CN111552884A

CN111552884A - Method and apparatus for content recommendation

Info

Publication number: CN111552884A
Application number: CN202010402143.0A
Authority: CN
Inventors: 张晗
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-05-13
Filing date: 2020-05-13
Publication date: 2020-08-18
Anticipated expiration: 2040-05-13
Also published as: CN111552884B

Abstract

A method for content recommendation is described, comprising: acquiring historical behavior data and a user portrait of a user; acquiring content data of a plurality of content files, wherein the content data of each content file comprises at least one content attribute of each content file; determining feature values of a plurality of features characterizing the user's interest in the each content file based on the at least one content attribute, the user representation, and the historical behavior data of the each content file, the plurality of features including click behavior features; determining a score for the each content file based at least on feature values of the plurality of features; selecting a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files.

Description

Method and apparatus for content recommendation

Technical Field

The present disclosure relates to the technical field of personalized recommendation, and in particular to a method and apparatus for content recommendation.

Background

With the development of internet technology, users can watch or listen to different types of contents such as video, audio, pictures and texts, albums and the like on various websites on the internet. Meanwhile, the website server can recommend the content meeting the user interest to the user in a personalized mode by deeply mining the user interest so as to improve the click rate of the user on the content.

Disclosure of Invention

In the related art, content is generally recommended to a user based on historical interests of the user in a past period of time, however, this easily causes that long-term interests of the user are heavily dependent, short-term interests of the user are insufficiently delineated, and changes in interests of the user cannot be captured in time when the content is recommended to the user. For example, from the long-term interest of the user, the user has a strong interest in "ironmen" and a weak interest in "Nezha", because the user has historically seen more "ironmen" and less "Nezha" content; from the user's short-term interest, the "ironmen" content was presented to the user 10 times most recently but only once, while the "Nezha" content was presented 3 times but clicked 3 times, due to the recent comparison of the movie "magic child descending Nezha". Obviously, the user is more interested in the content about the Nezha in a short period, but still a great amount of content about the ironmen is recommended to the user during content recommendation, but the content about the Nezha is rarely recommended to the user, so that the problems of poor content recommendation efficiency, low accuracy and poor user experience are caused.

In view of the above, the present disclosure provides methods and apparatus, computing devices, and computer-readable storage media for content recommendation, which desirably overcome some or all of the above-referenced deficiencies and possibly others.

According to a first aspect of the present disclosure, there is provided a method for content recommendation, comprising: obtaining historical behavior data of a user and a user representation, wherein the historical behavior data comprises data related to historical clicks of the user on a content file, the user representation comprises a plurality of interest classifications of the user, and each interest classification comprises a content attribute of the content file; acquiring content data of a plurality of content files, wherein the content data of each content file comprises at least one content attribute of each content file; determining feature values of a plurality of features for characterizing the user's interest in each content file based on the at least one content attribute of each content file, the user representation, and historical behavior data, the plurality of features including click behavior features related to the number of occurrences of each of the at least one content attribute in historical clicks of the user within a preset window of the number of recent clicks; determining a score for the each content file based at least on feature values of the plurality of features; selecting a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files.

In some embodiments, obtaining historical behavioral data and a user representation of the user comprises: in response to receiving a current content recommendation request for a user, historical behavior data and a user representation of the user are obtained.

In some embodiments, obtaining historical behavioral data and a user representation of a user comprises: acquiring historical behavior data of a user; and acquiring a user portrait of the user based on the historical behavior data of the user.

In some embodiments, each interest category further includes an interestingness corresponding to a content attribute of the content file, and the click behavior feature includes: the content attribute and the corresponding combination feature of the occurrence frequency of the content attribute in the historical clicks of the user in each corresponding preset closest click number quantum window in at least one preset closest click number quantum window, wherein the at least one preset closest click number quantum window is a sub-window of the preset closest click number window; and a respective combination feature of said each content attribute, a ranking of interestingness of said each content attribute in its corresponding interest category, and said number of occurrences of said each content attribute.

In some embodiments, the plurality of features further includes a click time feature associated with a click time of the user's historical clicks on the content file having the each content attribute within the preset number of most recent clicks window.

In some embodiments, the click time characteristics include: the time interval between the click time of the content file with each content attribute clicked in the preset latest click number window in the preset order and the time of the current content recommendation request; and the time interval between the click time of the content file with each content attribute clicked in the preset latest click number window in the preset order and the time of the current content recommendation request, and the combination characteristic of each content attribute.

In some embodiments, the click time feature further comprises: the content recommendation request of the content file with each content attribute is clicked in the preset latest click quantity window in the preset order and the request number interval of the current content recommendation request; and the content file with each content attribute is clicked in the preset latest click quantity window in a preset sequence, the request number interval of the content recommendation request and the current content recommendation request is positioned, and the combination characteristic of each content attribute is obtained.

In some embodiments, the historical behavioural data further comprises data relating to historical presentation of content files to the user, and the plurality of characteristics further comprises a presentation time characteristic relating to a presentation time at which a content file having each of the at least one content attribute was presented to the user in the historical presentation for a preset recent period.

In some embodiments, the presentation time characteristics include: a time interval between a presentation time at which the content file having the each content attribute is presented in a predetermined order in the historical presentation of the preset latest period and a time of a current content recommendation request; and a time interval between a presentation time at which the content file having said each content attribute is presented in a predetermined order in the historical presentation of said preset latest period and a time of a current content recommendation request, and a combination characteristic of both of said each content attribute.

In some embodiments, presenting the temporal features further comprises: the content recommendation request of the content file with each content attribute is presented in a preset order in the historical presentation of the preset recent period and the request number interval of the current content recommendation request; and the interval of the number of requests of the content recommendation request and the current content recommendation request of the content file with each content attribute when the content file is presented in the preset latest period of history presentation in the preset order, and the combined characteristics of the content file with each content attribute.

In some embodiments, obtaining content data for a plurality of content files comprises: acquiring content data of the plurality of content files based on the user image of the user.

In some embodiments, determining feature values for a plurality of features characterizing the user's interest in the each content file based on the at least one content attribute of the each content file, the user representation, and historical behavior data comprises: obtaining an original value for each of the plurality of features; and obtaining the characteristic value of each characteristic based on the original value of each characteristic and the corresponding characteristic name.

In some embodiments, obtaining the feature value of each feature based on the original value of each feature and the corresponding feature name includes: hashing the original value of each feature to obtain a first hash value of each feature; hashing the feature name character string of each feature to obtain a second hash value of each feature; and obtaining the characteristic value of each characteristic based on the first hash value and the second hash value of each characteristic.

In some embodiments, determining the score for each of the content files based at least on the feature values of the plurality of features comprises: inputting at least the feature values of the plurality of features into the trained intelligent scoring model to obtain a score of each content file; the trained intelligent scoring model is obtained by training according to positive sample data and negative sample data; in the process of presenting the plurality of content files to the user, the characteristic values of a plurality of characteristics representing the interest of the user in the clicked content files in the plurality of presented content files are used as positive sample data, and the characteristic values of a plurality of characteristics representing the interest of the user in the content files which are not clicked in the plurality of presented content files are used as negative sample data.

In some embodiments, selecting a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files comprises: based on the scores of the content files, sequencing the content files according to the scores to obtain an ordered sequence of the content files; a predetermined number of content files are selected for recommendation to the user starting with the first content file in the ordered sequence of content files.

According to a second aspect of the present disclosure, there is provided a method for content recommendation, comprising: obtaining a predetermined number of content files selected according to the method of the first aspect of the disclosure for recommendation to a user; presenting the predetermined number of content files.

According to a third aspect of the present disclosure, there is provided an apparatus for content recommendation, comprising: a first obtaining module configured to obtain historical behavior data of a user, the historical behavior data including data related to historical clicks of a content file by the user, and a user representation including a plurality of interest classifications of the user, each interest classification including a content attribute of a content file; a second obtaining module configured to obtain content data of a plurality of content files, the content data of each content file including at least one content attribute of the each content file; a first determination module configured to determine feature values of a plurality of features characterizing the user's interest in each content file based on at least one content attribute of the each content file, the user representation, and historical behavior data, the plurality of features including click behavior features related to the number of occurrences of each content attribute in the at least one content attribute in historical clicks of the user within a preset window of the number of most recent clicks; a second determination module configured to determine a score for the each content file based at least on feature values of the plurality of features; a selection module configured to select a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files.

