WO2023001093A1 - 文章推荐方法、装置、设备及存储介质 - Google Patents

文章推荐方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2023001093A1
WO2023001093A1 PCT/CN2022/106208 CN2022106208W WO2023001093A1 WO 2023001093 A1 WO2023001093 A1 WO 2023001093A1 CN 2022106208 W CN2022106208 W CN 2022106208W WO 2023001093 A1 WO2023001093 A1 WO 2023001093A1
Authority
WO
WIPO (PCT)
Prior art keywords
candidate
article
articles
style
feature
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2022/106208
Other languages
English (en)
French (fr)
Inventor
刘峰
李东亮
杨大威
翁源
叶双
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing ByteDance Network Technology Co Ltd
Original Assignee
Beijing ByteDance Network Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing ByteDance Network Technology Co Ltd filed Critical Beijing ByteDance Network Technology Co Ltd
Publication of WO2023001093A1 publication Critical patent/WO2023001093A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9538—Presentation of query results

Definitions

  • the embodiments of the present application relate to the technical field of artificial intelligence (AI), and in particular to an article recommendation method, device, device, and storage medium.
  • AI artificial intelligence
  • article ranking is to determine the display position of each candidate article on the web page.
  • the server sorts candidate articles based on user characteristics and content characteristics of candidate articles. This method leads to low article recommendation accuracy. question.
  • Embodiments of the present application provide an article recommendation method, device, device, and storage medium, thereby improving article recommendation accuracy.
  • an article recommendation device including: a first acquisition module, a first determination module, a second acquisition module, a second determination module, and a push module, wherein the first acquisition module is used to acquire user requests; the first The determination module is used to determine M placements and N candidate articles in the current webpage according to user requests, and M and N are both integers greater than 1; the second acquisition module is used to obtain at least one candidate article on the M placements respectively Corresponding feature information; the second determination module is used to determine the respective articles to be recommended on the M placements according to the feature information corresponding to at least one candidate article on the M placements; the push module is used to push M display to the terminal device
  • an electronic device including: a processor and a memory, the memory is used to store a computer program, the processor is used to call and run the computer program stored in the memory, and perform the same as the first aspect or various implementations thereof method in .
  • a computer-readable storage medium for storing a computer program, and the computer program causes a computer to execute the method in the first aspect or various implementations thereof.
  • a computer program product including computer program instructions, where the computer program instructions cause a computer to execute the method in the first aspect or various implementations thereof.
  • a computer program which enables a computer to execute the method in the first aspect or various implementations thereof.
  • the server can base on the user characteristics, the content characteristics of the candidate articles, the K style combination characteristics of the candidate articles, and the 1st display position to the i-1th display position
  • the style combination features of the determined articles to be recommended are used to sort the candidate articles, that is, to determine the articles to be recommended on the i-th display position, and then recommend the articles to be recommended on the i-th display position to users.
  • This method not only Considering the user characteristics and the content characteristics of candidate articles, the style combination characteristics of candidate articles and the style combination characteristics of sorted articles are also considered, so that the article recommendation accuracy can be improved.
  • FIGS 1 to 8 are schematic diagrams showing the articles provided by the embodiments of the present application.
  • FIG. 9 is a schematic diagram of an application scenario provided by an embodiment of the present application.
  • FIG. 10 is a flowchart of an article recommendation method provided in Embodiment 1 of the present application.
  • Fig. 11 is a schematic diagram of the first model provided by the embodiment of the present application.
  • FIG. 12 is a flowchart of an article recommendation method provided by another embodiment of the present application.
  • Fig. 13 is a schematic diagram of the second model provided by the embodiment of the present application.
  • FIG. 14 is a schematic diagram of an article recommendation device provided in an embodiment of the present application.
  • FIG. 15 is a schematic block diagram of an electronic device 1500 provided by an embodiment of the present application.
  • the style combination feature of the candidate article is a feature obtained by combining at least one style of the candidate article, for example: a style feature of the candidate article is: display a certain number of images, or display a certain number of images in the candidate article Big picture or small picture. Another style characteristic of the candidate article is: the user avatar is displayed or not displayed in the candidate article. Another style feature of the candidate article is: the candidate article displays or does not display an abstract.
  • Figures 1 to 8 are schematic diagrams of article display provided by the embodiment of the present application. As shown in Figure 1, the style combination features displayed by the candidate article include: no image is displayed in the candidate article, and user avatars and articles are not displayed Summary.
