embodiment of the present disclosure may match each scene attribute with the scene intent of the target media content. The matched primary media content is determined as a media element of the first candidate media content set. For example, in the case where it is determined that the scene of the target media content is both funny and serious, the first set of candidate media content may be determined based on the primary selection media content with a high degree of matching with the funny intent, and may also be determined based on the primary selected media content with a high degree of matching with the funny intent, and may also be based on the high degree of matching with the serious intent. The primary selection of media content determines a first set of candidate media content.

The entity intent in the embodiment of the present disclosure may correspond to a target entity in the target media content, and the target entity may be one or multiple. Here, taking a video of human-cat interaction as an example, a person can be used as a target entity, and a cat can also be used as a target entity.

In the embodiment of the present disclosure, after the first candidate media content set and the second candidate media content set are determined, a search result corresponding to the target media content may be sent to the user terminal.

For the first candidate media content set determined by the scene intent of the target media content, the scene intent corresponding to each first candidate media content in the candidate media content set may be one of the scene intents of the target media content or A plurality of scene attributes are related. Therefore, in this embodiment of the present disclosure, in order to facilitate consideration of various scene attributes, a first set of identification information may be corresponding to the first candidate media content set, so that the client can realize and The display of search results related to the scene intent.

Wherein, the first set identification information in the embodiment of the present disclosure may include a first thumbnail image used to indicate a scene, and may also be a first text description content used to describe the scene.

At the same time, for the second candidate media content set determined by the entity intention of the target media content, the entity intention corresponding to each second candidate media content in the candidate media content set may be the same as the target media content in the target media content. Entity-related, therefore, in order to facilitate searching for multiple target entities, in this embodiment of the present disclosure, at least one second candidate media content subset in the second candidate media content set may correspond to a second set of identification information respectively, and each The second candidate media content subset may correspond to a candidate entity that matches the entity's intent, so that the user terminal can display search results related to the entity's intent according to the second set of identification information.

The above-mentioned second set identification information may include a first thumbnail image used to indicate a candidate entity, and may also be a second text description content used to describe the candidate entity.

Here, the user terminal may display media content based on the received search results. The foregoing method for searching media content provided by the embodiments of the present disclosure will be described below with reference to the rendering effect diagrams of the user terminal interface shown in FIGS. 2( a ) to 2 ( c ).

As shown in FIG. 2( a ), the target media content screen (ie, the human-cat interactive screen) presented by the user terminal includes a search button (○). After the user triggers the search button, a search instruction about the target media content can be sent to the server. The server can then determine, based on the search instruction, the first candidate media content set (ie, multiple home videos) corresponding to the target media content, and its corresponding first set identification information (ie, the scene identifier), and the second candidate media content. A second candidate media content subset in the set (that is, multiple cat videos) and a corresponding second set identification information (that is, the cat identification), and another second candidate in the second candidate media content set A subset of media content (ie, multiple person videos) and a corresponding second set of identification information (ie, an identification of a person).

Among them, the above-mentioned scene identification, cat identification, and character identification may adopt the identification method of thumbnail images or the identification method of text description content, and may also combine the above two identification methods for identification display.

From the display of the search results shown in Figure 2(b), the three thumbnail image identifiers of scene image, cat image, and character image are shown as examples, and multiple home videos corresponding to the scene identifiers can be displayed, that is, the display with Search results corresponding to similar scenes. Through the switching operation, a plurality of cat videos (not shown) corresponding to the cat logos are also displayed, and it may also be a character video (not shown) corresponding to the character logos. From the display of the search results shown in Figure 2(c), the three thumbnail image identifiers of scene graph, cat image, and character image and the text description identifiers corresponding to each thumbnail image identifier are shown as examples. The displayed results are the same as those shown in FIG. 2( b ), and are not repeated here.

Here, the search results displayed corresponding to the multiple logos can be switched and displayed based on the sliding operation. For example, the three logos can be switched between the above three logos by sliding left and right, so as to realize the classified display of the corresponding display contents. After confirming the search On the premise of comprehensiveness, ensure the pertinence of the search.

It is worth noting that the specific display method of search results, such as displaying several results in one row, using vertical display or horizontal display, etc., can be selected based on different application requirements, and no specific restrictions are made here.

In order to further meet the user's customized search requirements, the embodiment of the present disclosure may also provide a manual selection button (shown in the upper right corners of FIG. 2(b) and FIG. 2(c) ) while providing the above search results. After selecting the button, you can jump to the target media content screen, so that the user can further select the target media content. For the process of triggering a search instruction based on the selection operation and searching for similar media content according to the search instruction, please refer to the above description. This will not be repeated here.

It is worth noting that the embodiments of the present disclosure can not only support the user's manual selection while displaying the search results, but also directly based on the user's manual selection to initiate a search instruction for the frame-selected media content to the server, so as to realize similar media content. search, the specific process will not be repeated here.

In addition, the embodiments of the present disclosure can not only support the above-mentioned classification and display methods, but also combine the first candidate media content set with the second candidate media content set, and then perform the combined display, that is, the indiscrimination is based on entity intent The searched similar media content is still the similar media content searched based on the scene intent, and other display manners may also be used, which are not specifically limited in this embodiment of the present disclosure.

In the embodiment of the present disclosure, the determination of the first candidate media content set and the second candidate media content set is a key step in searching for similar media content, which can be described separately below:

For the determination of the first candidate media content set, the embodiment of the present disclosure may be determined according to the matching result between the target media content feature vector corresponding to the target media content and each candidate media content feature vector in the scene index library, as shown in FIG. 3, the above-mentioned method for determining the first candidate media content set specifically includes the following steps:

S301, based on the feature information of the target media content under multiple preset dimensions, determine a target media content feature vector corresponding to the target media content;

S302: Determine at least one first candidate media content that matches the scene intent of the target media content by matching the target media content feature vector with each candidate media content feature vector in the scene index library, and match the at least one first candidate media content A first candidate media content set is formed.

Here, the above-mentioned feature vector of the target media content may be a feature vector extracted for the scene intent of the target media content. For the scene intention of the target media content, it may be related to the feature information under various preset dimensions. Taking video as media content as an example, the above-mentioned preset dimension may be a visual dimension related to the scene attention. The dimension can represent the relevant information of the application scene, such as information such as outdoor or home environment; it can also be the text information dimension related to the description of the media content; it can also be the music dimension related to the scene atmosphere, such as sad music or information such as cheerful music; it may also be other preset dimensions related to the scene intent, which are not specifically limited in this embodiment of the present disclosure.

In the embodiment of the present disclosure, the corresponding feature vector of the target media content can be determined directly based on the feature information of the target media content in the above-mentioned various preset dimensions, that is, the feature information in the various preset dimensions can be spliced together. Obtain the target media content feature vector of the target media content. For example, when there are three preset dimensions, which are visual dimension, text information dimension and music dimension, one preset dimension can correspond to one feature value or one feature vector, for example, the text information dimension can correspond to one feature vector ( For example, converting text into text vector), and the music dimension can correspond to a feature value (for example, the feature value corresponding to cheerful music is 1, and the feature value corresponding to sad music is 0). Here, the feature information corresponding to each preset dimension is processed. After splicing, the feature vector of the target media content can be obtained.

After the target media content feature vector is determined, at least one first candidate media content that matches the scene intent of the target media content can be determined by matching the target media content feature vector with each candidate media content feature vector in the scene index library , the first candidate media content set can be determined based on each first candidate media content.