According to a fourth aspect of the present disclosure, there is provided an apparatus for content recommendation, comprising: a content file acquisition module configured to acquire a predetermined number of content files for recommendation to a user from a device according to a third aspect of the present disclosure; a presentation module configured to present the predetermined number of content files.

According to a fifth aspect of the present disclosure, there is provided a computing device comprising a processor; and a memory configured to have computer-executable instructions stored thereon that, when executed by the processor, perform any of the methods described above.

According to a sixth aspect of the present disclosure, there is provided a computer-readable storage medium storing computer-executable instructions that, when executed, perform any of the methods described above.

In the method and the device for recommending content, which are claimed by the present disclosure, the click behavior characteristics related to the occurrence frequency of each content attribute of the user in the preset recent click number window are used during content recommendation, so that the short-term interest of the user can be fully embodied and the change of the user interest can be quickly embodied during content recommendation, thereby greatly improving the accuracy of content recommendation, and improving key indexes such as the click rate and the click amount of recommended content.

These and other advantages of the present disclosure will become apparent from and elucidated with reference to the embodiments described hereinafter.

Drawings

Embodiments of the present disclosure will now be described in more detail and with reference to the accompanying drawings, in which:

fig. 1 illustrates an exemplary application scenario in which a technical solution according to an embodiment of the present disclosure may be implemented;

FIG. 2 shows a schematic flow diagram of a method for content recommendation according to one embodiment of the present disclosure;

FIG. 3 shows a schematic diagram of an interest classification of a hierarchical structure of users according to one embodiment of the present disclosure;

FIG. 4 illustrates a schematic diagram of determining click behavior characteristics according to one embodiment of the present disclosure;

FIG. 5 illustrates an example diagram of determining click behavior characteristics according to one embodiment of the present disclosure;

FIG. 6 illustrates a schematic diagram of determining a click time feature and a presentation time feature in accordance with one embodiment of the present disclosure;

FIG. 7 shows a schematic flow chart diagram of a method for content recommendation according to another embodiment of the present disclosure;

FIG. 8 shows an architectural diagram of a method for content recommendation, according to one embodiment of the present disclosure;

FIG. 9 shows an architectural diagram of ordering videos according to one embodiment of the present disclosure;

FIG. 10 illustrates an overall flow of click rate prediction according to one embodiment of the present disclosure;

FIG. 11 shows an architectural diagram of model training according to one embodiment of the present disclosure;

FIG. 12 shows a schematic diagram of presenting recommended content files for a method for content recommendation according to an embodiment of the present disclosure;

FIG. 13 illustrates an exemplary block diagram of an apparatus for content recommendation according to one embodiment of the present disclosure;

FIG. 14 illustrates an exemplary block diagram of an apparatus for content recommendation according to another embodiment of the present disclosure; and

fig. 15 illustrates an example system that includes an example computing device that represents one or more systems and/or devices that may implement the various techniques described herein.

Detailed Description

The following description provides specific details of various embodiments of the disclosure so that those skilled in the art can fully understand and practice the various embodiments of the disclosure. It is understood that aspects of the disclosure may be practiced without some of these details. In some instances, well-known structures or functions are not shown or described in detail in this disclosure to avoid obscuring the description of the embodiments of the disclosure by these unnecessary descriptions. The terminology used in the present disclosure should be interpreted in its broadest reasonable manner, even though it is being used in conjunction with a particular embodiment of the present disclosure.

First, some terms referred to in the embodiments of the present application are explained so that those skilled in the art can understand that:

the content is as follows: any information or data that may be viewed, listened to, or perceived by a user herein, which may be, for example, video, audio, graphics, an album, or the like; accordingly, a content file refers to a carrier for carrying the content, such as a video file, an audio file, a web page, and so forth;

the combination characteristics are as follows: is to form cross-over features by combining individual features;

user portrait: a user model built on a series of attribute data may include a plurality of interest classifications for the user, which may be abstracted, for example, from historical behavioral data for the user. Each interest category includes content attributes of the content files (which may include, by way of example, content attributes of content files that the user has historically clicked on), and optionally includes interestingness corresponding to the content attributes of the content files, and fig. 3 shows a schematic diagram of a user representation. It should be noted that the user representation may be determined in various other ways, such as, but not limited to, by obtaining an interest classification entered by the user.

Fig. 1 illustrates an exemplary application scenario 100 in which a technical solution according to an embodiment of the present disclosure may be implemented. As shown in fig. 1, the application scenario 100 includes a server 110,

terminals

120, 130, and a network 140.

Terminals

120, 130 are communicatively coupled to server 110 via network 140. The user may view content, which may be, for example, video, audio, text, etc., through an application or client on the

terminal

120, 130. The server can recommend the content meeting the user interest to the user in a personalized mode through an application program or a client of the terminal by deeply mining the user interest.

For convenience of description, the user opens or logs in the corresponding client viewing content on the terminal 120 is taken as an example for description, and it should be understood that the user opens or logs in the corresponding client viewing content on the terminal 130 has the same effect. As an example, when the user opens or logs in a corresponding client on the terminal 120 to view the content, the terminal 120 may transmit a content recommendation request for the user to the server 110 through the network 140. The server 110 may obtain historical behavior data and a user portrait of the user after receiving the content recommendation request, and obtain content data of a plurality of content files, the content data of each content file including at least one content attribute of the each content file; determining feature values of a plurality of features for characterizing the user's interest in each content file based on the at least one content attribute of each content file, the user representation, and historical behavior data, the plurality of features may include, for example, a click behavior feature related to a number of occurrences of each content attribute in the at least one content attribute in historical clicks of the user within a preset window of a number of recent clicks. Server 110 may then determine a score for each of the content files based at least on the feature values of the plurality of features, then select a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files, and may transmit the predetermined number of content files to the client on terminal 120 via network 140 for presentation to the user to facilitate content recommendation to the user.

Alternatively, server 110 may be a content server of a content provider, a device associated with the content server, a system on a chip, and/or any other suitable computing device or computing system. The server 110 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 basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, a middleware service, a domain name service, a security service, a CDN (Content Delivery Network), a big data and artificial intelligence platform, and the like. The

terminals

120 and 130 may be, but are not limited to, a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, and the like. The

terminals

120 and 130 and the server 110 may be directly or indirectly connected through wired or wireless communication, and the application is not limited thereto. The network 140 may be, for example, a Wide Area Network (WAN), a Local Area Network (LAN), a wireless network, a public telephone network, an intranet, or any other type of network known to those skilled in the art. It should also be noted that the scenario described above is only one example in which the embodiments of the present disclosure may be implemented, and is not limiting.

Fig. 2 illustrates a schematic flow diagram of a method 200 for content recommendation according to one embodiment of the present disclosure. The method 200 may be implemented, for example, on the server 110 shown in fig. 1. As shown in fig. 2, the method 200 may include the following steps.

In step 201, historical behavior data and a user representation of a user are obtained. The historical behavior data of the user includes data related to historical clicks of content files by the user and optionally may include data related to historical presentations of content files presented to the user. Historical behavior data is typically recorded or saved in the form of a log. The data related to the user's historical clicks on the content file is, for example, data related to the user's clicks on the content file within a predetermined period of time in the past (e.g., 3 days in the past), data related to the user's clicks on the content file within a preset number of most recent clicks window in the past (e.g., 128 clicks in the last), and the like. Historical presentation-related data that presents content files to the user is, for example, data that presents content files to the user within a predetermined period of time in the past (e.g., within the past 3 days), data that presents content files to the user within a preset number of recent clicks window in the past (e.g., within the last 128 clicks), and so forth. Generally, data related to the user's historical clicks on a content file is saved in units of one click, which saves an identifier of the content file, a click time, and the like; the history of presenting content files to the user (also referred to as history presentation) is saved in units of one content recommendation request, which saves an identifier of a content file, a presentation time, etc., and a plurality of content files may be requested per content recommendation request, for example. Each content file has at least one content attribute, which may be queried, for example, with an identifier of the content file. The user representation includes a plurality of interest classifications for the user, each interest classification including a content attribute of a content file. Each interest category may further include an interest level corresponding to a content attribute of the content file. The interestingness corresponding to the content attribute can be determined according to the number of clicks or the click rate of the user on the content file with the content attribute. The higher the number of clicks or the click rate of a user on a content file having a certain content attribute is, the higher the interest level corresponding to the content attribute is.

