CN118093792A

CN118093792A - Method, device, computer equipment and storage medium for searching object

Info

Publication number: CN118093792A
Application number: CN202410504779.4A
Authority: CN
Inventors: 黄斌
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2024-04-25
Filing date: 2024-04-25
Publication date: 2024-05-28
Anticipated expiration: 2044-04-25
Also published as: CN118093792B

Abstract

The present application relates to a method, an apparatus, a computer device, a storage medium and a computer program product for object searching. The method comprises the following steps: acquiring an object search text aiming at a target field, and extracting search entity words in the object search text; searching candidate objects matched with the search entity words from the target field, and determining respective description text of each candidate object, wherein the description text is generated based on structural information of the candidate object in the target field; carrying out semantic matching on each description text and the object search text respectively to obtain semantic similarity of each description text and the object search text; and determining a search result of searching the text for the object based on the target object characterized by the descriptive text with the semantic similarity meeting the semantic filtering condition. By adopting the method, the accuracy of object searching can be improved.

Description

Method, device, computer equipment and storage medium for searching object

Technical Field

The present application relates to the field of internet technologies, and in particular, to a method, an apparatus, a computer device, and a storage medium for searching an object.

Background

With the rapid development of internet technology and the continuous improvement of the living standard of people, people can search objects directly through the internet, for example: searching for a desired video, searching for a desired song, searching for a desired article, etc., provides great convenience and enjoyment to the life of people. It is important how to accurately search for an object that the user really wants to find according to the search contents input by the user. At present, when searching an object, the search content input by the user is often matched with the keyword of the object, and the keyword cannot completely and clearly describe the accurate meaning of the object, but the user search content and the keyword can only describe the similarity of the text keyword, so that the searched object is possibly inconsistent with the object which the user really wants to search, and the accuracy of the object search is reduced.

Disclosure of Invention

In view of the foregoing, it is desirable to provide an object searching method, apparatus, computer device, and storage medium that can improve accuracy of object searching.

In a first aspect, the present application provides a method of object searching. The method comprises the following steps:

Acquiring an object search text aiming at a target field, and extracting search entity words in the object search text;

Searching candidate objects matched with the search entity words from the target field, and determining respective description text of each candidate object, wherein the description text is generated based on structural information of the candidate object in the target field;

carrying out semantic matching on each description text and the object search text respectively to obtain semantic similarity of each description text and the object search text;

and determining a search result of searching the text for the object based on the target object characterized by the descriptive text with the semantic similarity meeting the semantic filtering condition.

In a second aspect, the application further provides an object searching device. The device comprises:

the object search text acquisition module is used for acquiring an object search text aiming at the target field and extracting search entity words in the object search text;

The descriptive text determining module is used for searching candidate objects matched with the search entity words from the target field, determining respective descriptive text of each candidate object, and generating the descriptive text based on the structural information of the candidate object in the target field;

The semantic similarity matching module is used for carrying out semantic matching on each description text and the object search text respectively to obtain the semantic similarity of each description text and the object search text;

and the search result determining module is used for determining a search result of the text for searching the object based on the target object characterized by the description text with the semantic similarity meeting the semantic screening condition.

In a third aspect, the present application also provides a computer device. The computer device comprises a memory storing a computer program and a processor which when executing the computer program performs the steps of:

In a fourth aspect, the present application also provides a computer-readable storage medium. The computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps of:

In a fifth aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements the steps of:

According to the method, the device, the computer equipment, the storage medium and the computer program product for searching the objects, the search entity words in the object search text are extracted, candidate objects matched with the search entity words are searched in the target field, namely, candidate objects matched with the search entity words in the object search text in text are firstly searched in the target field, then natural language description processing is carried out on the candidate objects according to structural information of the target field, so that each description text of each candidate object is determined, natural language description can be carried out on the candidate object in the target field, therefore, semantic matching can be carried out on each description text and the object search text, the target object characterized by the description text with the semantic similarity meeting the semantic screening condition is selected, the selected target object meets the semantic requirement of the object search text, and therefore, the search results are determined from the two dimensions of text matching and semantic matching, and the accuracy of the object search can be improved.

Drawings

FIG. 1 is an application environment diagram of a method of object searching in one embodiment;

FIG. 2 is a system framework diagram of an object search system in one embodiment;

FIG. 3 is a flow diagram of a method of object searching in one embodiment;

FIG. 4 is a schematic diagram of a tag list in one embodiment;

FIG. 5 is a flow diagram of semantic similarity determination in one embodiment;

FIG. 6 is a schematic diagram of an ordered set of text in one embodiment;

FIG. 7 is a flow diagram that illustrates determining search results for searching text for objects, in one embodiment;

FIG. 8 is a flow diagram of determining search results for searching text for objects in another embodiment;

FIG. 9 is a flow chart of determining search results for searching text for objects in yet another embodiment;

FIG. 10 is a complete flow diagram of a method of object searching in one embodiment;

FIG. 11 is a block diagram of an object search apparatus in one embodiment;

fig. 12 is an internal structural diagram of a computer device in one embodiment.

Detailed Description

The present application will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present application more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the application.

With the rapid development of internet technology and the continuous improvement of the living standard of people, people can search objects directly through the internet, for example: searching for a desired video, searching for a desired song, searching for a desired article, etc., provides great convenience and enjoyment to the life of people. It is important how to accurately search for an object that the user really wants to find according to the search contents input by the user.

At present, when searching an object, the search content input by the user is often matched with the keyword of the object, because the keyword cannot completely and clearly describe the exact meaning of the object, and the user search content and the keyword can only describe the similarity of text keywords, for example, in the field of music search, whether the search box on various music Application devices (APP) or the dialogue of intelligent devices, the sound box product requests a song, the text is searched for the search word or the object input by the user, and the associated playlist is searched by a certain technology. The target searches the text based on the object, finds a song list similar to the text from the database, and finds a great number of results at this time, and sorts the results for the user to display by adopting a certain means, such as sorting based on song popularity values, sorting based on song release time, and the like. Therefore, the searched object may not coincide with the object that the user really wants to search, thereby reducing the accuracy of the object search.

The embodiment of the application provides an object searching method capable of realizing object searching efficiency. The object searching method provided by the embodiment of the application can be applied to an application environment shown in figure 1. Wherein the terminal 102 communicates with the server 104 via a network. The data storage system may store data that the server 104 needs to process. The data storage system may be integrated on the server 104 or may be located on the cloud or other servers.

Specifically, taking the terminal 102 in fig. 1 as an example, the terminal 102 first obtains the object search text for the target field, where the object search text for the target field may be obtained by text input by a user through a text input device such as a numeric keypad or an external keyboard, or may be obtained by performing voice text conversion processing on voice search data after voice input by the user through a voice input device, which is not particularly limited herein. Based on this, the terminal 102 may extract a search entity word in the object search text, find candidate objects matching the search entity word from target fields, which may be an audio-video field, a text field, and other search fields, and determine a respective description text for each candidate object. Then, the terminal 102 performs semantic matching on each description text and the object search text respectively to obtain semantic similarity between each description text and the object search text, and determines a search result for the object search text based on the target object represented by the description text whose semantic similarity satisfies the semantic screening condition. The method and the device can determine the search results from two dimensions of text matching and semantic matching, so that the accuracy of object searching can be improved.

The terminal 102 may be, but not limited to, various desktop computers, notebook computers, smart phones, tablet computers, internet of things devices, and portable wearable devices, where the internet of things devices may be smart speakers, smart televisions, smart air conditioners, smart vehicle devices, and the like. The portable wearable device may be a smart watch, smart bracelet, headset, or the like. The server 104 may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDNs, basic cloud computing services such as big data and artificial intelligence platforms, and the like. The terminal and the server may be directly or indirectly connected through wired or wireless communication, and the present application is not limited herein. And the object searching method provided by the application embodiment can be applied to various scenes, including but not limited to cloud technology, artificial intelligence and the like.

For easy understanding, the system framework provided by the embodiment of the present application is briefly described below, taking application to the field of music search as an example, a system framework diagram as shown in fig. 2 is used, first, an object search text needs to be acquired, then entity recognition is performed on the object search text to extract a search entity word, a music structured information search result matched with the search entity word is obtained by searching the search entity word from a music database, then natural language expression processing is performed on the music structured information search result, that is, the music structured information is changed into a text document described in natural language, and specifically, all the retrieved music can be documented and described by using natural language alignment. Based on the method, the object search text and the text document are input into the double-tower model, the text document which is most relevant is determined from the text documents by adopting the double-tower model aiming at the object search text and the text document, the text documents are ordered according to the relevance degree of the object search text, the ordered text documents are converted, namely the ordered text documents are converted into music structural information, and therefore the ordered music structural information meeting the conditions is input. The music structured information can be directly music, and the music structured information can also be music labels related to music, such as: the music name, music style, music singer (or music player), music distribution time, music producer, etc., are not limited herein.