In the embodiment of the present disclosure, as shown in FIG. 4 , the scene index library can be generated according to the following steps:

S401, performing scene intention identification on each primary selection media content, and determining a first candidate media content with scene intention in each primary selection media content;

S402, extracting feature information of the first candidate media content under multiple preset dimensions;

S403, based on the feature information of the first candidate media content in multiple preset dimensions, generate a candidate media content feature vector of the first candidate media content;

S404: Store the first media content identifier of the first candidate media content and the candidate media content feature vector of the first candidate media content in the scene index library correspondingly.

The manner of determining the feature vector of each candidate media content in the above scene index library is similar to the manner of determining the feature vector of the target media content, that is, in the embodiment of the present disclosure, it also needs to be determined based on feature information of various preset dimensions Candidate media content feature vector.

The difference is that the relative target media content feature vector can be done online, and the determination of the candidate media content feature vector can be done offline. In the embodiment of the present disclosure, before extracting feature information of various preset dimensions, it is necessary to first identify the scene intent of each media content in the media content library, so as to filter out the first candidate media content that matches the scene intent . In this way, feature information of various preset dimensions can be extracted for the screened media content to determine a candidate media content feature vector of the first candidate media content according to the extracted feature information in multiple preset dimensions. For the process of determining the feature vector of the candidate media content, please refer to the relevant description of the feature vector of the target media content, which will not be repeated here.

Wherein, the embodiment of the present disclosure can realize the screening of the first candidate media content matching the scene intent based on the scene intent model. The scene intent model here can be pre-trained, and can be trained based on each primary media content in the media content library and its corresponding scene intent annotation information. In this way, the media content library can be determined based on the trained scene intent model. Whether each primary media content in has a scene intent.

In a specific application, if the primary selection video is used as the primary selection media content, the information related to the scene intent recognition in the primary selection video can be extracted first, and then the extracted information related to the scene intent recognition can be used as the scene intent model to be trained. Input features, and use the above-mentioned scene intent annotation information as the output result of the scene intent model to be trained to train model parameters of the scene intent model.

Among them, the information related to the scene intention selected in the embodiment of the present disclosure may include the authorized behavior characteristics of the video user corresponding to the primary selection video, such as interactive statistical indicators such as video clicks, video likes, etc.; the primary selection video may also be extracted. The corresponding video text feature, such as the title text of the video, utilizes optical character recognition (Optical Character Recognition, OCR), automatic speech recognition (Automatic Speech Recognition, ASR) technology recognizes the text information; can also include video visual features, such as Picture feature information after video frame extraction. The above behavioral features can represent the popularity of the video to a certain extent, the video text features can represent the content information related to the scene description to a certain extent, and the visual features can represent the detailed information of the pictures in the video. It can represent a user's scene intention for the currently annotated primary media content.

Based on the scene intent recognition operation and the feature vector extraction operation of the media content library, candidate media content feature vectors that can belong to the scene index library can be determined. For the scene index library, the embodiment of the present disclosure can not only determine each candidate media content feature vector, but also The source of the media content of the candidate media content feature vector is determined, that is, a corresponding relationship between the candidate media content feature vector and the media content identifier can be established, and the corresponding relationship should be stored in the scene index database. In this way, after the target media content feature vector is matched with each candidate media content feature vector, the first candidate media content whose vector matching result meets the preset requirement can be formed into a first candidate media content set.

In the embodiment of the present disclosure, the determination of the vector matching result can be determined by calculating the cosine similarity between the target media content feature vector of the target media content and each candidate media content feature vector in the scene index library, and the second is Determined based on a scene correlation model.

For the former, in this embodiment of the present disclosure, at least one candidate media content feature vector that matches the target media content feature vector may be determined based on the cosine similarity. In this way, the candidate media that meets the preset requirements (eg, the cosine similarity is greater than 0.8) may be selected. The first candidate media content corresponding to the content feature vector is classified into the first candidate media content set.

For the latter, in this embodiment of the present disclosure, the target media content feature vector of the target media content and each candidate media content feature vector in the scene index library may be input into the scene correlation model for processing to obtain at least one feature vector matching the target media content feature vector. Candidate media content feature vector.

The above scene correlation model may be pre-trained based on the training sample media content marked with the scene correlation matching result. In a specific application, training can be performed based on any two training sample media contents in the training media content library and the relevant scene correlation matching results set for the two training sample media contents. Here, two media contents with relatively high correlation between two scenes can be extracted for media content feature vectors, and then a set of corresponding media content feature vectors can be used as a set of training samples for training. The parameters of the scene correlation model are adjusted according to the comparison results between the correlation results of the scene and the marked scene correlation. Similarly, the two media contents with low correlation between the two scenes can also be used as a set of training samples for training. In order to adjust the parameters, the trained scene correlation model can be obtained when the model training cut-off condition is reached.

In this way, the target media content feature vector and each candidate media content feature vector in the scene index library are input into the trained scene correlation model, and the target media content feature vector and each candidate media content feature vector in the scene index library can be determined. match between. In the embodiment of the present disclosure, the higher the correlation between the two media content feature vectors, the better the matching degree to a certain extent. For the first candidate media content with a preset matching degree, for example, the primary selected media content corresponding to the feature vector of the candidate media content with a matching degree of 0.75 or more may be determined as the matching first candidate media content.

In this embodiment of the present disclosure, whether the candidate media content feature vector matching the target media content feature vector is determined based on the cosine similarity or the scene correlation model, when determining the first candidate media content feature vector corresponding to the matching candidate media content feature vector content, the first media content identifier corresponding to the matched candidate media content feature vector can be determined based on the correspondence between the relevant candidate media content feature vectors and the media content identifiers stored in the scene index library. The first media content identifier can search for the corresponding first candidate media content from the media content library.

Here, after the matching first candidate media content is determined, the first candidate media content obtained by matching may be directly combined to generate a first candidate media content set, and the first candidate media content may also be filtered based on user behavior information and then combined into the first candidate media content. A collection of candidate media content.

In this embodiment of the present disclosure, based on the user behavior information corresponding to the first candidate media content identified by the first media content identifier, the first candidate media content that meets the preset requirements may be selected from at least one matched first candidate media content as a media content element in the first candidate media content set.

Wherein, the above-mentioned user behavior information may be information such as the amount of likes, playback, and forwarding of the first candidate media content, which to a certain extent indicates the degree of attention of each user to the first candidate media content, so as to further satisfy the The search demand of users increases the traffic of media content search.

After the user behavior information is determined, the first candidate media content that meets the preset requirements may be determined as the first candidate media content set. For example, when the user behavior information is the amount of media content likes and the amount of media content playback, the amount of media content likes may be greater than the preset amount of likes (eg, greater than 50), and the amount of media content playback may be greater than the preset amount of playback (eg The first candidate media content greater than 35) determines the first candidate media content set.

In addition, in the embodiment of the present disclosure, after ranking at least one matched first candidate media content according to the user behavior information, the first candidate media content of the preset ranking may be selected as the first candidate media content in the first candidate media content set. media content elements. Here, the amount of media content likes and the amount of media content played are still used as user behavior information for description, and the first candidate media content set may be determined based on the top 20 first candidate media contents. This is mainly because the first candidate media content set determined by the embodiment of the present disclosure needs to be displayed on the user terminal. Therefore, a ranking display method can be adopted here. In this way, when it is pushed to the user terminal for display, it can be displayed based on the above ranking results. Display of media content.