It should be noted that "clicking" as described herein may refer to a user's conventional click on a content file (e.g., via a mouse, etc.), may also refer to various forms of input (e.g., via voice input, gesture input, visual input) that represent a user's click or confirmation on a content file, and so forth.

In some embodiments, the plurality of interest classifications of the user have a hierarchical structure, for example, a first class classification, a second class classification, a label classification, and the like may be included in sequence from an upper layer to a lower layer. By way of example, FIG. 3 illustrates a schematic diagram of interest classification for a hierarchical structure of users. As shown in FIG. 3, the user has two content attributes "entertainment" and "sports" in a class of categories, numbered in parentheses for their corresponding interestingness, 0.6 and 0.4, respectively. The content attribute of the first-level classification, namely entertainment, is provided with the content attributes of two second-level classifications of 'movie' and 'hedyotis', the content attribute of the second-level classification, namely 'movie' is provided with the content attributes of two label classifications of 'ironman' and 'Nezha', and the numbers in the brackets are the corresponding interestingness of the contents. It should be noted that the number of levels of the hierarchical structure is not limiting, e.g., there may also be three levels of classification, four levels of classification, etc., between the two levels of classification and the label classification.

In some embodiments, a current content recommendation request for a user may be received prior to obtaining historical behavior data and a user representation of the user, namely: historical behavior data and a user representation of a user are obtained in response to receiving a current content recommendation request for the user. As described above, the current content recommendation request for the user may be sent when the user opens or logs in the corresponding client to view content, or wants to view new content, although this is not limiting.

The user representation is typically determined based on historical behavioral data of the user. The user profile may be determined in advance from the overall historical behavior data of the user, and then the historical behavior data of the user (which may be, for example, a part of the overall historical behavior data of the user) and the user profile required in the content recommendation may be obtained. In some embodiments, in obtaining historical behavior data and a user representation of a user, the historical behavior data of the user may be obtained first, and then the user representation of the user may be obtained based on the obtained historical behavior data of the user, which is not limiting.

In step 202, content data of a plurality of content files is obtained, the content data of each content file comprising at least one content attribute of said each content file. The content attributes may be used to characterize a particular aspect of the content file, which may be content attributes in interest categories as described above, e.g., content attributes of a primary category, content attributes of a secondary category, content attributes of a tag category, and so forth.

In some embodiments, content data for a plurality of content files may be obtained based on a user image. For example, if the primary classification included in the user portrait has a "sports" content attribute, the content data of the content file related to "sports" can be acquired, and the content file related to "sports" is taken as a candidate content file for recommendation to the user, which can narrow the range of acquiring the content file during content recommendation, save processing resources, and improve the content recommendation efficiency.

In step 203, feature values of a plurality of features characterizing the user's interest in each content file are determined based on at least one content attribute of the each content file, the user representation, and historical behavior data. The plurality of features includes a click behavior feature that is a series of features related to a number of occurrences of each of the at least one content attribute in historical clicks by the user within a preset window of a number of recent clicks. The preset number of most recent clicks window may be set as desired, for example, may be set to be 128 most recent clicks. That is, the click behavior feature may be, for example, a feature related to the number of occurrences of each of the at least one content attribute in the user's last 128 clicks.

In some embodiments, the click behavior features may include: the content attribute and the corresponding combination feature of the occurrence frequency of the content attribute in the historical clicks of the user in each corresponding preset closest click number quantum window in at least one preset closest click number quantum window, wherein the at least one preset closest click number quantum window is a sub-window of the preset closest click number window; the content attribute, the rank of the interestingness of each content attribute in the corresponding interest classification of each content attribute, and the corresponding combination characteristics of the occurrence times of each content attribute in the historical clicks of the user in each corresponding preset closest click number sub-window in at least one preset closest click number sub-window. By way of example, the content attributes in the interest category may be ranked from high to low according to their corresponding interestingness, and then the ranking of the interestingness of each content attribute therein is obtained therefrom (i.e., its corresponding interestingness ranks in the second place).

Taking the latest click number window as the latest 128 clicks as an example, the at least one preset latest click number quantum window may be, for example, 4 preset latest click number quantum windows, which are respectively the sub-windows of the latest 5 clicks, the latest 20 clicks, the latest 50 clicks, and the latest 100 clicks. It should be noted that the number of at least one preset closest-click number quantum window is not limiting, and the number of preset closest clicks included in each sub-window is also not limiting, e.g., there may be such a preset closest-click number quantum window as the last 30 clicks.

Table 1 below shows click behavior characteristics in the case where there are 4 preset closest-click number sub-windows (the last 5 clicks, the last 20 clicks, the last 50 clicks, and the last 100 clicks, respectively) and three content attributes (a content attribute of a first class, a content attribute of a second class, and a content attribute of a tag class).

As shown in table 1, taking the sub-window of the last 5 clicks and the content attribute of the first class (which may be, for example, "sports" as shown in fig. 3) as an example, the click behavior feature includes a corresponding combination feature of both the content attribute of the first class and the occurrence number of the content attribute of the first class in the last 5 clicks of the user; and the content attribute of the first-level classification, the ranking of the interestingness of the content attribute of the first-level classification in the corresponding interest classification (namely, the first-level classification) of the content attribute of the first-level classification, and the corresponding combination characteristics of the occurrence times.

As an example, as shown in fig. 4, the content attributes in the tag categories involved in the last 5 clicks of the user are, from far to near, respectively, a handshah, a myzha, a grand red, a hogger, and a myzha, where the interestingness of the "myzha" content attribute is 0.001 and is ranked as 7 in the tag category. As can be seen from fig. 4, the "myzha" content attribute appears 2 times in the user's last 5 clicks. Thus, in determining the characteristic values of the features characterizing the user's interest in content files comprising the "Nezha" content attribute, the click behavior features may comprise: the combined characteristics of the "Nezha" content attribute and its number of occurrences in the last 5 clicks, 2; and a combined feature of ranking 7 of the interestingness of the "Nezha" content attribute in its corresponding interest category (i.e., label category) and the number of times it occurred in the last 5 clicks, 2.

By using the click behavior characteristics related to the occurrence frequency of each content attribute in the at least one content attribute in the historical clicks of the user in the preset recent click number window, the effects of depicting the short-term interest of the user and quickly capturing the change of the interest of the user by using the recent click number of the user are achieved, the accuracy of content recommendation can be greatly improved, and the key indexes of the recommended content, such as click rate, click amount and the like, are improved. In particular, the preset number of recent clicks sub-window can be optionally utilized to achieve the effects of more finely depicting the short-term interest of the user and more quickly capturing the change of the user interest, so that the accuracy of content recommendation is greatly improved.

In some embodiments, the plurality of features further includes a click time feature. The click time characteristics are, for example, a series of characteristics related to the click time of the user on the content file having each content attribute within the preset number of most recent clicks window. The preset number of most recent clicks window may be set as desired, for example, it may also be set as the most recent 128 clicks, although this is not restrictive. In this case, the click time characteristic may be, for example, a characteristic related to click times of historical clicks of the user on the content file having the each content attribute in the last 128 clicks.