Further, since the object searching method provided by the embodiment of the present application further relates to artificial intelligence (ARTIFICIAL INTELLIGENCE, AI), the following description will be made of AI. AI is a theory, method, technique, and application system that utilizes a digital computer or a digital computer-controlled machine to simulate, extend, and extend human intelligence, sense the environment, acquire knowledge, and use knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include, for example, sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, pre-training model technologies, operation/interaction systems, mechatronics, and the like. The pre-training model is also called a large model and a basic model, and can be widely applied to all large-direction downstream tasks of artificial intelligence after fine adjustment. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

The scheme provided by the embodiment of the application relates to natural language processing (Nature Language processing, NLP) technology and machine learning (MACHINE LEARNING, ML) technology under artificial intelligence technology. Among them, natural language processing is an important direction in the fields of computer science and artificial intelligence. It is studying various theories and methods that enable effective communication between a person and a computer in natural language. The natural language processing relates to natural language, namely the language used by people in daily life, and is closely researched with linguistics; and also to computer science and mathematics. An important technique for model training in the artificial intelligence domain, a pre-training model, is developed from a large language model (Large Language Model) in the NLP domain. Through fine tuning, the large language model can be widely applied to downstream tasks. Natural language processing techniques typically include text processing, semantic understanding, machine translation, robotic questions and answers, knowledge graph techniques, and the like.

Machine learning is a multi-domain interdisciplinary, and relates to multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. It is specially studied how a computer simulates or implements learning behavior of a human to acquire new knowledge or skills, and reorganizes existing knowledge structures to continuously improve own performance. Machine learning is the core of artificial intelligence, a fundamental approach to letting computers have intelligence, which is applied throughout various areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, confidence networks, reinforcement learning, transfer learning, induction learning, teaching learning, and the like. The pre-training model is the latest development result of deep learning, and integrates the technology.

The object search method is specifically described by the following examples: in one embodiment, as shown in fig. 3, an object searching method is provided, where the method is applied to the terminal 102 in fig. 1, and is illustrated by way of example, it is understood that the method may also be applied to the server 104, and may also be applied to a system including the terminal 102 and the server 104, and implemented through interaction between the terminal 102 and the server 104. In this embodiment, the method includes the steps of:

Step 302, obtaining an object search text aiming at a target field, and extracting search entity words in the object search text.

The target domain is a domain to which the searched object belongs, and the searched object may be a music object, a video object, an image object, an article object, or the like, and thus the target domain may be a music search domain, a video search domain, a map search image domain, an article search domain, or the like. Based on this, the object search text is used to describe object information that the user wants to search for an object, and the object search text may be text input by the user when performing the object search, or text obtained by performing text conversion on speech input by the user when performing the object search, which is not limited herein. Secondly, the search entity word is an entity descriptor for describing the searched object, and the entity in this embodiment may be: name of person, place name, organization name, trade name, clause name, time, etc.

Specifically, when the user needs to perform object searching, the user can perform text input on the terminal through a text input device such as a numeric keyboard, an external keyboard and the like, so that the terminal obtains an object searching text aiming at the target field. It can be understood that, after the user performs voice input of the voice search data through the voice input device, voice text conversion processing is performed on the voice search data, so that the terminal obtains the object search text for the target field. And under other application scenarios, the terminal may also acquire the object search text for the target domain from the server side, and no specific limitation is made here on how to acquire the object search text for the target domain.

Based on this, the introduction is made with a phonetic text conversion process with phonetic input, and in an alternative embodiment, obtaining object search text for a target area includes: and acquiring voice search data aiming at the target field, and performing voice text conversion processing on the voice search data to obtain an object search text.

That is, the user needs to perform voice description on the object in the target field through the voice input device, so that the terminal obtains voice search data for the target field through the voice input device, and the voice input device refers to a man-machine interface device for directly inputting voice information of a person into the terminal, so that the voice input device can be a microphone built in the terminal, or a microphone externally connected with the terminal, and the like, and the voice input device is not limited herein. Based on this, after the terminal obtains the voice search data, it needs to perform voice text conversion processing on the voice search data to obtain an object search text corresponding to the voice search data, for example, the voice search data is "AAA of a version where me does not want to hear somehow", and then the object search text corresponding to the voice search data that can be obtained after performing the voice text conversion processing: "I do not want to hear some version of AAA".

Specifically, a voice-text conversion process is used to translate voice data into text year data, and the voice-text conversion process generally employs: conversion strategies for speech recognition, sentence spelling, and text translation, therefore, the speech-to-text conversion process may be: determining a phoneme sequence and a voice recognition result corresponding to the voice search data, then performing translation processing on the voice recognition result to obtain a voice translation result, performing text conversion processing on the phoneme sequence to obtain a text conversion result, and finally determining an object search text corresponding to the voice search data through the voice translation result and the text conversion result. It will be appreciated that there is no timing limitation between the foregoing steps. And other manners may be adopted in practical applications, such as: speech translation based on translation software, statistical-based speech translation, deep learning-based speech translation, and frame-based speech translation. Wherein, the statistical-based speech translation is trained by using a large amount of speech data, thereby establishing a statistical model for translation. Whereas deep learning based speech translation refers to translation using deep neural networks. And frame-based speech translation refers to decomposing the translation problem into a plurality of subtasks and organizing them into a frame for translation.

Further, after the object search text for the target field is obtained, the terminal needs to perform entity recognition on the object search text to extract the search entity words in the object search text, and since the entity recognition is specifically used for extracting the entities in a sentence, the search entity words are the entities in the object search text, that is, the search entity words specifically describe an entity, then the words used for expressing the relationship between emotion or connection words in the object search text are not recognized as object tags. For example, taking the target domain as the music search domain as an example, the object search text is "play a song of a shellfish", then the search entity word may be "shellfish". The object search text is "listen to classical music of some kind", and then the search entity words may be "peri-some kind" and "classical music". And the object search text is "i like music is classical music, but i now want to listen to popular music", then the search entity word may be "classical music" as well as "popular music".

Step 304, searching candidate objects matched with the search entity words from the target field, and determining respective description text of each candidate object, wherein the description text is generated based on the structural information of the candidate object in the target field.

The candidate objects are specifically: the text similarity between the object tag corresponding to the candidate object and the search entity word satisfies the text filtering condition, and the candidate objects are usually a plurality of. And the objects in different fields all have corresponding structured information, the structured information is specifically object labels of the objects, the object labels can represent description information of the objects in the target field, and the object labels can be single or multiple.

Secondly, the description text is generated based on the structural information of the candidate object in the target field, that is, different structural information exists in different fields, the structural information is used for describing key information of each object in the target field, for example, taking the target field as a music search field, the structural information can be: music singer a, music name B, music album name C, music distribution time D, and music heat value E. Taking the object domain as an article search domain as an example, the structured information may be: an article publisher a, an article title B, an article abstract C, an article publication time D, and an article download rate E. It can be seen that the key information of each description dimension of the object in the target field can be described through the structural information.

Specifically, the terminal searches candidate objects matched with the search entity words from the target field. That is, the terminal can search the object with higher association degree in the database based on the search entity word in the database searching mode to serve as the candidate object. For example, taking the target domain as the music search domain again as an example, if the object search text is "play AAA", the search entity word "AAA" in the music search domain may be extracted from the "play AAA", and then the candidate object matched with the search entity word may be "AAA" sung by a plurality of singers, for example: the candidate objects include: AAA in edition of a recording studio, AAA in singing on site, AAA in singing by singer A1, AAA in singing by singer A2, and AAA in singing by singer A3.

The manner in which the candidate object is found will be described in detail below: in one particular embodiment, finding candidate objects that match the search entity word from the target domain includes: acquiring a plurality of selectable objects in the target field and respective object labels of each selectable object; respectively carrying out text similarity matching on each search entity word and each object label of each selectable object to obtain text similarity of each search entity word and each object label; and determining the selectable object corresponding to the object label with the text similarity meeting the text screening condition as a candidate object matched with the search entity word.