For the determination of the second candidate media content set, the embodiment of the present disclosure may be determined according to the matching result between the target entity feature vector corresponding to the target media content and each candidate entity feature vector in the entity index library, as shown in FIG. 5 . As shown, the above-mentioned method for determining the second candidate media content set specifically includes the following steps:

S501. Determine the target entity feature vector of the target entity corresponding to the entity intent in the target media content;

S502: Match the target entity feature vector of the target entity with each candidate entity feature vector in the entity index library to determine a second candidate media content set that matches the target entity; the second candidate media content set includes at least one second candidate media content.

Here, the feature vector of the target entity in the embodiment of the present disclosure may be the feature vector of the target entity determined after the intent recognition of the entity in the target media content, that is, the feature vector of the target entity in the embodiment of the present disclosure corresponds to The target entity corresponding to the entity intent, the target entity may be part or all of the entities in the target media content. This is mainly considering that for the target media content, as a multimedia element that can cover sounds and images, it may contain various entities. However, entities that do not meet the intent of the entity will not only search for media content itself The amount of calculation is increased, and the viewing traffic of the search results will be reduced. Based on this, in the embodiment of the present disclosure, the intent of the target media content can be identified, and then the target entity feature vector of the target entity corresponding to the entity intent can be determined, as shown in FIG. 6 . , the above determination of the feature vector of the target entity can be achieved by the following steps:

S601. Detecting an entity in the target media content;

S602. Perform intent recognition on the detected at least one entity to determine at least one target entity;

S603. For each target entity, generate a target entity feature vector corresponding to the target entity based on image information corresponding to the target entity in the target media content.

Here, each entity in the target media content can be detected first, and then the intent recognition can be performed on each detected entity to determine the target entity from the various entities. Finally, for the target entity, a target entity feature vector corresponding to the target entity may be generated based on the image information corresponding to the target entity in the target media content.

Here, still taking the video as the media content as an example, the entity in the above-mentioned target media content may be the entity in the current picture frame selected by the user based on the trigger of the search instruction when watching the target video, or the entity in the current picture frame based on the trigger of the search instruction. The selected entity in the current picture frame or several picture frames before and after the picture frame may also be an entity in each picture frame in the entire target video. The entity in the embodiment of the present disclosure can be understood as the foreground target of the target video. At this time, the optical flow method, the frame difference method, the background difference method, etc. can be used to detect the foreground target, and the machine learning method can also be used to realize the detection of the foreground target. , which is not specifically limited in the embodiments of the present disclosure.

After each entity is detected, the entity can be input into the pre-trained entity intent model to determine the target entity corresponding to the user's intent, and then based on the corresponding image information of the target entity in the target media content, the corresponding Target entity feature vector.

Wherein, the above-mentioned target entity corresponding to the entity intent may be an entity determined directly based on a triggering operation of a frame selection button that initiates a search instruction, or an entity determined based on an entity intent model. The entity intent model here can be pre-trained, that is, it can be trained based on each primary selected media content in the media content library and its corresponding entity intent annotation information, so that it can be determined based on the trained entity intent model. Whether each media content in the media content library has a target entity.

In a specific application, if the primary selection video is used as the primary selection media content, the information related to entity intent recognition in the primary selection video can be extracted first, and then the extracted information related to entity intent recognition can be used as the information of the entity intent model to be trained. Input features, and use the entity intent annotation information as the output result of the entity intent model to be trained to train model parameters of the entity intent model.

Wherein, the information related to the entity intent selected in the embodiment of the present disclosure may include the authorized behavior characteristics of the video user corresponding to the primary selection video, such as interactive statistical indicators such as video clicks, video likes, etc.; the primary selection video may also be extracted. The corresponding video text features, such as the title text of the video, the text information identified by OCR technology and ASR technology; it can also be the visual features of the primary video, such as the picture feature information after the video frame is extracted; it can also include the initial video. Visual features of the entities in the selected video; it can also include the relationship between the entities in the primary video and the video, such as the area the entity occupies in the video picture frame, the location of the entity in the video picture frame, the frequency of the entity in the video, etc. information. The above information can definitely represent the entity intention of a user for the currently marked primary selected media content.

In the embodiment of the present disclosure, a target entity may correspond to a media content object. Here, taking a target media content interacting with a human and a cat as an example, a human can be used as a video object, and a cat can also be used as a video object. After the entity intent annotation is performed, it can be determined whether the target entity of the cat exists in the target media content.

The image information corresponding to the target entity in the target media content may be the image position information where the target entity is located, the entity motion information determined based on multiple picture frames where the target entity is located, or other information that can characterize the target entity. The image information of the entity, which is not specifically limited in this embodiment of the present disclosure. In the embodiment of the present disclosure, the above-mentioned various image information is represented in the form of a vector (for example, input into a trained feature extraction network), the target entity feature vector corresponding to the target entity can be obtained, for example, for 16* 16 picture frames, you can use 10111 to indicate that the center of gravity of the target entity is located in the second row and third column of the picture frame, that is, the position information can be vectorized. Similarly, other image information can also be vectorized. The disclosed embodiments are not described in detail here.

The description of the value is that the embodiment of the present disclosure may determine one or more target entities for the target media content, and for each target entity, its corresponding target entity feature vector may be determined based on the above method.

After the target entity feature vector is determined, the second candidate media content set matching the target entity is determined by matching the target entity feature vector of the target entity with each candidate entity feature vector in the entity index library.

In the embodiment of the present disclosure, as shown in FIG. 7 , the entity index library can be generated according to the following steps:

S701. Determine entities included in each primary selection media content;

S702, performing intention identification on entities included in the primary selection media content, and determining a candidate entity and a second candidate media content where the candidate entity is located;

S703, based on the image information of the candidate entity in the corresponding second candidate media content, generate a candidate entity feature vector corresponding to the candidate entity;

S704: Store the second media content identifier of the second candidate media content where the candidate entity is located, and the candidate entity feature vector of the candidate entity in the entity index library correspondingly.

The manner of determining the feature vectors of each candidate entity in the entity index library is similar to the manner of determining the feature vector of the target entity. The determined candidate entities are extracted for feature vector.

The difference is that, compared with the target entity feature vector, which can be done online, the determination of the candidate entity feature vector can be done offline. In the embodiment of the present disclosure, before performing entity intent recognition, frame extraction processing may be performed on each primary selected media content in the media content library. A candidate entity matching the entity's intent and the second candidate media content where the candidate entity is located are screened out. In this way, the corresponding candidate entity feature vector can be determined based on the image information of the candidate entity in the corresponding second candidate media content. For the determination process of the candidate entity feature vector, please refer to the above-mentioned relevant description of the target entity feature vector, which will not be repeated here.

Among them, the embodiment of the present disclosure can realize the screening of the second candidate media content that matches the entity's intent based on the entity model. For the entity intent model here, refer to the above description for details, and details are not repeated here.

Based on the entity intent recognition operation and the feature vector extraction operation of the above media content library, the candidate entity feature vectors that can belong to the entity index library can be determined. For the entity index library, the embodiment of the present disclosure can not only determine each candidate entity feature vector, but also determine the entity index library. The source of the media content of the candidate entity feature vector, that is, the corresponding relationship between the candidate entity feature vector of the candidate entity and the second media content identifier can be established, and the corresponding relationship should be stored in the scene index database. In this way, after the target entity feature vector is matched with each candidate entity feature vector, the second candidate media content whose vector matching result meets the preset requirement can be formed into a second candidate media content set.

In the embodiment of the present disclosure, the determination of the vector matching result can be determined by calculating the cosine similarity between the target entity feature vector of the target entity and each candidate entity feature vector in the entity index library, and the second is based on entity correlation sex model to determine.