Consider the following example, as shown in fig. 5, if a user clicks on a video with a tag classification with content attribute of "Nezha" at 12 am Monday, and then does not request a recommended video until such time as the recommended video is re-requested at 12 am Monday, his last click is the click at 12 am Monday. Further, if the user clicks a video of which content attribute is "Nezha" in a tag classification at noon on weekdays, when he requests a recommended video again 10 minutes later, the latest click is a click 10 minutes ago. The effect of these two clicks with different time intervals is the same from the previous click behavior characteristics. In the embodiment, the click time characteristics are used, so that the difference between the click times of the two clicks can be reflected, and the accuracy of short-term interest portrayal of the user is further improved.

In some embodiments, the click time characteristics include: the time interval between the click time of the content file with each content attribute clicked in the preset latest click number window in the preset order and the time of the current content recommendation request; the time interval between the click time when the content file with each content attribute is clicked in the preset number-of-most-clicks window in the preset sequence and the time of the current content recommendation request, and the combined characteristics of the content file with each content attribute.

As an example, the predetermined order may be specified as needed, and may be, for example, the most recent 1 time, the most recent 2 nd time, the most recent 1 time, or the like. Table 2 below shows click time characteristics in the case where the preset number-of-most-clicks window is the most recent 128 clicks, the predetermined order is the most recent 1 and the most distant 1 (i.e., the 1 st and 128 th clicks), and there are three content attributes (a content attribute of one primary classification, a content attribute of one secondary classification, a content attribute of one tag classification).

As shown in table 2, taking the content attributes of the first-level classification as an example, the click time characteristics include: the time interval between the click time of the content file with the content attribute of the primary classification at the latest 1 click and the time of the current content recommendation request; the time interval between the click time of the latest 1 click and the time of the current content recommendation request of the content file with the content attribute of the primary classification and the combined characteristics of the content attribute; the time interval between the click time of the farthest 1 click and the time of the current content recommendation request of the content file with the content attribute of the primary classification; the time interval between the click time of the farthest 1 click and the time of the current content recommendation request of the content file with the content attribute of the primary classification and the combined characteristics of the content attribute.

In some embodiments, the click time feature may further include: the content recommendation request of the content file with each content attribute is clicked in the preset latest click quantity window in the preset order and the request number interval of the current content recommendation request; the content file with each content attribute is clicked in the preset latest click quantity window in a preset sequence, the request number interval of the content recommendation request and the current content recommendation request is the same, and the combination characteristic of each content attribute is obtained. By utilizing such a click time feature, the bias that may be caused by a click time feature that is purely time interval dependent may be reduced.

As an example, the predetermined order may be specified as needed as described above, for example, may be the most recent 1 time, the most recent 2 nd time, the most recent 1 time, and the like. Table 3 below shows click time characteristics in the case where the preset number-of-most-clicks window is the most recent 128 clicks, the predetermined order is the most recent 1 and the most distant 1 (i.e., the 1 st and 128 th clicks), and there are three content attributes (a content attribute of one primary classification, a content attribute of one secondary classification, a content attribute of one tag classification).

As shown in table 3, taking the content attributes of the first-level classification as an example, in addition to the features shown in table 2 (not shown in table 3 for simplicity), the click time feature may further include: the request number interval between the content recommendation request of the content file with the content attribute of the primary classification and the current content recommendation request at the latest 1 click; the combined characteristics of the content attribute and the request number interval of the content recommendation request and the current content recommendation request of the content file with the content attribute of the primary classification at the latest 1 click; the request number interval between the content recommendation request of the content file with the content attribute of the primary classification at the farthest 1-time click and the current content recommendation request; the content file with the content attribute of the primary classification has the combined characteristics of the interval of the request number of the content recommendation request and the current content recommendation request at the farthest 1 click and the content attribute.

In some embodiments, the historical behavior data further comprises data relating to historical presentations of content files presented to the user. In this case, the plurality of features may include a presentation time feature in addition to the click time feature. The presentation time characteristic is a series of characteristics related to a presentation time at which the content file having each of the at least one content attribute is presented to the user in the historical presentation for a preset recent period. The preset latest period may be set as needed, for example, may be set to the latest 3 days, although this is not restrictive. In this case, the click time characteristic may be, for example, a characteristic related to a presentation time at which the user presented the content file having each of the at least one content attribute to the user in the last 3 days of historical presentation. The presentation time characteristic is used on the basis of the click time characteristic, so that the positive effect of the click on the content recommendation and the negative effect of the presentation but not the click on the content recommendation can be fully utilized, the accuracy of the content recommendation is further improved, key indexes such as click rate, click amount and the like of the recommended content are improved, and the experience of the user in watching the content is also enhanced.

In some embodiments, the presentation time characteristics include: a time interval between a presentation time at which the content file having the each content attribute is presented in a predetermined order in the historical presentation of the preset latest period and a time of a current content recommendation request; the time interval between the presentation time of the content file with each content attribute being presented in the preset latest period in the historical presentation in the predetermined order and the time of the current content recommendation request, and the combined characteristics of the two of each content attribute.

As an example, the predetermined order may be specified as needed as described above, for example, may be the most recent 1 time, the most recent 2 nd time, the most recent 1 time, and the like. Table 4 below shows the presentation time characteristics in the case where the preset latest period is the latest 3 days, the predetermined order is the latest 1 times and the farthest 1 times, respectively, and there are three content attributes (one primary-class content attribute, one secondary-class content attribute, one tag-class content attribute).

As shown in table 4, taking the content attributes of the first-level classification as an example, presenting the time characteristic may include: a time interval between a presentation time of the content file having the content attribute of the primary classification at the latest 1 presentation time in the latest 3 days of presentations and a time of a current content recommendation request; a time interval between a presentation time of the content file having the content attribute of the primary classification at the latest 1 presentation time and a time of a current content recommendation request in the last 3 days of presentation, and a combined characteristic of both the content attribute of the primary classification; a time interval between a presentation time of the content file having the content attribute of the primary classification at the farthest presentation time of 1 presentation in the last 3 days of presentation and a time of a current content recommendation request; the time interval between the presentation time of the content file having the content attribute of the primary classification at the farthest presentation time of 1 presentation in the last 3 days of presentation and the time of the current content recommendation request, and the combined characteristics of both the content attribute of the primary classification.

In some embodiments, presenting the temporal features may further include: the content recommendation request of the content file with each content attribute is presented in a preset order in the historical presentation of the preset recent period and the request number interval of the current content recommendation request; the content file with each content attribute is presented in a preset order in the historical presentation of the preset recent period, the request number interval of the content recommendation request and the current content recommendation request is between the content recommendation request and the current content recommendation request, and the combined characteristic of each content attribute is obtained.

As an example, the predetermined order may be specified as needed as described above, for example, may be the most recent 1 time, the most recent 2 nd time, the most recent 1 time, and the like. Table 5 below shows the presentation time characteristics in the case where the preset latest period is the latest 3 days, the predetermined order is the latest 1 times and the farthest 1 times, respectively, and there are three content attributes (one primary-class content attribute, one secondary-class content attribute, one tag-class content attribute).

As shown in table 5, taking the content attributes of the first-level classification as an example, presenting the time characteristic may include: the content recommendation request of the content file with the content attribute of the primary classification in the latest 1-time presentation in the latest 3 days of presentation is separated from the request number of the current content recommendation request; the content file with the content attribute of the primary classification is presented for the latest 1 time in the latest 3 days, the request number interval of the content recommendation request and the current content recommendation request is positioned, and the combined characteristic of the content attribute of the primary classification is obtained; the content recommendation request of the content file with the content attribute of the primary classification is separated from the request number of the current content recommendation request in the farthest 1-time presentation in the last 3 days of presentation; the content file with the content attribute of the primary classification is presented for the farthest 1 time in the last 3 days, the request number interval of the content recommendation request and the current content recommendation request and the combined characteristic of the content attribute of the primary classification.