Wherein, the selectable objects are all objects which belong to the target domain and can be searched, and taking the target domain as the music searching domain as an example, the selectable objects are all music which can be searched in the music searching domain. Taking the target domain as an article searching domain as an example, the selectable objects are all articles which can be searched in the article searching domain. Secondly, as can be seen from the foregoing description, the object tag may represent the description information of the object in the target domain, for convenience of understanding, taking the target domain as a music search domain as an example, the object is music, and then the music tag of the music may represent at least any one of the following: music singer, music name, music album name, music distribution time, music popularity value, music style, music lyrics, music producer, etc. Taking the target domain as an article search domain as an example, the object is an article, and then the article tag of the article may characterize at least any one of the following: an article publisher, an article title, an article abstract, an article publication time, an article download rate/article click rate, an article domain, and the like.

Based on this, text similarity is used to describe: searching description information of the object represented by the entity word and the object label in the target field, and similarity in text dimension. The text filtering condition may be that the text similarity reaches a text similarity threshold, or may also be a plurality of preset number of text similarities with higher values, which is not limited herein specifically.

Specifically, the terminal firstly acquires a plurality of selectable objects in the target field and respective object labels of each selectable object. The terminal may acquire a plurality of selectable objects in the target field through communication interaction with the server, and respective object tags of each selectable object, and the terminal may also acquire the selectable object and respective object tags of each selectable object through external data input, which is not limited herein to a manner of acquiring the foregoing information. Based on the text similarity matching is carried out on the entity word characteristics and each label characteristic, so that the text similarity of each search entity word and the object label is obtained. Text similarity matching may employ similarity algorithms available for Cosine distance (Cosine) computation, euclidean distance (Euclidean Distance) computation, and the like, and is not limited herein.

Therefore, the terminal screens the selectable objects corresponding to the object tags through the text similarity between each search entity word and the object tag so as to select the selectable objects of which the text similarity between the corresponding object tag and each search entity word meets the text screening condition, and determines the selectable objects as candidate objects matched with the search entity word.

In practical application, the structural information of each candidate object can be determined, that is, the description information represented by the object label of each candidate object is extracted, an information basis is provided for the subsequent determination of the description text, and the description information represented by the object label of each candidate object can be stored in a label list mode. For ease of understanding, taking the object search text as "play AAA", and the extracted search entity word "AAA" as an example, as shown in fig. 4, the search entity word "AAA" is extracted from the object search text "play AAA", and thus a plurality of candidate objects, and description information represented by object tags of each candidate object, such as description information 401 represented by object tags of candidate object 1, and description information 402 represented by object tags of candidate object 2, are obtained, and at this time, description information represented by object tags of candidate object 1 and candidate object 2 is stored in a tag list 403, that is, the tag list 403 includes:

Description information 401 characterized by a plurality of object tags of candidate object 1: song name "AAA", singer name "Wednesday", album name "BBB", heat value 10000.

Description information 402 characterized by a plurality of object tags of candidate object 2: song name "AAA", artist name "artist 111", album name "AAA", hotness value 100.

Further, after searching candidate objects matched with the search entity word, respective description text of each candidate object needs to be determined, and through the description, the structural information is specifically an object tag of the object, and the object tag can represent the description information of the object in the target field, that is, the description text of each candidate object can be obtained by filling the object tag of each candidate object in the target field into the description text template in the target field.

In a specific embodiment, determining the respective descriptive text for each candidate object includes: determining a description text template in the target field, wherein the description text template comprises blank fields matched with a plurality of object labels respectively; and filling description information represented by the object labels of each candidate object in the target field into blank fields matched with the object labels in the description text template to obtain respective description text of each candidate object.

The description text template comprises blank fields which are respectively matched with a plurality of object labels, and natural language description text connected with each blank field, wherein the blank fields are used for filling description information represented by the object labels. For ease of understanding, taking the target domain as an example of the music search domain, the description text templates under the music search domain may be: ____ songs ____ belong to music album ____, and the music style of the songs belongs to ____, the release time of the songs is ____, and the heat value of the songs in one month is ____. Thus, the blank field "____" in "____ singing" in "____ singing song" ____ "is used to fill in a music singer, and the blank field" ____ "in" ____ "is used to fill in a music name.

Similarly, taking the target domain as an article search domain as an example, the description text template under the article search domain may be: ____ article ____, article abstract information ____, article release time ____, and article click rate ____ in one month. Thus, the blank field "____" in the article "____" published by "_____" in "____" is used to fill in the article publisher, and the blank field "____" in the article "____" is used to fill in the article title. It can be seen that the blank area "____" in the above description text template is a blank field for filling the description information represented by the object tag.

It will be appreciated that the description text templates in different fields are different, and the specific description text templates need to be flexibly determined based on actual requirements, so that the positions of blank fields and specific connection information matched by the object labels respectively need to be flexibly configured, and therefore the foregoing examples should not be construed as specific limitations of the present application. And because the description information characterized by the object label of each candidate object can be stored in a label list mode, the obtained description text of each candidate object can also be stored in a text list mode.

Specifically, the terminal determines a description text template in the target field aiming at the target field, and then matches and fills in the description text template with blank fields matched with each object tag in the description text template through description information represented by the object tag of each candidate object in the target field, so that respective description text of each candidate object is obtained. For example, taking the target domain as the music search domain as an example, the description text templates under the music search domain are: ____ songs ____ belong to music album ____, and the music style of the songs belongs to ____, the release time of the songs is ____, and the heat value of the songs in one month is ____. The description information represented by the object tags of the candidate object 1 includes: song name "AAA", singer name "week someplace", album name "BBB", release time "10 months 15 days in 2008", music style "hip-hop and ballad", heat value 10000. From the foregoing analysis, blank field "____" in "____ sings" matches the music singer, whereas the descriptive information characterized by the object tags of the candidate object 1 is that the specific information of the singer name is "periclase", the blank field "____" in "____ singing" is filled with "periclase". The rest object labels are similarly filled in, so that the description text of the candidate object 1 can be obtained: song "AAA" of a singing in week belongs to music album "BBB", and music style of song belongs to hip hop and ballad, release time of song is 10 months 15 days in 2008, and heat value of song in one month is 10000.

Similarly, if the description information characterized by the object tags of the candidate object 2 includes: song name "AAA", singer name "singer 111", album name "AAA", heat value 100, release time "2021, 11 months, 11 days", music style "ballad", heat value 100, the description text of candidate 2 can be obtained in the same manner as described above: the song "AAA" by singer 111 belongs to music album "AAA", and the music style of the song belongs to ballad, the release time of the song is 2021, 11, and the popularity value of the song in one month is 100.

And 306, carrying out semantic matching on each description text and the object search text respectively to obtain the semantic similarity of each description text and the object search text.

The semantic similarity is used for representing similarity of the object search text and the object search text in semantic dimension. The method for performing semantic matching may also use available similarity algorithms such as cosine distance calculation and euclidean distance calculation, which are not limited herein. Specifically, the terminal extracts object search text semantic features of the object search text, and extracts description text semantic features corresponding to each of the description texts of each candidate object, so that similarity between the object search text semantic features and each of the description text semantic features is calculated through a similarity algorithm, and the similarity between the semantic features is determined to be the semantic similarity between each of the description texts and the object search text.

Considering that under the object retrieval scene, feature codes of different dimensions in different fields can be completed through a double-tower model, the features of two different dimensions are respectively encoded into a vector, and recall is carried out through vector similarity. Therefore, the manner of performing semantic similarity calculation through the dual-tower model will be described below, in a specific embodiment, performing semantic matching on each description text and the object search text respectively to obtain the semantic similarity between each description text and the object search text, where the method includes: extracting object searching text semantic features of an object searching text through a question vector tower in the double-tower model, and extracting description text semantic features corresponding to the description text of each candidate object through an answer vector tower in the double-tower model; and calculating cosine distances between semantic features of each description text and semantic features of the object search text through the semantic features of each description text in the double-tower model, and determining the obtained cosine distance result as the semantic similarity of each description text and the object search text.

The double-tower model comprises a question vector tower and an answer vector tower, the semantic features of the object search text for extracting the object search text can pass through the question vector tower, and the semantic features of the description text corresponding to the extracted description text can pass through the answer vector tower. Secondly, the double-tower model can be optimized based on a transformation architecture, and the text encoder is used for encoding the object search text, namely, the text encoder can be used for outputting the semantic features of the object search text. Similarly, the text encoder is adopted to encode the descriptive text, namely descriptive text semantic features corresponding to the descriptive text can be output through the text encoder, after the object search text semantic features and the descriptive text semantic features are obtained, the object search text semantic features and the descriptive text semantic features can be combined to combine the semantic features for output, or the object search text semantic features and the descriptive text semantic features can be independently output respectively.