For the former, in this embodiment of the present disclosure, at least one candidate entity feature vector that matches the target entity feature vector of the target entity may be determined based on the cosine similarity. In this way, the candidate entity that meets the preset requirements (eg, the cosine similarity is greater than 0.8) can be selected. The second candidate media content corresponding to the entity feature vector is classified into the second candidate media content set.

For the latter, in this embodiment of the present disclosure, the target entity feature vector of the target entity and each candidate entity feature vector in the entity index library can be input into the entity correlation model for processing to obtain at least one candidate entity feature vector matching the target entity feature vector. .

The above entity correlation model may be obtained by pre-training, and may be obtained by training based on the training sample media content of the standard good entity correlation matching result. In a specific application, training may be performed based on entities in any training sample media content in the training media content library and the related entity correlation matching results set for the two training sample media contents. Here, the entity feature vector can be extracted from the two media contents with high correlation between the two entities, and then the corresponding set of entity feature vectors can be used as a set of training samples for training. The comparison results between the marked entity correlation matching results are used to adjust the parameters of the entity correlation model. Similarly, two media contents with relatively low entity correlations can also be used as a set of training samples for training. The parameters are adjusted, so that the trained entity correlation model can be obtained when the model training cut-off condition is reached.

In this way, by inputting the target entity feature vector and each candidate entity feature vector in the entity index library into the trained entity correlation model, the matching between the target entity feature vector and each candidate entity feature vector in the entity index library can be determined. Spend. Referring to the relevant description of the above scene correlation model, the second candidate media content may also be determined based on the matching degree here.

In the embodiment of the present disclosure, whether the candidate entity feature vector matching the target entity feature vector is determined based on the cosine similarity or the entity correlation model, when determining the second candidate media content corresponding to the matching candidate entity feature vector, All can be based on the corresponding relationship between the relevant candidate entity feature vector and the media content identifier stored in the entity index library, to determine the second media content identifier corresponding to the candidate entity feature vector obtained by matching. In this way, based on the second media content identifier That is, the corresponding second candidate media content can be searched from the media content library.

Here, after the matching second candidate media content is determined, the second candidate media content obtained by matching can be directly combined to generate a second candidate media content set, and the user behavior information authorized by the user can also be filtered first and then combined. into a second candidate media content set.

In this embodiment of the present disclosure, based on user behavior information corresponding to the second candidate media content identified by the second media content identifier, a second candidate media content that meets preset requirements may be selected from at least one matched second candidate media content as a media content element in the second candidate media content set.

The method for determining the user behavior information corresponding to the second candidate media content is the same as the method for determining the user behavior information corresponding to the first candidate media content, and details are not described herein again. In addition, the method for determining the second candidate media content set based on the user behavior information is similar to the method for determining the first candidate media content set, that is, it can be determined based on either a preset requirement or a ranking result, which is not repeated here. Repeat.

Considering that the second candidate media content set determined by the embodiment of the present disclosure needs to be displayed on the user end, a ranking display method may be adopted here. In this way, when the set of media content is pushed to the user end for display, the media content may also be displayed based on the above ranking results. 's display.

In order to facilitate further understanding of the foregoing media content search method provided by the embodiments of the present disclosure, the foregoing media content search method may be described with reference to the application diagram shown in FIG. 8 .

As shown in FIG. 8 , the above-mentioned method for searching for media content can be implemented by an entity offline module, a scene offline module and an online module. Here, video is used as an example of media content for description.

Among them, the entity offline module can extract the picture frame based on the frame extraction operation of each video in the video library, and then can determine the candidate entity through entity detection and entity intent model. In this way, based on the image information of the target entity in each video After the candidate entity feature vectors are determined, the determined candidate entity feature vectors are stored in the entity index library. Among them, the corresponding relationship between each candidate entity feature vector and each video identifier that can be stored in the entity index library, such as the candidate entity feature vector 1 and video 21 shown in the entity index library in FIG. 21) The correspondence between the candidate entity feature vector 2 and the video 22 (ie, vector 2->video 22), and the candidate entity feature vector 3 and video 23 (ie, vector 3->video 23). At the same time, user behavior information corresponding to each video can also be recorded to prepare for subsequent entity sorting.

In addition, the above-mentioned scene offline module can input each video in the video library into the scene intent model to determine the video that matches the scene intent, through the extraction of feature vectors under the preset dimensions of visual features, text information features, and music features, Corresponding candidate video feature vectors may be determined, and each determined candidate video feature vector may be stored in a scene index library. Among them, the corresponding relationship between each candidate video feature vector and each video identifier that can be stored in the scene index library, such as the candidate video feature vector 1 and video 11 shown in the scene index library in FIG. 8 (that is, vector 1 -> video 11) Correspondence between candidate video feature vector 2 and video 12 (ie, vector 2->video 12), and candidate video feature vector 3 and video 13 (ie, vector 3->video 13). At the same time, user behavior information corresponding to each video can also be recorded to prepare for subsequent scene sorting.

For the online module, correlation processing can be performed in two ways, one way is to determine the first candidate video set based on the scene intent, and the other way is to determine the second candidate video set based on the entity intent.

For the determination of the first candidate video set, the target video feature vector under the preset dimensions of visual features, text information features, and music features can be extracted first, and the spliced target video feature vectors are input into the scene correlation model, That is, the target video feature vector can be matched with the candidate video feature vector in the scene index library to obtain the matched first candidate video. At this time, the matched first candidate videos can be sorted based on the scene sorting model and the user behavior information corresponding to the recorded videos, and the first candidate videos that meet the preset ranking can be screened out as the final first candidate videos. gather.

For the determination of the second candidate video set, first, entity detection and intent recognition under the entity intent model can be performed based on the picture frame clicked by the user. As shown in Figure 8, a target entity (only a specific example), where the corresponding target entity feature vector can be determined. Based on the entity correlation model, the target entity feature vector can be matched with the candidate entity feature vector in the entity index library to obtain a matched second candidate video. At this time, the matched second candidate videos can be sorted based on the entity sorting model and the user behavior information corresponding to the recorded videos, and the second candidate videos that meet the preset ranking can be screened out as the final second candidate videos. gather.

Next, a method for searching media content provided by an embodiment of the present disclosure will be described with the execution subject as the client.

Embodiment 2

Referring to FIG. 9, which is a flowchart of a method for searching media content according to Embodiment 2 of the present disclosure, the method includes steps S901-S903, wherein:

S901, sending a search instruction for target media content to a server;

S902, receive the search result fed back by the server; the search result includes a first candidate media content set that matches the scene intent of the target media content, and/or a second candidate media content set that matches the entity intent of the target media content;

S903 , based on the search result, display a search result display page.

Here, a search instruction for the target media content may be sent to the server first, and then a search result determined by the server according to the media content search method shown in Embodiment 1 may be received. The first candidate media content set and the second candidate media content set matching the entity intent of the target media content are used to display the search result display page.

Wherein, the above-mentioned search instruction may be initiated by the user terminal to the server in response to the trigger operation of the search button on the target media content screen, or in response to the trigger operation of the frame selection button acting on the target media content screen. For the specific initiation process of the search instruction and the above-mentioned related descriptions about the triggering operations of the search button and the frame selection button, please refer to the application schematic diagrams involved in FIGS. Repeat.

In order to facilitate taking into account the attributes of each scene and realize the search of multiple target entities, the search result in the embodiment of the present disclosure may further include a first set identification information corresponding to the first candidate media content set, and the second candidate media content set in the At least one second candidate media content subset corresponds to one second set identification information respectively. For a specific description, please refer to the relevant description in Embodiment 1, which will not be repeated here.