By way of example, fig. 6 illustrates a schematic diagram of determining a click time feature and a presentation time feature in accordance with one embodiment of the present disclosure. As shown in fig. 6, the current time of making the current content recommendation request is 18:00, which is recorded as 0 th content recommendation request; the time of the last click on the content file with the content attribute of "Nezha" with tag classification is 16:13, which is the latest 4 th content recommendation request; the time of the farthest click is 13:07, which is the 16 th most recent content recommendation request. Therefore, the time interval between the time of the last click and the current time is 1:47, the request number interval is 4, and the time interval can be discretized by half an hour, namely 3; the time interval of the farthest 1 click from the current time is 4:53, the request number interval is 16, where the time interval can be discretized in half an hour, i.e., 9. The determination process of the presentation time characteristic in fig. 6 is similar and will not be described in detail here.

In some embodiments, in determining the feature values of the plurality of features for characterizing the interest of the user in each of the content files, the original value of each of the plurality of features may be first obtained, and then the feature value of each of the features may be obtained based on the original value of each of the features and the corresponding feature name. The original value of each feature may be the value of the feature itself (i.e., the input value) or obtained by indexing the input value of the feature as described below. As indicated above, embodiments of the present disclosure relate to combined features and single features (i.e., features that are not combined features). For a single feature, there is only one input value, such as the discretized value of 3 for the time interval of the last click time to the current time, the value of 4 for the request number interval, and so on, as described above. The input values are typically of the uint64 (64-bit unsigned integer) type, float (floating point) type. Input values such as content attributes, request time intervals, request number intervals and the like are generally of the type of agent 64, and the original value of the single feature is the input value; for example, the input value such as click rate is generally float characteristic, and the original value of the single characteristic is the input value 10000.

For a combined feature, there are multiple input values since it is a combination of multiple single features. For example, the interval between the request number of the content recommendation request and the current content recommendation request at the latest 1 click and the sum of the request number of the content recommendation request and the sum of the content recommendation request and the content recommendation request for the content file with the content attribute of the primary classification in the click behavior feature described aboveA combined feature of both the content attributes having input values of a request number interval and two single features of the content attributes. In this case, the original values of the single features can be recorded as

、

Then, obtaining an original value y of the combined feature by means of prime number multiplication, namely:

. Similarly, the way in which the original values are calculated can be extended to a combination feature having three or more input values. For example, for three input values

、

、

In the case of (a) in (b),

。

as an example, when the feature value of each feature is obtained based on the original value of each feature and the corresponding feature name, the original value of each feature may be hashed to obtain a first hash value of each feature; hashing the feature name character string of each feature to obtain a second hash value of each feature; and obtaining the characteristic value of each characteristic based on the first hash value and the second hash value of each characteristic.

For example, to increase feature distinctiveness and allow for online performance, embodiments of the present disclosure may map feature values to a 64-bit hash space. Using the high 16 bits of 64-bit space to reflect the feature type information, wherein the feature type information is obtained by hashing (hash) a feature name string (feature _ name) and shifting the low 16 bits; then, using the lower 48 bits to reflect the feature index information, the lower 48 bits are obtained by hashing the original value (feature _ value _ o) of the feature, that is, the feature value Y of the feature can be calculated according to the following formula:

Y=hash(feature_name)&0xFFFF<<48+hash(feature_value_o)&0xFFFFFFFFFFFF。

at step 204, a score for each of the content files is determined based at least on the feature values of the plurality of features. In some embodiments, at least feature values of the plurality of features may be input into the trained intelligent scoring model, resulting in a score for the each content file. The score for each content file may be used to represent a predicted click probability for the user for the each content file.

In practical applications, the trained intelligent scoring model can be selected according to practical needs, for example: logistic Regression (LR) model, depth factorization Machine (deep fm) model, and the like. Taking a selected logistic regression model for scoring as an example, respectively inputting the feature values of the plurality of features into the trained logistic regression model (classification model) to obtain the score of each content file. The formula used may be:

(ii) a Wherein,

for the nth characteristic value of the content file,

is composed of

Z is the score of the content file, z ∈ [0,1]。

In practical applications, the trained intelligent scoring model used may also be a deep learning model, for example, as is common in the field of artificial intelligence. The Artificial Intelligence (AI) is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making. The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. 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 the like.

As an example, the trained intelligent scoring model may be trained according to positive sample data and negative sample data; in the process of presenting the plurality of content files to the user, the characteristic values of a plurality of characteristics representing the interest of the user in the clicked content files in the plurality of presented content files are used as positive sample data, and the characteristic values of a plurality of characteristics representing the interest of the user in the content files which are not clicked in the plurality of presented content files are used as negative sample data.

It should be noted that the score of each of the content files may be determined in addition to the feature values of the plurality of features in combination with the feature values of the environmental features, the user features, the content features, and the like. The environmental characteristics represent attributes related to the content recommendation request, such as a request region, a request device, a used network and the like. The user characteristics mainly include the demographic characteristics of the user, such as the gender, age, address, occupation, and the like of the user; and long-term portrait characteristics of the user, such as the user's deep interests. The content characteristics mainly comprise topics, quality and the like of the content, and the data are generally marked when the content is stored in a library; and also includes collections, comments, etc. of content, which are typically counted offline after the content is rendered.

At step 205, a predetermined number of content files are selected from the plurality of content files for recommendation to the user based on the scores of the plurality of content files. Additionally, the predetermined number of content files may be sent to a client of the user for presentation.

In some embodiments, the selection of the predetermined number of content files may be performed by: based on the scores of the content files, sequencing the content files according to the scores to obtain an ordered sequence of the content files; a predetermined number of content files are selected for recommendation to the user starting with the first content file in the ordered sequence of content files.

The description will be given taking as an example a score of 10 content files, the predetermined number being 3. The scores of the 10 content files are all between 0 and 1, the 10 scores are sorted in the order of the scores from high to low, such as 0.9, 0.85, 0.83, 0.8, 0.75, 0.7, 0.66, 0.64, 0.6 and 0.58, and the content files corresponding to the scores of 0.9, 0.85 and 0.83 are selected for recommendation to the user.

In the embodiment of the disclosure, the click behavior characteristics related to the occurrence frequency of each content attribute of the user in the preset recent click number window are used during content recommendation, so that the short-term interest of the user can be fully embodied and the change of the interest of the user can be quickly embodied during content recommendation, the accuracy of content recommendation can be greatly improved, and key indexes such as click rate, click amount and the like of recommended content are improved. In particular, the preset number of recent clicks quantum window can be optionally utilized to achieve the effects of more finely depicting the user's short-term interests and more rapidly capturing the changes in the user's interests. In addition, the click time characteristic and the presentation time characteristic can be used for representing the difference in time during content recommendation, the accuracy of content recommendation is further improved, and key indexes such as click rate and click amount of recommended content are improved.

Fig. 7 illustrates a schematic flow chart diagram of a method 700 for content recommendation in accordance with one embodiment of the present disclosure. The method 700 may be implemented, for example, in an associated client on the terminal 120 or 130 shown in fig. 1. As shown in fig. 7, the method 700 may include the following steps.

In step 701, a predetermined number of content files for making a recommendation to a user are obtained. The predetermined number of content files is selected, for example, by the method 200 described with reference to fig. 2. In some embodiments, a content recommendation request may be first sent, for example, to a server, and then a predetermined number of content files sent, for example, by the server for recommendation to the user may be obtained.

In step 702, the predetermined number of content files are presented. This may be presented, for example, in the relevant client on

terminal

120 or 130 shown in fig. 1. The predetermined number of content files may be presented in various ways, such as by video, audio, and so forth.

By using the method for recommending the content of the embodiment of the disclosure, the recommended content can fully reflect the short-term interest of the user and quickly reflect the change of the interest of the user, so that the accuracy of content recommendation can be greatly improved, and key indexes such as click rate, click amount and the like of the recommended content are improved.

Next, a description will be given of a method for content recommendation according to an embodiment of the present disclosure, taking as an example a point that a content file is a video and a client on a terminal is Tencent. Fig. 8 is a schematic architecture diagram of a method for content recommendation provided by an embodiment of the present disclosure, and referring to fig. 8, the method for content recommendation provided by the embodiment of the present disclosure mainly includes an offline part and an online part; the off-line part mainly calculates a user portrait according to historical behavior data of a user and trains an intelligent scoring model, wherein the user portrait mainly comprises portraits of different dimensions such as primary classification and secondary classification; the online part mainly comprises recalls of candidate videos, ranking scores of the videos, video diversity presentation and the like.