Specifically, the terminal extracts object search text semantic features of the object search text through a question vector tower in the double-tower model, extracts description text semantic features corresponding to the description text of each candidate object through an answer vector tower in the double-tower model, calculates similarity with the object search text semantic features through a preset similarity algorithm in the double-tower model, and determines the obtained similarity calculation result as the semantic similarity of each description text and the object search text. If the preset similarity algorithm is cosine distance calculation, the cosine distance can be calculated between semantic features of each description text and semantic features of the object search text in the double-tower model, and then the obtained cosine distance result is determined to be the semantic similarity between each description text and the object search text.

In order to facilitate understanding, in the stage of constructing the dual-tower model, the features corresponding to the questions (i.e., the object search text) and the answers (i.e., the description text) are required to be embedded (Embedding), and then a neural network is used for training, so as to obtain the dual-tower model, where the neural network may be a complex deep learning network such as a convolutional neural network (Convolutional Neural Network, CNN) or a cyclic neural network (Recurrent Neural Networks, CNN), and the method is not limited herein. As shown in the semantic similarity determination flowchart of fig. 5, feature embedding is performed on an object search text (Feature Embedding), object search text semantic features of the object search text are extracted through a neural network, feature embedding is performed on description text, description text semantic features of the description text are extracted through the neural network, and cosine distance calculation is performed on the object search text semantic features and the description text semantic features, so that the obtained cosine distance result is determined to be the semantic similarity of each description text and the object search text. As can be seen from fig. 5, the dual-tower model is composed of two tower structures, one is a question vector tower 502 and the other is an answer vector tower 504, and the relevance of the obtained corresponding features can be calculated by the question vector tower 502 and the answer vector tower 504, so that the obtained dual-tower model can obtain semantic feature vectors of texts (such as description text and object search text) and can predict semantic similarity between the two texts (such as description text and object search text).

Step 308, determining a search result of the text for the object search based on the target object characterized by the descriptive text with semantic similarity satisfying the semantic filtering condition.

The semantic filtering condition may be that the semantic similarity reaches a semantic similarity threshold, or may also be a plurality of preset number of semantic similarities with higher values, and the like, which is not limited herein. Secondly, the search result can be an object in the target field or description information of the object in the target field, and the description information is structural information of the object in the target field. For example, in the field of music search, the search result may be a song, and may also be descriptive information including a song title, a song artist, and a song release time.

Specifically, considering semantic screening conditions, the terminal selects semantic similarity meeting the semantic screening conditions from the description texts through the semantic similarity between each description text and the object search text, determines the description text corresponding to the semantic similarity meeting the semantic screening conditions, and determines a search result of the object search text through the target object represented by the description text. As can be seen from the foregoing description, the target object may be an object in the target domain, and the target object may also be description information of the object in the target domain.

In an alternative embodiment, after determining the search result for searching text for the object, the method of object searching may further include: and outputting the search result of the text for searching the object in a voice broadcasting mode. Or may also display search results for the object search text in a highlighted manner at the display interface of the terminal. The manner in which the search results are output to the user is not specifically limited herein.

It can be seen from the foregoing embodiments that the extraction of the text semantic features and the description text semantic features of the object search can be performed by the double-tower model, and then the process of determining the search result can also be performed by the double-tower model. In a specific embodiment, determining a search result for searching text for an object based on a target object characterized by descriptive text whose semantic similarity satisfies a semantic filtering condition includes: according to the semantic similarity of each description text and the object search text, ordering each description text from large to small through a double-tower model to obtain an ordered text set; selecting a target object characterized by descriptive text with semantic similarity meeting semantic screening conditions from the ordered text set, and determining a search result for searching the text for the object based on the target object.

The ranking of each description text may be that the numerical values based on the semantic similarity are arranged in sequence from large to small, or that the numerical values based on the semantic similarity are arranged in sequence from small to large. Based on this, the ordered text set is used to describe the magnitude relation of the semantic similarity between each description text and the object search text, and since the manner of sorting can be as described in the previous example, the obtained ordered text set can be the description text with the corresponding numerical value of the semantic similarity arranged sequentially from large to small, or can also be the description text with the corresponding numerical value of the semantic similarity arranged sequentially from small to large.

Specifically, the terminal sorts each description text from large to small through a double-tower model according to the semantic similarity of each description text and the object search text, and a sequential text set is obtained. Since the ordered text set is used to describe the magnitude relation of semantic similarity between each description text and the object search text, for convenience of understanding, the ordered text set is arranged from large to small in order, for example, as shown in fig. 6, where the ordered text set 602 includes: descriptive text with the greatest numerical value of semantic similarity to the object search text: "Song" AAA "of a singing in week belongs to music album" BBB ", and music style of Song belongs to hip-hop and ballad, release time of Song is 10 months 15 days in 2008, and heat value of Song in one month is 10000.", and the latter descriptive text arranged in order from large to small: "Song" AAA "by singer 111 belongs to music album" AAA ", and the music style of song belongs to ballad, the release time of song is 2021, 11, and the popularity value of song in one month is 100.", and so on.

Based on the above, the terminal considers different semantic filtering conditions, selects a target object characterized by the descriptive text with semantic similarity meeting the semantic filtering conditions from the ordered text set, and determines a search result for the object search text based on the target object.

It should be understood that the illustrations in the embodiments of the application and the respective examples are for understanding the present application, but are not to be construed as specific limitations of the present application.

In the method for searching the object, the search entity words in the object search text are extracted, candidate objects matched with the search entity words are searched in the target field, namely, candidate objects which can be matched with the search entity words in the object search text in the text are firstly searched in the target field, then natural language description processing is carried out on the candidate objects according to structural information of the target field, so that the description text of each candidate object is determined, and the natural language description can be carried out on the candidate objects in the target field, so that semantic matching can be carried out on each description text and the object search text respectively, the target objects characterized by the description texts with semantic similarity meeting semantic screening conditions are selected, the semantic requirements of the object search texts are guaranteed to be met, and therefore, the search result determination can be carried out from the two dimensions of text matching and semantic matching, and the accuracy of the object search can be improved.

Since the semantic filtering condition may be that the semantic similarity reaches the semantic similarity threshold, or may also be a plurality of preset number of semantic similarities with higher values, a manner how to determine the search result for the search semantic under different semantic filtering conditions is described below, first, a case that the semantic filtering condition is that the semantic similarity reaches the semantic similarity threshold is described, and in one embodiment, determining the search result for the object search text based on the target object represented by the description text whose semantic similarity satisfies the semantic filtering condition, as shown in fig. 7, includes:

And step 702, determining the description text with the semantic similarity reaching the semantic similarity threshold as a target description text.

The semantic similarity threshold is a preset minimum value for describing the semantic similarity between the text and the object search text, the semantic similarity threshold can be 60%,65%,55% and the like, and specific semantic similarity thresholds need to be flexibly determined based on actual conditions. The semantic similarity between the object description text and the object search text reaches a semantic similarity threshold, and the object description text can be the description text or comprise a plurality of description texts. Specifically, the terminal compares the semantic similarity between each description text and the object search text with a semantic similarity threshold value respectively, and determines the description text corresponding to the semantic similarity as the target description text when the semantic similarity is greater than or equal to (i.e. reaches) the semantic similarity threshold value.

For ease of understanding, the object search text B1, the description text C2, the description text C3 to the description text B10 exist, and the semantic similarity between the object search text B1 and the description text C1 is 30%, the semantic similarity between the object search text B1 and the description text C2 is 65%, the semantic similarity between the object search text B1 and the description text C3 is 80%, the semantic similarity between the object search text B1 and the description text C4 is 35%, the semantic similarity between the object search text B1 and the description text C5 is 25%, the semantic similarity between the object search text B1 and the description text C6 is 60%, the semantic similarity between the object search text B1 and the description text C7 is 70%, the semantic similarity between the object search text B1 and the description text C8 is 90%, the semantic similarity between the object search text B1 and the description text C9 is 10%, and the semantic similarity between the object search text B1 and the description text C10 is 50%.