Here, the user terminal can receive the above-mentioned search results, and display the search result display page including the first set identification information and the second set identification information. In this way, when the user triggers any set of identification information, it can display and display any set of identification information. The candidate media content corresponding to the set identification information.

Still taking the application diagram shown in FIG. 2( b ) as an example, when the scene marker is triggered, multiple home videos corresponding to the marker can be displayed, that is, search results corresponding to similar scenes can be displayed. Similarly, when other identifiers are triggered, the corresponding candidate media content (not shown) can be switched and displayed.

In the embodiment of the present disclosure, the candidate media content displayed corresponding to the multiple logos can be switched and displayed based on the sliding operation. Still in FIG. 2(b), the three logos can be switched by sliding left and right, so as to realize the display of the candidate media content. The classified display of the corresponding display content ensures the pertinence of the search under the premise of determining the comprehensiveness of the search.

In addition, the embodiments of the present disclosure can not only support the above-mentioned classification display methods, but also combine the first candidate media content set and the second candidate media content set, and then perform combined display. , display the first search result display area containing the first set identification information and the second set identification information on the search result display page, and the media including each media content in the first candidate media content set and the second candidate media content set The second search result display area of the content list can make no distinction between similar media content searched based on entity intent or similar media content searched based on scene intent.

It should be noted that the embodiments of the present disclosure may also adopt other display manners, which are not specifically limited in the embodiments of the present disclosure.

Those skilled in the art can understand that in the above method of the specific implementation, the writing order of each step does not mean a strict execution order but constitutes any limitation on the implementation process, and the specific execution order of each step should be based on its function and possible Internal logic is determined.

Based on the same inventive concept, an apparatus for searching media content corresponding to the method for searching media content is also provided in the embodiment of the present disclosure, because the principle of solving the problem by the apparatus in the embodiment of the present disclosure is the same as the above-mentioned method for searching media content in the embodiment of the present disclosure. Similar, therefore, the implementation of the apparatus may refer to the implementation of the method, and repeated descriptions will not be repeated.

Embodiment 3

Referring to FIG. 10, which is a schematic diagram of an apparatus for searching media content according to Embodiment 3 of the present disclosure, the apparatus includes: an instruction receiving module 1001, a set determining module 1002, and a result searching module 1003; wherein,

an instruction receiving module 1001, configured to receive a search instruction for target media content;

a set determination module 1002, configured to determine, based on the search instruction, a first candidate media content set that matches the scene intent of the target media content, and a second candidate media content set that matches the entity intent of the target media content;

The result search module 1003 is configured to send a search result corresponding to the target media content to the user terminal based on the first candidate media content set and the second candidate media content set.

The embodiments of the present disclosure perform automatic search for similar media content from both entities and scenes in the target media content, and do not need to search for media content through text input, which can improve search efficiency and search accuracy, and improve the ability to meet user intent. the probability of a search result.

In a possible implementation, the set determination module 1002 is configured to determine the first candidate media content set that matches the scene intent of the target media content according to the following steps:

Determine the target media content feature vector corresponding to the target media content based on the feature information of the target media content in multiple preset dimensions;

By matching the target media content feature vector with each candidate media content feature vector in the scene index library, at least one first candidate media content that matches the scene intent of the target media content is determined, and the at least one first candidate media content is composed of the first candidate media content. A set of candidate media content.

In a possible implementation, the media content is a video, and the multiple preset dimensions include multiple of the following dimensions:

Visual dimension, textual information dimension, musical dimension.

In a possible implementation, the set determination module 1002 is configured to generate a scene index library according to the following steps:

Perform scene intention identification on each primary selection media content, and determine the first candidate media content with scene intention in each primary selection media content;

extracting feature information of the first candidate media content under multiple preset dimensions;

generating a candidate media content feature vector of the first candidate media content based on feature information of the first candidate media content in multiple preset dimensions;

The first media content identifier of the first candidate media content and the candidate media content feature vector of the first candidate media content are correspondingly stored in the scene index library.

In a possible implementation, the set determination module 1002 is configured to determine at least one first candidate media content that matches the scene intent of the target media content according to the following steps:

Matching the target media content feature vector of the target media content with each candidate media content feature vector in the scene index library, and determining at least one candidate media content feature vector that matches the target media content feature vector;

At least one first candidate media content is determined based on the first media content identifier corresponding to the determined feature vector of the candidate media content in the scene index library.

In a possible implementation, the set determination module 1002 is configured to form at least one first candidate media content into a first candidate media content set according to the following steps:

obtaining user behavior information corresponding to the first candidate media content identified by the first media content identifier;

Based on the user behavior information, a first candidate media content set that meets a preset requirement is selected from the first candidate media content identified by the first media content identifier to form a first candidate media content set.

In a possible implementation, the set determination module 1002 is configured to match the target media content feature vector of the target media content with each candidate media content feature vector in the scene index library according to the following steps:

Inputting the target media content feature vector of the target media content and each candidate media content feature vector in the scene index library into the scene correlation model for processing to obtain at least one candidate media content feature vector matching the target media content feature vector; scene correlation The model is trained based on the training sample media content marked with the scene relevance matching results; or,

At least one candidate media content feature vector matching the target media content feature vector is determined by calculating the cosine similarity between the target media content feature vector of the target media content and each candidate media content feature vector in the scene index library.

In a possible implementation manner, the set determination module 1002 is configured to determine the second candidate media content set that matches the entity intent of the target media content according to the following steps:

Determine the target entity feature vector of the target entity corresponding to the entity intent in the target media content;

Matching the target entity feature vector of the target entity with each candidate entity feature vector in the entity index library to determine a second candidate media content set that matches the target entity; the second candidate media content set includes at least one second candidate media content .

In a possible implementation, the set determination module 1002 is configured to determine the target entity feature vector of the target entity corresponding to the entity intent in the target media content according to the following steps:

detecting entities in the target media content;

Perform intent recognition on the detected at least one entity, and determine at least one target entity;

For each target entity, a target entity feature vector corresponding to the target entity is generated based on the image information corresponding to the target entity in the target media content.

In a possible implementation, the set determination module 1002 is configured to generate an entity index library according to the following steps:

determine the entities included in each primary media content;

Perform intent recognition on entities included in the primary selection media content, and determine the candidate entity and the second candidate media content where the candidate entity is located;

generating a candidate entity feature vector corresponding to the candidate entity based on the image information of the candidate entity in the corresponding second candidate media content;

The second media content identifier of the second candidate media content where the candidate entity is located and the candidate entity feature vector of the candidate entity are correspondingly stored in the entity index library.

In a possible implementation, the set determination module 1002 is configured to determine the second candidate media content set matching the target entity according to the following steps:

Matching the target entity feature vector of the target entity in the target media content with each candidate entity feature vector in the entity index library, and determining at least one candidate entity feature vector matching the target media content;

Based on the second media content identifier corresponding to the determined candidate entity feature vector in the entity index library, the second candidate media content set is determined.

In a possible implementation manner, the set determination module 1002 is configured to determine the second candidate media content set according to the following steps:

obtaining user behavior information corresponding to the second candidate media content identified by the second media content identifier;

Based on the user behavior information, the second candidate media content that meets the preset requirements is selected from the second candidate media content identified by the second media content identifier to form a second candidate media content set.

In a possible implementation, the set determination module 1002 is configured to match the target entity feature vector of the target entity with each candidate entity feature vector in the entity index library according to the following steps:

The target entity feature vector of the target entity and each candidate entity feature vector in the entity index library are input into the entity correlation model for processing, and at least one candidate entity feature vector matching the target entity feature vector is obtained; The training sample media content of the entity correlation matching result is obtained by training; or,

At least one candidate entity feature vector matching the target entity feature vector is determined by calculating the cosine similarity between the target entity feature vector of the target entity and each candidate entity feature vector in the entity index library.