For the offline portion, the user representation is a long-term accumulation of user interest, and may have a hierarchical structure, as shown in fig. 3, from the top level down, a first-level classification, a second-level classification, and a label classification. The intelligent scoring model may then be trained based on the user's historical behavior data and the user's image.

For the online part, taking the content attribute of "department" in the label classification as an example, videos related to "department" in the video library can be recalled as candidate videos. In actual implementation, when the user starts the recommendation service, the server may perform user portrait calculation on the user based on the user identifier carried in the request sent by the client, so as to recall the relevant video. The recalled videos may then be scored and ranked and a predetermined number of videos selected for recommendation to the user based on the scoring and ranking.

Fig. 9 is a schematic diagram of an architecture for sorting videos provided by an embodiment of the present disclosure, which mainly includes a resource adaptation, a feature extraction, and a score sorting section. As shown in fig. 9, in the resource adaptation part, the user portrait adaptation and the adaptation of the user behavior are performed first, that is, the historical behavior data and the user portrait of the user are obtained as described above.

Next, feature extraction will be explained. The feature extraction mainly relates to three parts of feature design, feature index and feature coding. The feature design is mainly to design various features representing the interest of users in videos based on video data of the videos so as to facilitate subsequent scoring. Specifically, the click behavior feature, click time feature, presentation time feature, and the like may be designed as described above.

The features typically have one or more input values, as described above. Feature indexing is primarily a consistent index of feature values for the convenience of computing feature values to obtain the original values of features as described above. A single feature typically has one input value, typically of the type uint64 (64-bit unsigned integer), float (floating point). For a combined feature, there are multiple input values since it is a combination of multiple single features. In this case, the original values of the single features may be obtained separately, and then the original values of the combined features may be obtained by means of prime number multiplication. Similarly, the way in which the original values are calculated can be extended to a combination feature having three or more input values.

Feature encoding is mainly to encode the original value of a feature to obtain a feature value. To increase feature distinctiveness and allow for online performance, embodiments of the present disclosure may map feature values to a 64-bit hash space. Reflecting feature type information by using the upper 16 bits of a 64-bit space, wherein the feature type information is obtained by hashing a feature name string (feature _ name) and shifting the lower 16 bits; then, using the lower 48 bits to reflect the feature index information, the lower 48 bits are obtained by hashing the original value (feature _ value _ o) of the feature, that is, the feature value Y of the feature can be calculated according to the following formula:

Y=hash(feature_name)&0xFFFF<<48+hash(feature_value_o)&0xFFFFFFFFFFFF。

next, feature values of a plurality of features obtained by feature extraction may be input into, for example, the trained LR model described above, and a score corresponding to the video may be calculated. In practical implementation, the unordered _ map container access parameter of stl (Standard template library) can be used, but the search time is too high, and the container access parameter of google dense _ map can also be used, so that the search time can be reduced by about 2/3 in terms of space time. After the scores of the videos are obtained, the videos are sorted based on the scores, and a preset number of videos are selected based on the sorting result to be recommended.

The flow of using the intelligent scoring model described above will be further described below. The intelligent scoring model essentially utilizes CTR (Click-through rate) estimation for estimating how large the probability that a content file is clicked by a user after being recommended is a very important link in an industrial-level recommendation system. FIG. 10 illustrates an overall flow of click-through rate estimation, which mainly includes four aspects of data, features, models, online, etc., according to an embodiment of the present disclosure. As shown in fig. 10. The data aspect mainly comprises the steps of obtaining original data, wherein the original data mainly comprise a click log and a presentation log of a user; the characteristic aspect is mainly characteristic engineering, and mainly comprises three categories of calculation for acquiring various user figures, content attributes and various characteristics as described above; the model aspect is mainly an intelligent scoring model which can be various linear models and nonlinear models; the online aspect is mainly online service, and relates to the extraction of the features, the calculation of scores, and the final sorting and recommendation. The invention is primarily directed to feature engineering module deployment.

Fig. 11 is a schematic diagram of an architecture of model training provided in an embodiment of the present invention, and referring to fig. 11, the model training mainly includes three parts, namely log merging, feature extraction, and model training, which are described below.

Log merging is mainly the aggregation of all information of a content recommendation request according to a click log (saving data related to the historical clicks of a user on a content file), a presentation log (saving data related to the historical presentations of a content file to a user). Because the click is typically delayed relatively to the presentation, there is a time window problem. For example, a 15-minute time window may be used in the embodiments of the present invention, and it is considered that all clicks presented at one time occur within 15 minutes, and if the time is out, it is considered that there is no click. For each content file requested each time, whether the content file is clicked and corresponding click and presentation data are searched, and the combined log data is written on Kafka which is a distributed, partitioned, multi-copy and multi-subscriber distributed log system.

And performing feature extraction according to the combined log data, and respectively extracting feature values of the clicked content files in the log, which correspond to the user, to construct a positive sample and a negative sample of model training, wherein as described above, feature values of a plurality of features corresponding to the clicked content files in the plurality of content files presented by the user are used as positive sample data, and feature values of a plurality of features corresponding to the content files not clicked in the plurality of content files presented are used as negative sample data. The feature extraction depends on the positive data (for example, for inquiring content attributes), the user image and the historical statistical information (for example, the information such as user inquiry click history and presentation history) of the content file, wherein the positive data and the historical statistical information are updated every hour, and the user image is updated every day. In the embodiment of the invention, 99% of sample data is randomly extracted as a training sample, the rest 1% is a test sample, and the training sample and the test sample can be written on two themes (topic) of kafka respectively for model training.

The method can use all training samples to train the model, and uses an online learning sparse algorithm to train a large-scale sparse logistic regression model. The logistic regression model trained offline in the invention can be derived every 30 minutes and pushed to the online environment.

Fig. 12 shows a schematic diagram of presentation of a content file recommended by a method for content recommendation according to an embodiment of the present disclosure, which is presented, for example, in a client on a terminal of a user. As shown in fig. 12, since the content attribute of the soccer ball exists in the secondary classification of the user profile of the user and the content hit frequency to the content attribute is the largest among the last 100 hits, a soccer-related video is recommended to the user in the main recommendation 1201. Optionally, after the user clicks on the football-related video, a three-drag scene 1202 may be entered, presenting a series of videos related to the videos in the main recommendation 1201.

Fig. 13 illustrates an exemplary block diagram of an apparatus 1300 for content recommendation according to one embodiment of the present disclosure. As shown in fig. 13, the apparatus 1300 for content recommendation includes a first obtaining module 1301, a second obtaining module 1302, a first determining module 1303, a second determining module 1304, and a selecting module 1305.

The first obtaining module 1301 is configured to obtain historical behavior data of a user, the historical behavior data comprising data relating to historical clicks of the user on a content file, and a user profile comprising a plurality of interest classifications of the user, each interest classification comprising a content attribute of a content file. In some embodiments, each interest category may further include an interest level corresponding to a content attribute of the content file. In some embodiments, the first obtaining module 701 may be configured to obtain historical behavior data and a user representation of the user in response to receiving a current content recommendation request for the user.

The second obtaining module 1302 is configured to obtain content data of a plurality of content files, the content data of each content file comprising at least one content attribute of said each content file. In some embodiments, the second obtaining module 1302 may be configured to obtain content data for a plurality of content files based on the user imagery.

The first determining module 1303 is configured to determine feature values of a plurality of features characterizing the interest of the user in the each content file based on the at least one content attribute of the each content file, the user representation and historical behavior data, wherein the plurality of features includes click behavior features related to the occurrence number of each content attribute in the at least one content attribute in the historical clicks of the user within a preset window of the number of most recent clicks. In some embodiments, the plurality of features may further include a click time feature, the click time feature being associated with a click time of the user's historical clicks on the content file having the each content attribute within the preset number of most recent clicks window. In some embodiments, the historical behavioural data further comprises data relating to historical presentation of content files to the user, and the plurality of characteristics further comprises a presentation time characteristic relating to a presentation time of a content file having each of the at least one content attribute presented to the user in the historical presentation for a preset recent period of time.