  • the style combination features displayed by the candidate article include: a pair of images are displayed in the candidate article, and user avatars and article abstracts are not displayed.
  • the style combination features displayed by the candidate article include: a large picture is displayed in the candidate article, and user avatars and article summaries are not displayed.
  • the style combination features displayed by the candidate article include: three small pictures are displayed in the candidate article, and user avatars and article summaries are not displayed.
  • the style combination features displayed by the candidate article include: two small images are displayed in the candidate article, and user avatars are displayed instead of article summaries. As shown in FIG.
  • the style combination features displayed by the candidate article include: the candidate article displays a large picture, and displays the user avatar, but does not display the abstract of the article.
  • the style combination features displayed by the candidate article include: the candidate article does not display images, and does not display user avatars, but displays article summaries.
  • the style combination features displayed by the candidate article include: three small pictures are displayed in the candidate article, and the user avatar is not displayed, but the article summary is displayed. In short, the embodiment of the present application does not limit the style features and style combination features of the candidate articles.
  • the click-through rate of a candidate article refers to the ratio of the number of times the candidate article is clicked to the number of times it is displayed on a web page, that is, clicks/views, which is a percentage. It reflects the attention degree of the candidate article on the webpage, which can be used to measure the attractiveness of the candidate article.
  • the server sorts candidate articles based on user features and content features of the candidate articles. After the server sorts the candidate articles based on the user characteristics and the content characteristics of the candidate articles, the server then determines the display style of the candidate articles on each display position. On the one hand, this sorting method does not consider the style combination characteristics of candidate articles, which leads to the problem of low article recommendation accuracy. On the other hand, the current article ranking method does not consider the style combination characteristics of the sorted articles, which leads to the problem of low article recommendation accuracy. For example, if 3 articles without pictures are displayed in succession, then one article with a large picture is displayed and 4 articles with a large picture are displayed in succession, the last article with a large picture in the former is more likely to be clicked. This is because the former three articles without pictures are relatively monotonous, while the articles with pictures are richer in information than those without pictures, while the information of the three articles in the latter has little change.
  • the embodiment of the present application will consider the style combination features of the candidate articles and the style combination features of the previous sorted articles, so as to improve the accuracy of article recommendation.
  • FIG. 9 is a schematic diagram of an application scenario provided by the embodiment of the present application.
  • the terminal device 910 can communicate with the server 920, wherein, The terminal device 910 can obtain and send a user request to the server 920.
  • the user request is used to request to browse a certain webpage.
  • the server 920 that is, the webpage browsing server corresponding to the webpage, obtains the user request
  • the technical solution provided according to the embodiment of the present application is as follows: Each candidate article determines a display position, and finally pushes the articles on each display position to the central terminal device 910 .
  • the terminal device may be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a wearable device, etc., but is not limited thereto, and this embodiment of the present application does not limit it.
  • the server may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.
  • the communication between the terminal device and the server, or between the servers may be performed using a cellular mobile network or wireless fidelity (Wireless Fidelity, WiFi), which is not limited in this embodiment of the present application.
  • a cellular mobile network or wireless fidelity Wireless Fidelity, WiFi
  • Fig. 10 is a flow chart of an article recommendation method provided in Embodiment 1 of the present application.
  • the method can be executed by the server in Fig. 9, but is not limited thereto.
  • the following uses the server to execute the article recommendation method as an example.
  • the technical solution of the application embodiment is described, as shown in Figure 10, the method includes the following steps:
  • S1010 Obtain a user request.
  • S1020 Determine M display positions and N candidate articles in the current webpage according to the user request, where both M and N are integers greater than 1.
  • S1040 Determine respective articles to be recommended on the M placements according to feature information corresponding to at least one candidate article on the M placements.
  • the user request is used to request to browse a certain webpage, and the user request may be triggered by the user clicking on the webpage link on the terminal device, or the user enters the webpage URL on the terminal device and presses the Enter key triggered, the embodiment of this application does not limit how to trigger or obtain the user request.
  • the above M display positions are positions where candidate articles can be displayed in the current web page, and the current web page is a web page requested by the user to browse.
  • the N candidate articles may be N candidate articles on the same topic, for example: the N candidate articles are all introductions about a certain person, or the N candidate articles may not be the N candidate articles on the same topic, for example :
  • the webpage displayed by the terminal device includes: articles on various popular topics, etc.
  • M is less than or equal to N. That is, the number of candidate articles is greater than or equal to the display positions on the current web page where the candidate articles can be displayed.