Based on this, taking the example that the semantic similarity threshold is 60%, the semantic similarity reaches 60% as follows: semantic similarity between the object search text B1 and the description text C2, semantic similarity between the object search text B1 and the description text C3, semantic similarity between the object search text B1 and the description text C6, semantic similarity between the object search text B1 and the description text C7, and semantic similarity between the object search text B1 and the description text C8. Thus, the resulting target description text includes: description text C2, description text C3, description text C6, description text C7, and description text C8.

In step 704, search results for searching text for objects are determined based on the target objects characterized by the target descriptive text in the target domain.

Specifically, the terminal determines a search result for searching text for an object based on a target object characterized by the target descriptive text in the target field. I.e. the terminal may determine the object corresponding to the target description text in the target field as the target object. Or the terminal can also determine the description information of the object corresponding to the target description text in the target field as the target object. It should be understood that the corresponding examples in the embodiments of the present application are all for understanding the present solution, but should not be construed as specific limitations of the present solution.

In this embodiment, the description text is screened through the semantic similarity threshold, so that the semantics of the selected target description text and the object search text are ensured to be closer, and therefore, the search result is determined through the target object represented by the target description text, so that the semantic matching degree of the search result and the object search text is ensured, the problem of result deviation caused by semantic deviation is avoided, and the object search accuracy is further improved.

In the following, a description will be given of a case where the semantic filtering condition is a plurality of preset number of semantic similarities with higher values, and in one embodiment, as shown in fig. 8, determining a search result for an object search text based on a target object characterized by a description text whose semantic similarity satisfies the semantic filtering condition, includes:

Step 802, sorting is performed based on each semantic similarity, and a semantic similarity sorting result is obtained.

Similar to the foregoing sorting manner, the sorting of each semantic similarity may be performed sequentially from large to small according to the numerical value of the semantic similarity, or may be performed sequentially from small to large according to the numerical value of the semantic similarity. Based on this, the semantic similarity sorting result is used to describe the magnitude relation of the semantic similarity between each description text and the object search text, and since the sorting manner can be as in the previous example, the obtained semantic similarity sorting result can be the semantic similarity arranged sequentially from large to small, or can also be the semantic similarity arranged sequentially from small to large.

Specifically, the terminal ranks according to the semantic similarity between each description text and the object search text according to a preset ranking manner to obtain a semantic similarity ranking result, for easy understanding, the ranking manner is that the semantic similarity between each description text and the object search text is sequentially ranked from big to small, and the semantic similarity between each description text and the object search text is as in the previous example, that is, the semantic similarity between the object search text B1 and the description text C1 to the description text B10 respectively, then the obtained semantic similarity ranking result is: 90% (semantic similarity between object search text B1 and description text C8), 80% (semantic similarity between object search text B1 and description text C3), 70% (semantic similarity between object search text B1 and description text C7), 65% (semantic similarity between object search text B1 and description text C2), 60% (semantic similarity between object search text B1 and description text C6), 50% (semantic similarity between object search text B1 and description text C10), 35% (semantic similarity between object search text B1 and description text C4), 30% (semantic similarity between object search text B1 and description text C1), 25% (semantic similarity between object search text B1 and description text C5), 10% (semantic similarity between object search text B1 and description text C9).

Step 804, selecting a preset number of description texts as target description texts based on the semantic similarity sorting result.

The preset number is the maximum number of texts in the target description texts, for example, the preset number is 2, and the target description texts comprise at most 2 description texts. If the preset number is 6, the target description text includes at most 6 description texts. It will be appreciated that if the preset number is greater than the number of description texts obtained, then all description texts are directly determined as target description texts. Specifically, the terminal selects a preset number of semantic similarities from the large value to the small value of the semantic similarities based on the size relation of the semantic similarities described by the semantic similarity sequencing result, and takes a description text corresponding to the semantic similarities as a target description text.

For ease of understanding, taking a preset number of 6 as an example, the semantic similarity ranking results "90%, 80%, 70%, 65%, 60%, 50%, 35%, 30%, 25%, 10%" in the foregoing examples indicate that the semantic similarity "90%, 80%, 70%, 65%, 60%, 50%" of 6 before numerical ranking of the semantic similarity is selected, and the semantic similarity "90%, 80%, 70%, 65%, 60%, 50%" is specifically: semantic similarity between the object search text B1 and the description text C8, semantic similarity between the object search text B1 and the description text C3, semantic similarity between the object search text B1 and the description text C7, semantic similarity between the object search text B1 and the description text C2, semantic similarity between the object search text B1 and the description text C6, and semantic similarity between the object search text B1 and the description text C10. Thus, the target description text may be determined to include: description text C8, description text C3, description text C7, description text C2, description text C6, and description text C10.

At step 806, search results for searching text for objects are determined based on the target objects characterized by the target descriptive text in the target domain.

Specifically, similar to the foregoing embodiments, the terminal determines a search result for searching text for an object based on a target object characterized by the target descriptive text in the target field. I.e. the terminal may determine the object corresponding to the target description text in the target field as the target object. Or the terminal can also determine the description information of the object corresponding to the target description text in the target field as the target object. It should be understood that the corresponding examples in the embodiments of the present application are all for understanding the present solution, but should not be construed as specific limitations of the present solution.

In this embodiment, sorting is performed through semantic similarity, filtering is performed on description texts through a threshold value, so that semantics between a selected target description text and an object search text are guaranteed to be closer, and therefore, determination of search results is performed through a target object represented by the target description text, so that the matching degree of the search results and the object search text in terms of semantics is guaranteed, the problem of result deviation caused by semantic deviation is avoided, and the accuracy of object search is further improved.

Based on the foregoing example, since the target object in the present solution may be a corresponding object in the target domain, and may also be description information of the corresponding object in the target domain, how to determine the target object will be described in detail below, in one embodiment, the method includes first introducing that the target object may be a corresponding object in the target domain, as shown in fig. 9, determining, based on the target object represented by the target description text in the target domain, a search result for the target search text, including:

And step 902, determining a candidate object corresponding to the target description text in the target field as a search result of searching the text for the object.

The candidate objects corresponding to the target description text in the target field are target objects characterized by the target description text in the target field. Specifically, the terminal first determines candidate objects, which are included in the target description text and correspond to the description text in the target field, and determines the candidate objects, which correspond to the description text in the target field, as target objects, thereby determining the target objects as search results for searching the text for the objects.

To facilitate understanding, further to the foregoing example, the description text of candidate 1 "song" AAA "singed somewhere around, belongs to music album" BBB ", and the music style of the song belongs to hip-hop and ballad, the release time of the song is 10 months 15 days in 2008, and the heat value of the song in one month is 10000. "and the descriptive text of the candidate object 2" song "AAA" by singer 111, belongs to music album "AAA", and the music style of the song belongs to ballad, the release time of the song is 2021, 11, and the popularity value of the song in one month is 100.". If the target description text is the description text "song with a certain performance" AAA "belongs to the music album" BBB ", the music style of the song belongs to hip-hop and ballad, the release time of the song is 10 months and 15 days in 2008, and the heat value of the song in one month is 10000. If the target description text corresponds to the candidate object 1, the candidate object 1 is the target object characterized by the target description text in the target field, and the candidate object 1 can be determined as the search result of searching the text for the object.

Since it is necessary to determine the candidate objects corresponding to the description texts in the target field, how to determine the candidate objects corresponding to the description texts may be performed by recording the mapping relationship between the candidate objects and the corresponding description texts when determining the respective description texts of each candidate object in the foregoing embodiment, and then completing the conversion from the description texts to the candidate objects by considering the mapping relationship between the candidate objects and the corresponding description texts. Based on this, in an alternative embodiment, the method of object searching further comprises: an object text mapping relationship between each candidate object and the descriptive text of each candidate object in the target domain is created.

The object text mapping relation is used for describing the candidate objects and one-to-one mapping relation between the candidate objects and the description texts of the candidate objects in the target field. Specifically, when determining the respective description text of each candidate object, the terminal creates an object text mapping relation between each candidate object and the description text of each candidate object in the target field according to the record. For example, it is determined that the description text of the candidate object D1 in the target field is the description text E1, and at this time, an object text mapping relationship between the candidate object D1 and the description text E1 in the target field can be created. Similarly, if the description text of the candidate object D2 in the target area is determined to be the description text E2 and the description text of the candidate object D3 in the target area is determined to be the description text E3, then an object text mapping relationship between the candidate object D2 and the description text E2 in the target area and an object text mapping relationship between the candidate object D3 and the description text E3 in the target area may also be created.