The second determining module 1304 is configured to determine a score for each of the content files based at least on the feature values of the plurality of features. In some embodiments, the second determination module 1304 may be configured to input at least feature values of the plurality of features into a trained intelligent scoring model, resulting in a score for each of the content files. The score for each content file may be used to represent a predicted click probability for the user for the each content file.

The selecting module 1305 is configured to select a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files.

Fig. 14 illustrates an exemplary block diagram of an apparatus 1400 for content recommendation according to one embodiment of the present disclosure. As shown in fig. 14, the apparatus 1400 for content recommendation includes a content file obtaining module 1401 and a presenting module 1402.

The file acquisition module 1401 is configured to acquire a predetermined number of content files for recommendation to a user from the apparatus for content recommendation 1300. In some embodiments, the file obtaining module 1401 may first send a content recommendation request and then obtain a predetermined number of content files for recommendation to a user as a response to the content recommendation request.

The rendering module 1402 is configured to render the predetermined number of content files. By way of example, the rendering module 1402 may be configured to render the predetermined number of content files in various manners, such as via video, audio, and so forth.

Fig. 15 illustrates an example system 1500 that includes an example computing device 1510 that represents one or more systems and/or devices that can implement the various techniques described herein. The computing device 1510 may be, for example, a server of a service provider, a device associated with a server, a system on a chip, and/or any other suitable computing device or computing system. Any of the apparatus for content recommendation 1300 and the apparatus for content recommendation 1400 described above with reference to fig. 13 and 14, respectively, may take the form of a computing device 1510. Alternatively, any one of the apparatus for content recommendation 1300 and the apparatus for content recommendation 1400 may be implemented as a computer program in the form of a content recommendation application 1516.

The example computing device 1510 as illustrated includes a processing system 1511, one or more computer-readable media 1512, and one or more I/O interfaces 1513 communicatively coupled to each other. Although not shown, the computing device 1510 may also include a system bus or other data and command transfer system that couples the various components to one another. A system bus can include any one or combination of different bus structures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and/or a processor or local bus that utilizes any of a variety of bus architectures. Various other examples are also contemplated, such as control and data lines.

The processing system 1511 represents functionality to perform one or more operations using hardware. Thus, the processing system 1511 is illustrated as including hardware elements 1514 that may be configured as processors, functional blocks, and so forth. This may include implementation in hardware as an application specific integrated circuit or other logic device formed using one or more semiconductors. Hardware element 1514 is not limited by the material from which it is formed or the processing mechanisms employed therein. For example, a processor may be comprised of semiconductor(s) and/or transistors (e.g., electronic Integrated Circuits (ICs)). In such a context, processor-executable instructions may be electronically-executable instructions.

Computer-readable medium 1512 is illustrated as including memory/storage 1515. Memory/storage 1515 represents memory/storage capacity associated with one or more computer-readable media. Memory/storage 1515 may include volatile media (such as Random Access Memory (RAM)) and/or nonvolatile media (such as Read Only Memory (ROM), flash memory, optical disks, magnetic disks, and so forth). Memory/storage 1515 may include fixed media (e.g., RAM, ROM, a fixed hard drive, etc.) as well as removable media (e.g., flash memory, a removable hard drive, an optical disk, and so forth). The computer-readable medium 1512 may be configured in a variety of other ways, as described further below.

The one or more I/O interfaces 1513 represent functionality that allows a user to enter commands and information to the computing device 1510 using various input devices and optionally also allows information to be presented to the user and/or other components or devices using various output devices. Examples of input devices include a keyboard, a cursor control device (e.g., a mouse), a microphone (e.g., for voice input), a scanner, touch functionality (e.g., capacitive or other sensors configured to detect physical touch), a camera (e.g., motion that may not involve touch may be detected as gestures using visible or invisible wavelengths such as infrared frequencies), and so forth. Examples of output devices include a display device (e.g., a monitor or projector), speakers, a printer, a network card, a haptic response device, and so forth. Thus, the computing device 1510 may be configured in various ways to support user interaction, as described further below.

The computing device 1510 also includes a content recommendation application 1516. The content recommendation application 1516 may be, for example, a software instance of any of the device for content recommendation 1300 and the device for content recommendation 1400, and in combination with other elements in the computing device 1510 implement the techniques described herein.

Various techniques may be described herein in the general context of software hardware elements or program modules. Generally, these modules include routines, programs, objects, elements, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The terms "module," "functionality," and "component" as used herein generally represent software, firmware, hardware, or a combination thereof. The features of the techniques described herein are platform-independent, meaning that the techniques may be implemented on a variety of computing platforms having a variety of processors.

An implementation of the described modules and techniques may be stored on or transmitted across some form of computer readable media. Computer readable media can include a variety of media that can be accessed by computing device 1510. By way of example, and not limitation, computer-readable media may comprise "computer-readable storage media" and "computer-readable signal media".

"computer-readable storage medium" refers to a medium and/or device, and/or a tangible storage apparatus, capable of persistently storing information, as opposed to mere signal transmission, carrier wave, or signal per se. Accordingly, computer-readable storage media refers to non-signal bearing media. Computer-readable storage media include hardware such as volatile and nonvolatile, removable and non-removable media and/or storage devices implemented in a method or technology suitable for storage of information such as computer-readable instructions, data structures, program modules, logic elements/circuits or other data. Examples of computer readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical storage, hard disks, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or other storage devices, tangible media, or an article of manufacture suitable for storing the desired information and accessible by a computer.

"computer-readable signal medium" refers to a signal-bearing medium configured to transmit instructions to the hardware of computing device 1510, such as via a network. Signal media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave, data signal or other transport mechanism. Signal media also includes any information delivery media. The term "modulated data signal" means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared and other wireless media.

As previously mentioned, the hardware element 1514 and the computer-readable medium 1512 represent instructions, modules, programmable device logic, and/or fixed device logic implemented in hardware that, in some embodiments, may be used to implement at least some aspects of the techniques described herein. The hardware elements may include integrated circuits or systems-on-chips, Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), Complex Programmable Logic Devices (CPLDs), and other implementations in silicon or components of other hardware devices. In this context, a hardware element may serve as a processing device that performs program tasks defined by instructions, modules, and/or logic embodied by the hardware element, as well as a hardware device for storing instructions for execution, such as the computer-readable storage medium described previously.

Combinations of the foregoing may also be used to implement the various techniques and modules described herein. Thus, software, hardware, or program modules and other program modules may be implemented as one or more instructions and/or logic embodied on some form of computer-readable storage medium and/or by one or more hardware elements 1514. The computing device 1510 may be configured to implement particular instructions and/or functions corresponding to software and/or hardware modules. Thus, implementing a module as a module executable by the computing device 1510 as software may be implemented at least in part in hardware, for example, using computer-readable storage media and/or hardware elements 1514 of the processing system. The instructions and/or functions may be executable/operable by one or more articles of manufacture (e.g., one or more computing devices 1510 and/or processing systems 1511) to implement the techniques, modules, and examples described herein.

In various implementations, the computing device 1510 may take on a variety of different configurations. For example, the computing device 1510 may be implemented as a computer-like device including a personal computer, a desktop computer, a multi-screen computer, a laptop computer, a netbook, and so forth. The computing device 1510 may also be implemented as a mobile device class device including mobile devices such as mobile phones, portable music players, portable gaming devices, tablet computers, multi-screen computers, and the like. Computing device 1510 may also be implemented as a television-like device that includes or is connected to a device having a generally larger screen in a casual viewing environment. These devices include televisions, set-top boxes, game consoles, and the like.

The techniques described herein may be supported by these various configurations of the computing device 1510 and are not limited to specific examples of the techniques described herein. The functionality may also be implemented in whole or in part on the "cloud" 1520 through the use of a distributed system, such as through the platform 1522 as described below.