In a possible implementation, the result search module 1003 is configured to send the search result corresponding to the target media content to the client according to the following steps:

generating first set identification information corresponding to the first candidate media content set, and generating second set identification information respectively corresponding to at least one second candidate media content subset in the second candidate media content set; wherein, each second candidate media The content subset corresponds to a candidate entity that matches the entity's intent;

The first candidate media content set and the first set identification information corresponding to the first candidate media content set, at least one second candidate media content subset and each second candidate media content subset in the second candidate media content set respectively The corresponding second set identification information is sent to the user terminal as the search result.

In a possible implementation manner, the first set identification information includes a first thumbnail image and/or first text description information;

The second set identification information includes second thumbnail images and/or second text description information.

As shown in FIG. 11 , it is a schematic diagram of another apparatus for searching media content according to Embodiment 3 of the present disclosure. The apparatus includes: an instruction sending module 1101 , a result receiving module 1102 , and a page displaying module 1103 ; wherein,

an instruction sending module 1101, configured to send a search instruction for the target media content to the server;

The result receiving module 1102 is configured to receive the search result fed back by the server; the search result includes a first candidate media content set that matches the scene intent of the target media content, and/or a second candidate media that matches the entity intent of the target media content content collection;

The page display module 1103 is configured to display a search result display page based on the search result.

In a possible implementation, the instruction sending module 1101 is configured to send a search instruction for the target media content to the server according to the following steps:

In response to the trigger operation of the search button on the target media content screen, send a search instruction for the target media content to the server; or,

In response to the trigger operation of the frame selection button acting on the target media content screen, a search instruction for the frame-selected media content is sent to the server.

In a possible implementation manner, the search result further includes first set identification information corresponding to the first candidate media content set, and identification information corresponding to at least one second candidate media content subset in the second candidate media content set respectively. The second set of identification information; wherein, each second candidate media content subset corresponds to a candidate entity that matches the entity's intent;

The page display module 1103 is used to display the search result display page according to the following steps:

Based on the search result, displaying a search result display page containing the first set of identification information and the second set of identification information;

The content display module 1104 is configured to display the candidate media content corresponding to any set identification information in response to a trigger operation for any set identification information after displaying the search result display page; wherein, any set identification information is the first set identification information or any second set of identification information.

In one embodiment, the page display module 1103 is configured to display a search result display page containing the first set of identification information and at least one second set of identification information according to the following steps:

Based on the search result, displaying a search result display page including the first search result display area and the second search result display area;

Wherein, the first search result display area includes the first set identification information and the second set identification information; the second search result display area includes a media content list, and the media content list includes the first candidate media content set and the second candidate media Individual media content in a content collection.

In one embodiment, the above device further includes:

The content switching module 1105 is configured to, after displaying the candidate media content corresponding to any set identification information in response to a trigger operation for any set identification information, switch to display other candidates corresponding to other set identification information in response to the sliding trigger operation media content.

For the description of the processing flow of each module in the apparatus and the interaction flow between the modules, reference may be made to the relevant descriptions in the foregoing method embodiments, which will not be described in detail here.

Embodiment 4

An embodiment of the present disclosure also provides a computer device, where the computer device may be a server or a client. When a server is used as a computer device, as shown in FIG. 12 , a schematic structural diagram of a computer device provided by an embodiment of the present disclosure includes: a processor 1201 , a memory 1202 , and a bus 1203 . The memory 1202 stores machine-readable instructions executable by the processor 1201 (in the device shown in FIG. 10, the instructions executed by the instruction receiving module 1001, the set determination module 1002 and the result search module 1003), when the computer device is running, The communication between the processor 1201 and the memory 1202 is through the bus 1203, and the machine-readable instructions are executed by the processor 1201 to perform the following processing:

Receive a search instruction for target media content;

determining, based on the search instruction, a first candidate media content set that matches the scene intent of the target media content, and a second candidate media content set that matches the entity intent of the target media content;

Based on the first candidate media content set and the second candidate media content set, a search result corresponding to the target media content is sent to the user terminal.

In a possible implementation manner, in the instructions executed by the processor 1201, the first candidate media content set that matches the scene intent of the target media content is determined, including:

Determine the target media content feature vector corresponding to the target media content based on the feature information of the target media content in multiple preset dimensions;

By matching the target media content feature vector with each candidate media content feature vector in the scene index library, at least one first candidate media content that matches the scene intent of the target media content is determined, and the at least one first candidate media content is composed of the first candidate media content. A set of candidate media content.

In a possible implementation, the media content is a video, and the multiple preset dimensions include multiple of the following dimensions:

Visual dimension, textual information dimension, musical dimension.

In a possible implementation manner, in the instructions executed by the processor 1201, the scene index library is generated according to the following steps:

Perform scene intention identification on each primary selection media content, and determine the first candidate media content with scene intention in each primary selection media content;

extracting feature information of the first candidate media content under multiple preset dimensions;

generating a candidate media content feature vector of the first candidate media content based on feature information of the first candidate media content in multiple preset dimensions;

The first media content identifier of the first candidate media content and the candidate media content feature vector of the first candidate media content are correspondingly stored in the scene index library.

In a possible implementation, in the instructions executed by the processor 1201, by matching the target media content feature vector with each candidate media content feature vector in the scene index library, it is determined that at least the target media content is matched with the scene intent of the target media content. A first candidate media content, including:

Matching the target media content feature vector of the target media content with each candidate media content feature vector in the scene index library, and determining at least one candidate media content feature vector that matches the target media content feature vector;

At least one first candidate media content is determined based on the first media content identifier corresponding to the determined feature vector of the candidate media content in the scene index library.

In a possible implementation manner, in the instructions executed by the processor 1201, at least one first candidate media content is formed into a first candidate media content set, including:

obtaining user behavior information corresponding to the first candidate media content identified by the first media content identifier;

Based on the user behavior information, a first candidate media content set that meets a preset requirement is selected from the first candidate media content identified by the first media content identifier to form a first candidate media content set.

In a possible implementation, in the instructions executed by the processor 1201, the target media content feature vector of the target media content is matched with each candidate media content feature vector in the scene index library, including:

Inputting the target media content feature vector of the target media content and each candidate media content feature vector in the scene index library into the scene correlation model for processing to obtain at least one candidate media content feature vector matching the target media content feature vector; scene correlation The model is trained based on the training sample media content marked with the scene relevance matching results; or,

At least one candidate media content feature vector matching the target media content feature vector is determined by calculating the cosine similarity between the target media content feature vector of the target media content and each candidate media content feature vector in the scene index library.

In a possible implementation manner, in the instructions executed by the processor 1201, the second candidate media content set that matches the entity intent of the target media content is determined, including:

Determine the target entity feature vector of the target entity corresponding to the entity intent in the target media content;

Matching the target entity feature vector of the target entity with each candidate entity feature vector in the entity index library to determine a second candidate media content set that matches the target entity; the second candidate media content set includes at least one second candidate media content .

In a possible implementation manner, in the instructions executed by the processor 1201, the target entity feature vector of the target entity corresponding to the entity intent in the target media content is determined, including:

detecting entities in the target media content;

Perform intent recognition on the detected at least one entity, and determine at least one target entity;

For each target entity, a target entity feature vector corresponding to the target entity is generated based on the image information corresponding to the target entity in the target media content.