Based on this, in a specific embodiment, determining a candidate object corresponding to the target descriptive text in the target field as a search result for searching the text for the object includes: determining a candidate object corresponding to the target description text in the target field based on the object text mapping relation; and determining the candidate objects corresponding to the target description text in the target field as search results of searching the text for the objects.

Specifically, the terminal determines a candidate object corresponding to the target description text in the target field based on the object text mapping relation, namely, the terminal determines a one-to-one mapping relation between the candidate object described by the object text mapping relation and the description text of the candidate object in the target field, and determines the description text included in the target description text and the corresponding candidate object in the target field. Based on this, the candidate object corresponding to the target description text in the target field is determined as the search result of searching the text for the object in the manner described in the foregoing embodiment.

For ease of understanding, if the target description text includes description text E2 and description text E3, since the foregoing example creates the target field: the object text mapping relationship between the candidate object D1 and the description text E1, the object text mapping relationship between the candidate object D2 and the description text E2, and the object text mapping relationship between the candidate object D3 and the description text E3, so that it can be determined that the candidate object corresponding to the target description text in the target field includes: a candidate object D2 corresponding to the descriptive text E2 in the target domain, and a candidate object D3 corresponding to the descriptive text E3 in the target domain.

Description information of the target object as a corresponding object in the target domain will be described below, optionally, in step 904, description information of a candidate object corresponding to the target description text in the target domain is determined as a search result of searching the text for the object.

The description information of the candidate object corresponding to the target description text in the target field is the description information of the candidate object characterized by the target description text in the target field. Specifically, the terminal first determines candidate objects corresponding to the description texts included in the target description texts in the target field, and determines description information of the candidate objects corresponding to the description texts in the target field as target objects, thereby determining the target objects as search results for the object search text.

To facilitate understanding, further introduce with the foregoing example, if the target description text is the description text "song" AAA "of somehow singing every week, belongs to the music album" BBB ", and the music style of the song belongs to hip-hop and ballad, the release time of the song is 10 months 15 days in 2008, and the popularity value of the song in one month is 10000. As can be seen from the foregoing example, the description information of the candidate object 1 includes: song name "AAA", singer name "week someplace", album name "BBB", release time "10 months 15 days in 2008", music style "hip-hop and ballad", heat value 10000. At this time, the description information "song name" AAA ", singer name" somehow around ", album name" BBB ", release time" 10 months and 15 days in 2008 ", music style" hip-hop and ballad ", and heat value 10000" of the candidate object 1 may be determined as the target object represented by the target description text in the target field, that is, the search result of the object search text is the description information of the candidate object 1.

Since it is known in the foregoing embodiment how to determine the description information of the candidate object corresponding to the description text, when determining the description text of each candidate object, the description information represented by the object tag of the candidate object in the target field may be specifically filled into a blank data segment matched with the object tag in the description text template to obtain the description text of each candidate object, and then the mapping relationship between the description information represented by the object tag of the candidate object in the target field and the obtained description text may be recorded, and then the conversion from the description text to the description information may be completed by considering the mapping relationship between the description text and the corresponding description information. Based on this, in an alternative embodiment, the method of object searching further comprises: and creating the descriptive information of each candidate object in the target field and the information text mapping relation between the descriptive text of each candidate object in the target field.

The information text mapping relation is used for describing the description information of the candidate object in the target field and the one-to-one mapping relation between the description text of the candidate object in the target field. Specifically, since the terminal can fill the description information represented by the object tag of the candidate object in the target field into the blank data segment matched by the object tag in the description text template to obtain the respective description text of each candidate object, when the description information is filled into the description text template to obtain the respective description text of each candidate object, the corresponding record creates the description information of each candidate object in the target field and the information text mapping relation between the description text of each candidate object in the target field.

For easy understanding, the description information F1 represented by the object tag of the candidate object D1 in the target domain is filled into the blank data segment matched with the object tag in the description text template, so as to obtain the description text of the candidate object in the target domain as the description text E1, and at this time, an information text mapping relationship between the description information F1 of the candidate object D1 in the target domain and the description text E1 of the candidate object D1 can be created. Similarly, the description information F2 represented by the object tag of the candidate object D2 in the target field is filled into a blank data segment matched with the object tag in the description text template, so that the description text of the candidate object in the target field is obtained as the description text E2, and then an information text mapping relation between the description information F2 of the candidate object D2 in the target field and the description text E2 of the candidate object D2 can be created. And filling description information F3 represented by the object tag of the candidate object D3 in the target field into a blank data segment matched with the object tag in the description text template to obtain a description text E3 of the candidate object in the target field, and creating an information text mapping relation between the description information F3 of the candidate object D3 in the target field and the description text E3 of the candidate object D3.

Based on this, in a specific embodiment, determining a candidate object corresponding to the target descriptive text in the target field as a search result for searching the text for the object includes: determining description information of a candidate object corresponding to the target description text in the target field based on the information text mapping relation; and determining the candidate objects corresponding to the target description text in the target field as search results of searching the text for the objects.

Specifically, based on the information text mapping relation, the description information of the candidate object corresponding to the target description text in the target field is determined, namely, the description information of the candidate object in the target field, which is described by the terminal through the object text mapping relation, is determined, and the one-to-one mapping relation between the description text of the candidate object in the target field and the description information of the candidate object corresponding to the description text included in the target description text in the target field is determined. Based on this, the candidate object corresponding to the target description text in the target field is determined as the search result of searching the text for the object in the manner described in the foregoing embodiment.

For ease of understanding, if the target description text includes description text E2 and description text E3, since the foregoing example creates the target field: an information text mapping relationship between the description information F1 of the candidate object D1 and the description text E1 of the candidate object D1, an information text mapping relationship between the description information F2 of the candidate object D2 and the description text E2 of the candidate object D2, and an information text mapping relationship between the description information F3 of the candidate object D3 and the description text E3 of the candidate object D3. Therefore, the candidate objects corresponding to the target description text in the target field can be determined according to the information text mapping relation, and the candidate objects comprise: description text E2 corresponds to description information F2 in the target area, and description text E3 corresponds to description information F3 in the target area.

If the confirmation of the description information is performed without considering the mapping relationship, since the description information represented by the object tag of the candidate object in the target field is described in the foregoing embodiment, the description information is filled into the blank data segment matched with the object tag in the description text template to obtain the respective description text of each candidate object, in a similar manner, since the blank data segment matched with the object tag exists in the description text template, the description information filled in the blank data segment in the description text is extracted with reference to the blank data segment matched with the object tag exists in the description text template, so as to determine the corresponding description information. In a specific embodiment, determining the description information of the candidate object corresponding to the target description text in the target field as the search result of searching the text for the object includes: extracting description information of the target description text through blank fields in a description text template in the target field to determine the description information of a candidate object corresponding to the target description text in the target field; candidate objects corresponding to the target descriptive text in the target field are determined as search results of searching the text for the objects,

The description information extraction is specifically used for extracting the filled description information from each blank field. Specifically, the terminal firstly determines a description text template in the target field, and then extracts description information of the target description text based on blank fields in the description text template in the target field so as to determine the description information of the candidate object corresponding to the target description text in the target field. That is, the parts filled in by the description information of the candidate objects in the blank fields in the description text template are extracted, and the description information of the candidate objects corresponding to the target description text in the target field is obtained by back-pushing.

For ease of understanding, taking the target domain as an example of the music search domain, the description text templates under the music search domain are: "____ Song" ____ ", belongs to music album" ____ ", and the music style of Song belongs to ____, the release time of Song is ____, and the popularity of Song in one month is ____. "for example, if the target description text is: "Song" AAA "of a singing in week belongs to music album" BBB ", and music style of Song belongs to hip-hop and ballad, release time of Song is 10 months 15 days in 2008, and heat value of Song in one month is 10000. As can be seen from the foregoing analysis, the blank fields in "____ singing song" ____ "are blank fields to be filled with description information, and the text of the target description text here is" song "AAA" of somehow singing, "the description information of the song name" AAA "and the singer name" somehow about "can be known through the blank fields, and similarly, it can also be determined in a similar manner: album name "BBB", release time "10 months 15 days in 2008", music genre "hip hop and ballad", description information of heat value 10000, therefore, description information of candidate objects corresponding to the target description text in the music search field is: song name "AAA", singer name "week someplace", album name "BBB", release time "10 months 15 days in 2008", music style "hip-hop and ballad", heat value 10000.