The cloud 1520 includes and/or is representative of a platform 1522 for resources 1524. The platform 1522 abstracts underlying functionality of hardware (e.g., servers) and software resources of the cloud 1520. The resources 1524 may include applications and/or data that may be used when executing computer processes on servers remote from the computing device 1510. The resources 1524 may also include services provided over the internet and/or over a subscriber network such as a cellular or Wi-Fi network.

The platform 1522 may abstract resources and functionality to connect the computing device 1510 with other computing devices. The platform 1522 may also be used to abstract a hierarchy of resources to provide a corresponding level of hierarchy encountered for the demand of the resources 1524 implemented via the platform 1522. Thus, in interconnected device embodiments, implementation of functions described herein may be distributed throughout the system 1500. For example, the functionality may be implemented in part on the computing device 1510 as well as through the platform 1522 that abstracts the functionality of the cloud 1520.

It will be appreciated that embodiments of the disclosure have been described with reference to different functional units for clarity. However, it will be apparent that the functionality of each functional unit may be implemented in a single unit, in a plurality of units or as part of other functional units without departing from the disclosure. For example, functionality illustrated to be performed by a single unit may be performed by a plurality of different units. Thus, references to specific functional units are only to be seen as references to suitable units for providing the described functionality rather than indicative of a strict logical or physical structure or organization. Thus, the present disclosure may be implemented in a single unit or may be physically and functionally distributed between different units and circuits.

It will be understood that, although the terms first, second, third, etc. may be used herein to describe various devices, elements, components or sections, these devices, elements, components or sections should not be limited by these terms. These terms are only used to distinguish one device, element, component or section from another device, element, component or section.

Although the present disclosure has been described in connection with some embodiments, it is not intended to be limited to the specific form set forth herein. Rather, the scope of the present disclosure is limited only by the accompanying claims. Additionally, although individual features may be included in different claims, these may possibly advantageously be combined, and the inclusion in different claims does not imply that a combination of features is not feasible and/or advantageous. The order of features in the claims does not imply any specific order in which the features must be worked. Furthermore, in the claims, the word "comprising" does not exclude other elements, and the words "a" or "an" do not exclude a plurality. Reference signs in the claims are provided merely as a clarifying example and shall not be construed as limiting the scope of the claims in any way.

Claims

1. A method for content recommendation, comprising:

obtaining historical behavior data of a user and a user representation, wherein the historical behavior data comprises data related to historical clicks of the user on a content file, the user representation comprises a plurality of interest classifications of the user, and each interest classification comprises a content attribute of the content file;

acquiring content data of a plurality of content files, wherein the content data of each content file comprises at least one content attribute of each content file;

determining feature values of a plurality of features for characterizing the user's interest in each content file based on the at least one content attribute of each content file, the user representation, and historical behavior data, the plurality of features including click behavior features related to the number of occurrences of each of the at least one content attribute in historical clicks of the user within a preset window of the number of recent clicks;

determining a score for the each content file based at least on feature values of the plurality of features;

selecting a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files.

2. The method of claim 1, wherein obtaining historical behavioral data and a user representation of a user comprises:

in response to receiving a current content recommendation request for a user, historical behavior data and a user representation of the user are obtained.

3. The method of claim 1, wherein each interest category further includes an interestingness corresponding to a content attribute of the content file, and the click behavior feature includes:

the content attribute and the corresponding combination feature of the occurrence frequency of the content attribute in the historical clicks of the user in each corresponding preset closest click number quantum window in at least one preset closest click number quantum window, wherein the at least one preset closest click number quantum window is a sub-window of the preset closest click number window; and

the ranking of each content attribute, the interestingness of each content attribute in its corresponding interest classification, and the corresponding combination feature of the number of occurrences of each content attribute.

4. The method of claim 2, wherein the plurality of features further includes a click time feature related to a click time of a historical click of the user on a content file having the each content attribute within the preset number of most recent clicks window.

5. The method of claim 4, wherein the click time characteristics comprise:

the time interval between the click time of the content file with each content attribute clicked in the preset latest click number window in the preset order and the time of the current content recommendation request; and

the time interval between the click time when the content file with each content attribute is clicked in the preset number-of-most-clicks window in the preset sequence and the time of the current content recommendation request, and the combined characteristics of the content file with each content attribute.

6. The method of claim 5, wherein the click time feature further comprises:

the content recommendation request of the content file with each content attribute is clicked in the preset latest click quantity window in the preset order and the request number interval of the current content recommendation request; and

the content file with each content attribute is clicked in the preset latest click quantity window in a preset sequence, the request number interval of the content recommendation request and the current content recommendation request is the same, and the combination characteristic of each content attribute is obtained.

7. The method of claim 4, wherein the historical behavior data further comprises data relating to a historical presentation of content files to the user, and the plurality of features further comprises a presentation time feature relating to a presentation time of a content file having each of the at least one content attribute to the user in the historical presentation for a preset recent period of time.

8. The method of claim 7, wherein the presenting temporal features comprises:

a time interval between a presentation time at which the content file having the each content attribute is presented in a predetermined order in the historical presentation of the preset latest period and a time of a current content recommendation request; and

the time interval between the presentation time of the content file with each content attribute being presented in the preset latest period in the historical presentation in the predetermined order and the time of the current content recommendation request, and the combined characteristics of the two of each content attribute.

9. The method of claim 8, wherein presenting temporal features further comprises:

the content recommendation request of the content file with each content attribute is presented in a preset order in the historical presentation of the preset recent period and the request number interval of the current content recommendation request; and

the content file with each content attribute is presented in a preset order in the historical presentation of the preset recent period, the request number interval of the content recommendation request and the current content recommendation request is between the content recommendation request and the current content recommendation request, and the combined characteristic of each content attribute is obtained.

10. The method of claim 1, wherein obtaining content data for a plurality of content files comprises:

acquiring content data of the plurality of content files based on the user image of the user.

11. The method of claim 1, wherein determining feature values for a plurality of features characterizing the user's interest in the each content file based on the at least one content attribute of the each content file, the user representation, and historical behavior data comprises:

obtaining an original value for each of the plurality of features;

and obtaining the characteristic value of each characteristic based on the original value of each characteristic and the corresponding characteristic name.

12. The method of claim 1, wherein determining a score for the each content file based at least on feature values of the plurality of features comprises:

inputting at least the feature values of the plurality of features into the trained intelligent scoring model to obtain a score of each content file; the trained intelligent scoring model is obtained by training according to positive sample data and negative sample data; in the process of presenting the plurality of content files to the user, the characteristic values of a plurality of characteristics representing the interest of the user in the clicked content files in the plurality of presented content files are used as positive sample data, and the characteristic values of a plurality of characteristics representing the interest of the user in the content files which are not clicked in the plurality of presented content files are used as negative sample data.

13. An apparatus for content recommendation, comprising:

a first obtaining module configured to obtain historical behavior data of a user, the historical behavior data including data related to historical clicks of a content file by the user, and a user representation including a plurality of interest classifications of the user, each interest classification including a content attribute of a content file;

a second obtaining module configured to obtain content data of a plurality of content files, the content data of each content file including at least one content attribute of the each content file;

a first determination module configured to determine feature values of a plurality of features characterizing the user's interest in each content file based on at least one content attribute of the each content file, the user representation, and historical behavior data, the plurality of features including click behavior features related to the number of occurrences of each content attribute in the at least one content attribute in historical clicks of the user within a preset window of the number of most recent clicks;

a second determination module configured to determine a score for the each content file based at least on feature values of the plurality of features;

a selection module configured to select a predetermined number of content files from the plurality of content files for recommendation to the user based on the scores of the plurality of content files.

14. A computing device comprising

A memory configured to store computer-executable instructions;

a processor configured to perform the method of any one of claims 1-12 when the computer-executable instructions are executed by the processor.

15. A computer-readable storage medium storing computer-executable instructions that, when executed, perform the method of any one of claims 1-12.