In a possible implementation manner, in the instructions executed by the processor 1201, an entity index library is generated according to the following steps:

determine the entities included in each primary media content;

Perform intent recognition on entities included in the primary selection media content, and determine the candidate entity and the second candidate media content where the candidate entity is located;

generating a candidate entity feature vector corresponding to the candidate entity based on the image information of the candidate entity in the corresponding second candidate media content;

The second media content identifier of the second candidate media content where the candidate entity is located and the candidate entity feature vector of the candidate entity are correspondingly stored in the entity index library.

In a possible implementation, in the instructions executed by the processor 1201, the target entity feature vector of the target entity is matched with each candidate entity feature vector in the entity index library, and the second candidate media content matching the target entity is determined. Collection, including:

Matching the target entity feature vector of the target entity in the target media content with each candidate entity feature vector in the entity index library, and determining at least one candidate entity feature vector matching the target media content;

Based on the second media content identifier corresponding to the determined candidate entity feature vector in the entity index library, the second candidate media content set is determined.

In a possible implementation, in the instruction executed by the processor 1201, based on the second media content identifier corresponding to the determined candidate entity feature vector in the entity index library, the second candidate media content set is determined, including:

obtaining user behavior information corresponding to the second candidate media content identified by the second media content identifier;

Based on the user behavior information, the second candidate media content that meets the preset requirements is selected from the second candidate media content identified by the second media content identifier to form a second candidate media content set.

In a possible implementation manner, in the instructions executed by the processor 1201, the target entity feature vector of the target entity is matched with each candidate entity feature vector in the entity index library, including:

The target entity feature vector of the target entity and each candidate entity feature vector in the entity index library are input into the entity correlation model for processing, and at least one candidate entity feature vector matching the target entity feature vector is obtained; The training sample media content of the entity correlation matching result is obtained by training; or,

At least one candidate entity feature vector matching the target entity feature vector is determined by calculating the cosine similarity between the target entity feature vector of the target entity and each candidate entity feature vector in the entity index library.

In a possible implementation, in the instructions executed by the processor 1201, based on the first candidate media content set and the second candidate media content set, the search results corresponding to the target media content are sent to the client, including:

generating first set identification information corresponding to the first candidate media content set, and generating second set identification information respectively corresponding to at least one second candidate media content subset in the second candidate media content set; wherein, each second candidate media The content subset corresponds to a candidate entity that matches the entity's intent;

The first candidate media content set and the first set identification information corresponding to the first candidate media content set, and at least one second candidate media content subset and each second candidate media content subset in the second candidate media content set The corresponding second set identification information is sent to the user terminal as a search result.

In a possible implementation manner, the first set identification information includes a first thumbnail image and/or first text description information;

The second set identification information includes second thumbnail images and/or second text description information.

When a user terminal is used as a computer device, as shown in FIG. 13 , a schematic structural diagram of a computer device provided by an embodiment of the present disclosure includes: a processor 1301 , a memory 1302 , and a bus 1303 . The memory 1302 stores machine-readable instructions executable by the processor 1301 (in the device shown in FIG. 11, the instructions executed by the instruction sending module 1101, the result receiving module 1102, and the page display module 1103), when the computer device runs, The communication between the processor 1301 and the memory 1302 is through the bus 1303. When the machine-readable instructions are executed by the processor 1301, the following processing is performed:

Send a search instruction for the target media content to the server;

Receive a search result fed back by the server; the search result includes a first candidate media content set that matches the scene intent of the target media content, and/or a second candidate media content set that matches the entity intent of the target media content;

Based on the search results, display the search results display page.

In a possible implementation manner, in the instructions executed by the processor 1301, a search instruction for the target media content is sent to the server, including:

In response to the trigger operation of the search button on the target media content screen, send a search instruction for the target media content to the server; or,

In response to the trigger operation of the frame selection button acting on the target media content screen, a search instruction for the frame-selected media content is sent to the server.

In a possible implementation manner, the search result further includes first set identification information corresponding to the first candidate media content set, and first set identification information corresponding to at least one second candidate media content subset in the second candidate media content set respectively. Two sets of identification information; wherein, each second candidate media content subset corresponds to a candidate entity that matches the entity's intent;

In the instructions executed by the above-mentioned processor 1301, based on the search results, a search result display page is displayed, including:

Based on the search result, displaying a search result display page containing the first set of identification information and the second set of identification information;

After displaying the search result presentation page, the instructions executed by the processor 1301 further include:

In response to a triggering operation for any set of identification information, candidate media content corresponding to any of the set of identification information is displayed; wherein any of the set of identification information is the first set of identification information or any of the second set of identification information.

In a possible implementation, in the instructions executed by the processor 1301, based on the search results, displaying a search result display page containing the first set of identification information and at least one second set of identification information, including:

Based on the search result, displaying a search result display page including the first search result display area and the second search result display area;

Wherein, the first search result display area includes the first set identification information and the second set identification information; the second search result display area includes a media content list, and the media content list includes the first candidate media content set and the second candidate media Individual media content in a content collection.

In a possible implementation manner, the instructions executed by the processor 1301 further include:

After displaying candidate media content corresponding to any set identification information in response to a trigger operation for any set identification information, in response to a sliding trigger operation, switching to display other candidate media content corresponding to other set identification information.

An embodiment of the present disclosure further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is run by a processor, the steps of the method for searching for media content in the first embodiment of the above method are executed or Perform the steps of the method for searching for media content in the second embodiment of the above method. Wherein, the storage medium may be a volatile or non-volatile computer-readable storage medium.

The computer program product of the method for searching media content provided by the embodiments of the present disclosure includes a computer-readable storage medium storing program codes, and the instructions included in the program code can be used to execute the steps of the method for searching media content in the above method embodiments , for details, refer to the foregoing method embodiments, which will not be repeated here.

Embodiments of the present disclosure further provide a computer program, which implements any one of the methods in the foregoing embodiments when the computer program is executed by a processor. The computer program product can be specifically implemented by hardware, software or a combination thereof. In an optional embodiment, the computer program product is embodied as a computer storage medium, and in another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (Software Development Kit, SDK), etc. .

Those skilled in the art can clearly understand that, for the convenience and brevity of description, for the specific working process of the system and device described above, reference may be made to the corresponding process in the foregoing method embodiments, which will not be repeated here. In the several embodiments provided by the present disclosure, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. The apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some communication interfaces, indirect coupling or communication connection of devices or units, which may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.

The functions, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a processor-executable non-volatile computer-readable storage medium. Based on such understanding, the technical solutions of the present disclosure can be embodied in the form of software products in essence, or the parts that contribute to the prior art or the parts of the technical solutions. The computer software products are stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program codes.