  • M is greater than N. That is, the number of candidate articles is less than the display positions that can display candidate articles on the current web page.
  • the embodiment of the present application does not limit the size relationship between M and N.
  • At least one candidate article on the 1st placement is N candidate articles
  • at least one candidate article on the i+1th placement is at least one candidate article on the ith placement is deleted
  • the server may sort the above M placements so that all M placements have unique indexes.
  • the candidate articles on this position are the above N candidate articles, based on this, the server can first determine the articles to be recommended on the first placement, and further, for the second placement Specifically, the candidate article at this position is the candidate article after the above N candidate articles have been deleted from the article to be recommended on the first placement. Based on this, the server can determine the article to be recommended on the second placement , and so on, the server can determine the candidate articles on each placement, and determine the articles to be recommended on each placement among the candidate articles on each placement.
  • server may use any of the following practicable ways to sort the M placements, but is not limited thereto:
  • the server can sort the M placements according to their priorities.
  • the placements with higher priorities are ranked higher.
  • the placements with lower priorities are ranked higher.
  • the priority of the M placements is related to the attractiveness of the M placements to users. For example, the most central placement on the current webpage is the most attractive, so this placement has the highest priority, and the top leftmost placement on the current webpage The corner placement is the next most attractive placement, so that placement has the second priority relative to the centermost placement, and so on.
  • Realization method 2 The server can sort the M placements according to their priorities. The placements with higher priorities will be ranked higher. On the contrary, the placements with lower priorities will be ranked higher. The lower the sort. In other words, placements with higher priority have smaller indexes, and placements with lower priorities have larger indexes.
  • the server can determine the priorities of the M placements in order from left to right and top to bottom. For example, the placement at the upper left corner of the current web page has the highest priority, and the next display at the placement at the top left corner Location has second priority, and so on.
  • the server may sort the M display positions according to the attractiveness of the M display positions to the user. For example: the most center-most placement on the current page is the most attractive placement, so it ranks first, and the top-left most placement on the current page is the next most attractive placement, so it ranks first Two, and so on.
  • the server can sort the M placements in order from left to right and from top to bottom, for example: the placement in the upper left corner of the current web page is ranked first, and the placement in the upper left corner of the current web page is ranked first. The next placement for the position comes in second, and so on.
  • two adjacent display positions are arranged in random order, and they may be adjacent or not adjacent on the current web page, which is not limited in this embodiment of the present application.
  • the user characteristics that trigger the user request include at least one of the following, but are not limited thereto: the user's age, gender, preference, educational background, and the like.
  • the content characteristics of the candidate article include at least one of the following, but not limited thereto: the category to which the candidate article belongs, the keyword of the candidate article, the author, the click rate of the candidate article in the category to which it belongs, and the like.
  • the category of candidate articles can be divided according to the expression of the articles, including but not limited to the following items: narrative, explanatory, and argumentative.
  • the categories of candidate articles can be classified according to literary works, including, but not limited to: poetry, novels, prose (lyrical prose, narrative prose), script, etc.
  • the categories to which candidate articles belong can also be classified according to expression, rhetorical techniques, and expression techniques.
  • the above K style combination features of the candidate article may be all the style combination features of the candidate article, or may be some style combination features in all the style combination features of the candidate article. This is not limited.
  • the server determines that the two style combination features of the candidate article item are: the candidate article displays a large image, and displays a user avatar, but does not display an article summary. And, three small pictures are displayed in the candidate article, and the user avatar is not displayed, but the abstract of the article is displayed.
  • the style combination features of the determined articles to be recommended on the first placement to the i-1th placement include: the style combination features of the determined articles to be recommended on the first placement, the first The determined style combination features of the articles to be recommended on the 2 placement positions...the determined style combination features of the articles to be recommended on the i-1th placement position.
  • the style combination features of the determined articles to be recommended on the first to second placements include: the first The style combination characteristics of the determined article to be recommended on the first placement and the style combination characteristics of the determined article to be recommended on the second placement, for example: the style combination characteristics of the determined article to be recommended on the first placement
  • the style combination characteristics of the determined article to be recommended on the first placement For: a large picture is displayed in the article to be recommended, and the user's avatar is displayed, but the abstract of the article is not displayed.
  • the characteristics of the style combination of the article to be recommended determined in the second display position are: the article to be recommended displays two small pictures, displays the user's avatar, and displays the abstract of the article.