It will be appreciated that the corresponding examples in the embodiments of the present application are for understanding the present solution, but should not be construed as a specific limitation on the present solution.

In this embodiment, the candidate object may be directly determined as a search result, through which the selected object is explicitly determined, and the description information of the candidate object may also be determined as a search result, where the search result may also result in description information related to the object, thereby determining the search result in multiple manners, so as to improve flexibility of searching the object.

Based on the detailed description of the foregoing embodiments, the following describes a specific application manner when the embodiment of the present application is applied to the field of music search:

1. the object search text is "play a night song of something about the scallop".

According to the embodiment of the application, the search entity word ' Beiqi ' is extracted for the object search text ' play the night song ' of Beiqi ', then the song search of Beiqi ' is carried out based on the search entity word ' Beiqi ' to determine candidate songs corresponding to the search entity word ' Beiqi ', then the respective description texts of the candidate songs corresponding to the ' Beiqi ' are matched semantically through the double-tower model in the embodiment, and then the respective description texts of the candidate songs corresponding to the ' Beiqi ' are ranked, because the moon light and the night are higher in similarity degree semantically, the moon light music and the moon light music play music of the Beiqi can be ranked to the forefront, namely the moon light music and the moon light music play are determined as the search results of the object search text ' play the night song ' of the Beiqi '.

2. The object search text is "i don't want to hear AAA of some version around).

According to the embodiment of the application, firstly, searching the AAA of a version of the text "I don't want to hear" around "and" AAA "for the object, then, searching songs of the" around "and" AAA "based on the search entity words" around "and" AAA "to determine candidate songs corresponding to the search entity words" around "and candidate songs corresponding to the search entity words" AAA ", and then, performing semantic matching through the double-tower model in the embodiment and then sorting, and based on the semantic matching, obtaining whether" I don't want to hear "or not, so that the double-tower model can output the AAA of other singers in front, namely, can determine the search results of the AAA of other singers for the object searching the text" I don't want to hear "of the AAA of the version of the around".

3. The object search text is "i like music is classical type, but i now want to listen to popular music".

According to the embodiment of the application, firstly, the text which is liked by me is of classical type, but I want to listen to popular music, namely, the search entity words are extracted from the popular music, namely, the popular music is the required music style, then the song which is corresponding to the popular music is output in front of the search entity words, namely, the song which is corresponding to the popular music is determined to be the object search text, and then the popular music is the classical type, namely, I want to listen to the search result of the popular music, namely, the popular music is obtained by the search text which is liked by me, namely, the popular music is determined to be the object search text, and the popular music is obtained by the double-tower model in the embodiment.

It will be appreciated that the foregoing examples are provided for the understanding of the specific embodiments of the present solution when applied to the field of music search, but are not to be construed as limiting the present solution in any way.

Based on the foregoing detailed description of the embodiments, a complete flow of the method for searching objects in the embodiments of the present application will be described, and in one embodiment, as shown in fig. 10, a method for searching objects is provided, which is applied to the terminal 102 in fig. 1, for example, and is described by way of illustration, it will be understood that the method may also be applied to the server 104, and may also be applied to a system including the terminal 102 and the server 104, and implemented through interaction between the terminal 102 and the server 104. In this embodiment, the method includes the steps of:

In step 1001, an object search text for a target field is acquired, and a search entity word in the object search text is extracted.

Step 1002, a plurality of selectable objects in a target area, and respective object tags of each selectable object, are acquired.

In step 1003, text similarity matching is performed on each search entity word and each object tag of each selectable object, so as to obtain text similarity between each search entity word and each object tag.

Step 1004, determining the candidate object matched with the search entity word based on the selectable object corresponding to the object label with the text similarity meeting the text screening condition.

In step 1005, a description text template in the target area is determined.

And step 1006, filling description information represented by the object labels of each candidate object in the target field into blank fields matched with the object labels in the description text template to obtain respective description text of each candidate object.

Step 1007, performing semantic matching on each description text and the object search text respectively to obtain semantic similarity between each description text and the object search text.

Step 1008, determining a search result for the object search text based on the target object characterized by the descriptive text whose semantic similarity satisfies the semantic filtering condition.

It should be understood that the specific implementation of steps 1001 to 1008 is similar to the previous embodiments, and will not be repeated here.

It should be understood that, although the steps in the flowcharts related to the embodiments described above are sequentially shown as indicated by arrows, these steps are not necessarily sequentially performed in the order indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in the flowcharts described in the above embodiments may include a plurality of steps or a plurality of stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of the steps or stages is not necessarily performed sequentially, but may be performed alternately or alternately with at least some of the other steps or stages.

Based on the same inventive concept, the embodiment of the application also provides an object searching device for realizing the above related object searching method. The implementation of the solution provided by the device is similar to the implementation described in the above method, so the specific limitation in the embodiments of the object searching device or devices provided below may refer to the limitation of the object searching method hereinabove, and will not be repeated herein.

In one embodiment, as shown in fig. 11, there is provided an object searching apparatus including: an object search text acquisition module 1102, a descriptive text determination module 1104, a semantic similarity matching module 1106, and a search result determination module 1108, wherein:

an object search text obtaining module 1102, configured to obtain an object search text for a target field, and extract a search entity word in the object search text;

A description text determining module 1104, configured to find candidate objects that match the search entity word from the target domain, and determine respective description text of each candidate object, where the description text is generated based on structural information of the candidate object in the target domain;

The semantic similarity matching module 1106 is configured to perform semantic matching on each description text and the object search text, so as to obtain semantic similarity between each description text and the object search text;

The search result determining module 1108 is configured to determine a search result for the object search text based on the target object characterized by the descriptive text whose semantic similarity satisfies the semantic filtering condition.

In one embodiment, the descriptive text determination module 1104 is specifically configured to obtain a plurality of selectable objects in the target area, and a respective object tag of each selectable object; respectively carrying out text similarity matching on each search entity word and each object label of each selectable object to obtain text similarity of each search entity word and each object label; and determining the selectable object corresponding to the object label with the text similarity meeting the text screening condition as a candidate object matched with the search entity word.

In one embodiment, the descriptive text determining module 1104 is specifically configured to determine a descriptive text template in the target field, where the descriptive text template includes blank fields that are respectively matched by a plurality of object tags; and filling description information represented by the object labels of each candidate object in the target field into blank fields matched with the object labels in the description text template to obtain respective description text of each candidate object.

In one embodiment, the search result determining module 1108 is specifically configured to determine, as the target description text, the description text whose semantic similarity reaches the semantic similarity threshold; search results for searching text for objects are determined based on target objects characterized by the target descriptive text in the target domain.

In one embodiment, the search result determining module 1108 is specifically configured to perform ranking based on each semantic similarity, so as to obtain a semantic similarity ranking result; selecting a preset number of description texts as target description texts based on semantic similarity sorting results; search results for searching text for objects are determined based on target objects characterized by the target descriptive text in the target domain.

In one embodiment, the search result determining module 1108 is specifically configured to determine a candidate object corresponding to the target description text in the target field as a search result for searching the text for the object.

In one embodiment, the search result determining module 1108 is specifically configured to determine description information of a candidate object corresponding to the target description text in the target field as a search result of searching the text for the object.

In one embodiment, the object searching apparatus further includes a mapping relation creating module;

The mapping relation creation module is used for creating an object text mapping relation between each candidate object and the description text of each candidate object in the target field;

the search result determining module 1108 is specifically configured to determine, based on the object text mapping relationship, a candidate object corresponding to the target description text in the target field; and determining the candidate objects corresponding to the target description text in the target field as search results of searching the text for the objects.

In one embodiment, the mapping relation creating module is used for creating the information text mapping relation between the description information of each candidate object in the target field and the description text of each candidate object in the target field;

The search result determining module 1108 is specifically configured to determine, based on the information text mapping relationship, description information of a candidate object corresponding to the target description text in the target field; and determining the candidate objects corresponding to the target description text in the target field as search results of searching the text for the objects.

In one embodiment, the search result determining module 1108 is specifically configured to extract description information of the target description text through each blank field in the description text template in the target domain, so as to determine description information of a candidate object corresponding to the target description text in the target domain; and determining the candidate objects corresponding to the target description text in the target field as search results of searching the text for the objects.

In one embodiment, the search result determining module 1108 is specifically configured to sort each description text from large to small through the double-tower model according to the semantic similarity between each description text and the object search text, so as to obtain a sequential text set; selecting a target object characterized by descriptive text with semantic similarity meeting semantic screening conditions from the ordered text set, and determining a search result for searching the text for the object based on the target object.