Finally, it should be noted that the above-mentioned embodiments are only specific implementations of the present disclosure, and are used to illustrate the technical solutions of the present disclosure, but not to limit them. The protection scope of the present disclosure is not limited to this, although the aforementioned The embodiments describe the present disclosure in detail, and those skilled in the art should understand that: any person skilled in the art can still modify the technical solutions described in the foregoing embodiments within the technical scope disclosed by the present disclosure. Or can easily think of changes, or equivalently replace some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be covered in the present disclosure. within the scope of protection. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims (22)

1. A method for media content search, wherein the method comprises: Receive a search instruction for target media content; Based on the search instruction, determine a first candidate media content set that matches the scene intent of the target media content, and a second candidate media content set that matches the entity intent of the target media content. The scene intent is related to the feature information under various preset dimensions; Based on the first candidate media content set and the second candidate media content set, a search result corresponding to the target media content is sent to the user terminal. 2. The method according to claim 1, wherein the determining the first candidate media content set matching the scene intent of the target media content comprises: determining the target media content feature vector corresponding to the target media content based on the feature information of the target media content in multiple preset dimensions; By matching the target media content feature vector with each candidate media content feature vector in the scene index library, at least one first candidate media content matching the scene intent of the target media content is determined, and the at least one first candidate media content is determined. A candidate media content constitutes the first candidate media content set. 3. The method according to claim 2, wherein the media content is a video, and the multiple preset dimensions include multiples of the following dimensions: Visual dimension, textual information dimension, musical dimension. 4. The method according to claim 2, wherein the scene index library is generated according to the following steps: Perform scene intention identification on each primary selection media content, and determine the first candidate media content with scene intention in each primary selection media content; extracting feature information of the first candidate media content under multiple preset dimensions; generating a candidate media content feature vector of the first candidate media content based on feature information of the first candidate media content in multiple preset dimensions; The first media content identifier of the first candidate media content and the candidate media content feature vector of the first candidate media content are stored in the scene index library correspondingly. 5. The method according to claim 4, wherein, by matching the target media content feature vector with each candidate media content feature vector in the scene index library, determine the scene with the target media content At least one first candidate media content intended to match, including: Matching the target media content feature vector of the target media content with each candidate media content feature vector in the scene index library, and determining at least one candidate media content feature vector that matches the target media content feature vector; The at least one first candidate media content is determined based on the first media content identifier corresponding to the determined feature vector of the candidate media content in the scene index library. 6. The method according to claim 5, wherein the forming the at least one first candidate media content into the first candidate media content set comprises: acquiring user behavior information corresponding to the first candidate media content identified by the first media content identifier; Based on the user behavior information, first candidate media content that meets a preset requirement is selected from the first candidate media content identified by the first media content identifier to form the first candidate media content set. 7. The method according to claim 1, wherein determining the second candidate media content set matching the entity intent of the target media content comprises: Determine the target entity feature vector of the target entity corresponding to the entity intent in the target media content; Match the target entity feature vector of the target entity with each candidate entity feature vector in the entity index library, and determine a second candidate media content set that matches the target entity; the second candidate media content set includes at least A second candidate media content. 8. The method according to claim 7, wherein the determining the target entity feature vector of the target entity corresponding to the entity intent in the target media content comprises: detecting an entity in the target media content; Perform intent recognition on the detected at least one entity, and determine at least one of the target entities; For each target entity, the target entity feature vector corresponding to the target entity is generated based on the image information corresponding to the target entity in the target media content. 9. The method according to claim 7, wherein the entity index library is generated according to the following steps: determine the entities included in each primary media content; Perform intent identification on entities included in the primary selection media content, and determine a candidate entity and a second candidate media content where the candidate entity is located; generating the candidate entity feature vector corresponding to the candidate entity based on the image information of the candidate entity in the corresponding second candidate media content; The second media content identifier of the second candidate media content where the candidate entity is located and the candidate entity feature vector of the candidate entity are correspondingly stored in the entity index library. 10. The method according to claim 9, wherein the target entity feature vector of the target entity is matched with each candidate entity feature vector in the entity index library, and the first entity matching the target entity is determined. Two candidate media content sets, including: Matching the target entity feature vector of the target entity in the target media content with each candidate entity feature vector in the entity index library, and determining at least one candidate entity feature vector that matches the target media content; The second candidate media content set is determined based on the second media content identifier corresponding to the determined candidate entity feature vector in the entity index library. 11 . The method according to claim 10 , wherein the second media content candidate is determined based on a second media content identifier corresponding to the determined feature vector of the candidate entity in the entity index library. 12 . Collection, including: acquiring user behavior information corresponding to the second candidate media content identified by the second media content identifier; Based on the user behavior information, a second candidate media content that meets a preset requirement is selected from the second candidate media content identified by the second media content identifier to form the second candidate media content set. 12. The method according to claim 1, wherein, based on the first candidate media content set and the second candidate media content set, the search corresponding to the target media content is sent to the client Results, including: generating first set identification information corresponding to the first candidate media content set, and generating second set identification information respectively corresponding to at least one second candidate media content subset in the second candidate media content set; wherein each The second candidate media content subset corresponds to a candidate entity that matches the entity intent; Combining the first candidate media content set and the first set identification information corresponding to the first candidate media content set, at least one second candidate media content subset and each second candidate media content in the second candidate media content set The second set identification information respectively corresponding to the content subsets is sent to the user terminal as the search result. 13. The method according to claim 12, wherein the first set identification information comprises a first thumbnail image and/or first text description information; The second set identification information includes second thumbnail images and/or second text description information. 14. A method for media content search, wherein the method comprises: Send a search instruction for the target media content to the server; Receive a search result fed back by the server; the search result includes a first candidate media content set that matches the scene intent of the target media content, and/or a second candidate that matches the entity intent of the target media content A set of media content, where the scene intent of the target media content is related to feature information under various preset dimensions; Based on the search results, a search result presentation page is displayed. 15. The method according to claim 14, wherein the sending a search instruction for the target media content to the server comprises: In response to the trigger operation of the search button on the target media content screen, send a search instruction for the target media content to the server; or, In response to the trigger operation of the frame selection button acting on the target media content screen, a search instruction for the frame-selected media content is sent to the server. 16. The method according to claim 14, wherein the search result further comprises first set identification information corresponding to the first candidate media content set, and at least one of the second candidate media content set Second set identification information respectively corresponding to the second candidate media content subsets; wherein, each of the second candidate media content subsets corresponds to a candidate entity that matches the entity intent; The displaying a search result display page based on the search result, including: Based on the search result, displaying a search result display page containing the first set identification information and the second set identification information; After the displaying the search result display page, the method further includes: In response to a trigger operation for any set identification information, display candidate media content corresponding to the any set identification information; wherein, the any set identification information is the first set identification information or any of the second sets identification information. 17. The method according to claim 16, wherein, based on the search result, displaying a search result display page containing the first set of identification information and at least one second set of identification information, comprising: Based on the search result, displaying the search result display page including the first search result display area and the second search result display area; Wherein, the first search result display area includes the first set identification information and the second set identification information; the second search result display area includes a media content list, and the media content list includes the each media content in the first candidate media content set and the second candidate media content set. 18. The method of claim 16, wherein the method further comprises: After displaying candidate media content corresponding to any set identification information in response to a trigger operation for any set identification information, in response to a sliding trigger operation, switching to display other candidate media content corresponding to other set identification information. 19. An apparatus for searching media content, wherein the apparatus comprises: an instruction receiving module for receiving a search instruction for the target media content; a set determination module, configured to determine, based on the search instruction, a first candidate media content set that matches the scene intent of the target media content, and a second candidate media content set that matches the entity intent of the target media content, The scene intent of the target media content is related to feature information under various preset dimensions; A result search module, configured to send a search result corresponding to the target media content to the user terminal based on the first candidate media content set and the second candidate media content set. 20. An apparatus for searching media content, wherein the apparatus comprises: an instruction sending module, configured to send a search instruction for the target media content to the server; A result receiving module, configured to receive a search result fed back by the server; the search result includes a first candidate media content set that matches the scene intent of the target media content, and/or an entity that matches the target media content a set of second candidate media content for intent matching, where the scene intent of the target media content is related to feature information under various preset dimensions; A page display module, configured to display a search result display page based on the search result. 21. A computer device, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device runs, the processor and the The memories communicate with each other through a bus, and when the machine-readable instructions are executed by the processor, the steps of the method for searching media content according to any one of claims 1 to 18 are performed. 22. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program is executed by the processor when the media content search according to any one of claims 1 to 18 is executed. steps of the method.

Images