In one embodiment, the search result determination module 1108 is specifically configured to extract object search text semantic features of the object search text through a problem vector tower in the dual-tower model; extracting the semantic features of the descriptive text corresponding to the descriptive text of each candidate object through an answer vector tower in the double-tower model; and calculating cosine distances between semantic features of each description text and semantic features of the object search text through the semantic features of each description text in the double-tower model, and determining the obtained cosine distance result as the semantic similarity of each description text and the object search text.

In one embodiment, the object searching device further comprises a voice broadcasting module;

the object search text acquisition module 1102 is configured to acquire voice search data for a target field, and perform voice text conversion processing on the voice search data to obtain an object search text;

and the voice broadcasting module is used for outputting a search result of searching the text for the object in a voice broadcasting mode.

In one embodiment, a computer device is provided, which may be a server or a terminal, and in this embodiment, the computer device is taken as an example to describe the terminal, and the internal structure diagram may be as shown in fig. 12. The computer device includes a processor, a memory, an input/output interface, a communication interface, a display unit, and an input means. The processor, the memory and the input/output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input/output interface. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input/output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless mode can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a method of object searching. The display unit of the computer equipment is used for forming a visual picture, and can be a display screen, a projection device or a virtual reality imaging device, wherein the display screen can be a liquid crystal display screen or an electronic ink display screen, the input device of the computer equipment can be a touch layer covered on the display screen, can also be a key, a track ball or a touch pad arranged on a shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.

It will be appreciated by those skilled in the art that the structure shown in FIG. 12 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

In an embodiment, there is also provided a computer device comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the method embodiments described above when the computer program is executed.

In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored which, when executed by a processor, carries out the steps of the method embodiments described above.

In an embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the steps of the method embodiments described above.

It should be noted that, the object information (including, but not limited to, object device information, object personal information, etc.) and the data (including, but not limited to, data for analysis, stored data, presented data, etc.) related to the present application are both information and data authorized by the object or sufficiently authorized by each party, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related countries and regions.

Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, database, or other medium used in embodiments provided herein may include at least one of non-volatile and volatile memory. The nonvolatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical Memory, high density embedded nonvolatile Memory, resistive random access Memory (ReRAM), magneto-resistive random access Memory (Magnetoresistive Random Access Memory, MRAM), ferroelectric Memory (Ferroelectric Random Access Memory, FRAM), phase change Memory (PHASE CHANGE Memory, PCM), graphene Memory, and the like. Volatile memory can include random access memory (Random Access Memory, RAM) or external cache memory, and the like. By way of illustration, and not limitation, RAM can be in various forms such as static random access memory (Static Random Access Memory, SRAM) or dynamic random access memory (Dynamic Random Access Memory, DRAM), etc. The databases referred to in the embodiments provided herein may include at least one of a relational database and a non-relational database. The non-relational database may include, but is not limited to, a blockchain-based distributed database, and the like. The processor referred to in the embodiments provided in the present application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, or the like, but is not limited thereto.

The technical feature information of the above embodiments may be arbitrarily combined, and for brevity of description, all possible combinations of the technical feature information in the above embodiments are not described, however, as long as there is no contradiction between the combinations of the technical feature information, they should be considered as the scope of the description.

The foregoing examples illustrate only a few embodiments of the application and are described in detail herein without thereby limiting the scope of the application. It should be noted that it will be apparent to those skilled in the art that several variations and modifications can be made without departing from the spirit of the application, which are all within the scope of the application. Accordingly, the scope of the application should be assessed as that of the appended claims.

Claims

1. An object searching method, comprising:

Acquiring an object search text aiming at a target field, and extracting a search entity word in the object search text;

and determining a search result of searching text for the object based on the target object characterized by the descriptive text of which the semantic similarity meets the semantic screening condition.

2. The method of claim 1, wherein said finding candidate objects matching the search entity word from the target domain comprises:

acquiring a plurality of selectable objects in the target field and respective object labels of each selectable object;

Respectively carrying out text similarity matching on each search entity word and each object label of each selectable object to obtain text similarity of each search entity word and each object label;

And determining the candidate object matched with the search entity word based on the selectable object corresponding to the object label of which the text similarity meets the text screening condition.

3. The method of claim 1, wherein said determining respective descriptive text for each of said candidate objects comprises:

Determining a description text template in the target field, wherein the description text template comprises blank fields matched with a plurality of object labels respectively;

and filling description information represented by the object label of each candidate object in the target field into blank fields matched with the object label in the description text template to obtain respective description text of each candidate object.

4. The method of claim 1, wherein the determining a search result for searching text for a target object characterized by descriptive text for which the semantic similarity satisfies a semantic filtering condition based on the object comprises:

Determining the description text of which the semantic similarity reaches a semantic similarity threshold as a target description text;

Based on a target object characterized by the target descriptive text in the target domain, a search result for searching text for the object is determined.

5. The method of claim 1, wherein the determining a search result for searching text for a target object characterized by descriptive text for which the semantic similarity satisfies a semantic filtering condition based on the object comprises:

sorting is carried out based on each semantic similarity, and a semantic similarity sorting result is obtained;

selecting a preset number of description texts as target description texts based on the semantic similarity sorting result;

6. The method of claim 4 or 5, wherein the determining search results for searching text for the object based on the target object characterized by the target descriptive text in the target domain comprises:

Determining candidate objects corresponding to the target description text in the target field as search results of searching text for the objects;

Or alternatively, the first and second heat exchangers may be,

And determining the description information of the candidate object corresponding to the target description text in the target field as a search result of searching the text for the object.

7. The method of claim 6, wherein the method further comprises:

creating an object text mapping relation between each candidate object and the descriptive text of each candidate object in the target field;

The determining the candidate object corresponding to the target description text in the target field as the search result of searching the text for the object comprises the following steps:

Determining a candidate object corresponding to the target description text in the target field based on the object text mapping relation;

And determining a candidate object corresponding to the target description text in the target field as a search result of searching text for the object.

8. The method of claim 6, wherein the method further comprises:

Creating the description information of each candidate object in the target field and the information text mapping relation between the description text of each candidate object in the target field;

the determining the description information of the candidate object corresponding to the target description text in the target field as the search result of searching the text for the object comprises the following steps:

determining the description information of the candidate object corresponding to the target description text in the target field based on the information text mapping relation;

9. The method according to claim 6, wherein determining the description information of the candidate object corresponding to the target description text in the target field as the search result of searching text for the object includes:

Extracting description information of the target description text through blank fields in a description text template in the target field to determine the description information of a candidate object corresponding to the target description text in the target field;

10. The method of claim 1, wherein the determining a search result for searching text for a target object characterized by descriptive text for which the semantic similarity satisfies a semantic filtering condition based on the object comprises:

According to the semantic similarity between each description text and the object search text, ordering each description text from large to small through a double-tower model to obtain a sequential text set;

Selecting a target object characterized by the descriptive text with the semantic similarity meeting semantic filtering conditions from the ordered text set, and determining a search result of searching text for the object based on the target object.

11. The method according to claim 10, wherein said semantically matching each of said descriptive texts with said object search text to obtain a semantic similarity between each of said descriptive texts and said object search text, comprises:

Extracting object searching text semantic features of the object searching text through a problem vector tower in the double-tower model;

extracting the semantic features of the descriptive text corresponding to the descriptive text of each candidate object through an answer vector tower in the double-tower model;

And calculating cosine distances between semantic features of each description text and semantic features of the object search text through the semantic features of the double-tower model, and determining the obtained cosine distance result as the semantic similarity between each description text and the object search text.

12. The method of claim 1, wherein the obtaining the object search text for the target area comprises:

acquiring voice search data aiming at a target field, and performing voice text conversion processing on the voice search data to obtain an object search text;

the method further comprises the steps of:

and outputting a search result of searching the text for the object in a voice broadcasting mode.

13. An object searching apparatus, the apparatus comprising:

the description text determining module is used for searching candidate objects matched with the search entity words from the target field, and determining respective description text of each candidate object, wherein the description text is generated based on structural information of the candidate object in the target field;

And the search result determining module is used for determining a search result of searching the text for the object based on the target object characterized by the descriptive text of which the semantic similarity meets the semantic screening condition.

14. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 12 when the computer program is executed.

15. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any of claims 1 to 12.

16. A computer program product comprising a computer program, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any one of claims 1 to 12.