  • the server may determine articles to be recommended on the M placements in any of the following practicable ways, but is not limited thereto:
  • Achievable way 1 For any candidate article in at least one candidate article on the i-th display position, and any style combination feature in the K style combination features of the candidate article, the server combines the user characteristics, the content of the candidate article The features, style combination features and the style combination features of the articles to be recommended from the 1st to the i-1th placements are input into the first model to obtain the click-through rate of the candidate articles under the style combination features .
  • the server determines the maximum click-through rate of the candidate article under the K style combination features as the score of the candidate article.
  • the server determines the candidate article with the highest score on the i-th placement as the article to be recommended on the i-th placement.
  • Realization method 2 For any candidate article in at least one candidate article on the i-th display position, and any style combination feature in the K style combination features of the candidate article, the server combines the user characteristics, the content of the candidate article The features, style combination features and the style combination features of the articles to be recommended from the 1st to the i-1th placements are input into the first model to obtain the click-through rate of the candidate articles under the style combination features .
  • the server determines the average click rate of the candidate article under the K style combination features as the score of the candidate article.
  • the server determines the candidate article with the highest score on the i-th placement as the article to be recommended on the i-th placement.
  • FIG. 11 is a schematic diagram of the first model provided by the embodiment of the present application.
  • the input of the first model is user characteristics, content characteristics of candidate articles, certain style combination characteristics and The style combination features of the determined articles to be recommended from the placement position to the i-1th placement position.
  • the output of the first model is the click-through rate of the candidate article under the certain style combination feature.
  • the server can sequentially select an article to be recommended from the candidate articles for the first placement to the Mth placement and fill in the corresponding position.
  • the article to be recommended on the i-th display position needs to be determined currently, there are C candidate articles corresponding to the display position, assuming that the candidate set composed of these C candidate articles is ⁇ item 1 , item 2 ... item C ⁇ , let score(item j ) is the score of the candidate article item j .
  • the server can select the candidate article with the highest score and rank it in the i-th display position.
  • the score of the candidate article item j can be calculated according to the following formula (1):
  • Model multi () represents the first model.
  • max ⁇ means to take the maximum value.
  • F user represents a user feature, wherein, for all candidate articles, the corresponding user features are the same, and for the same candidate article in different placements, the corresponding user features are also the same.
  • (1,j 1 ) means to select the jth 1st style in the first style dimension
  • (2,j 2 ) means to select the jth style in the second style dimension j 2 styles
  • ...(N,j N ) means to select the j Nth style in the Nth style dimension.
  • the candidate article item j includes K style feature combinations.
  • the server can sequentially select an article to be recommended from the candidate articles for the first placement to the Mth placement and fill in the corresponding position.
  • the article to be recommended on the i-th display position needs to be determined currently, there are C candidate articles corresponding to the display position, assuming that the candidate set composed of these C candidate articles is ⁇ item 1 , item 2 ... item C ⁇ , let score(item j ) is the score of the candidate article item j .
  • the server can select the candidate article with the highest score and rank it in the i-th display position.
  • the score of the candidate article item j can be calculated according to the following formula (2):
  • Model multi () represents the first model.
  • mean ⁇ means to take the average value.
  • F user represents a user feature, wherein, for all candidate articles, the corresponding user features are the same, and for the same candidate article in different placements, the corresponding user features are also the same.
  • (1,j 1 ) means to select the jth 1st style in the first style dimension
  • (2,j 2 ) means to select the jth style in the second style dimension j 2 styles
  • ...(N,j N ) means to select the j Nth style in the Nth style dimension.
  • the candidate article itemj includes K style feature combinations.
  • the server determines the respective articles to be recommended on the M placements, it can push the corresponding articles to be recommended on the M placements to the terminal device, so that the terminal device can display the articles on the above M placements on the current webpage.
  • the corresponding articles to be recommended are displayed on the above.
  • the article to be recommended on the placement can be displayed with a target style combination feature, and the target style combination feature is the score of the article to be recommended when the server determines the , the style combination feature with the highest click-through rate, of course, the target style combination feature is the style combination feature with the second highest click-through rate when the server determines the score of the article to be recommended.
  • the manner of determining the target style combination feature is not limited to this.
  • FIG. 12 is a flow chart of an article recommendation method provided in another embodiment of the present application, as shown in FIG. 12 The method includes the following steps:
  • S4 Determine the article to be recommended on the i-th placement according to the feature information corresponding to at least one candidate article on the i-th placement;
  • At least one candidate article on the 1st placement is N candidate articles
  • at least one candidate article on the i+1th placement is at least one candidate article on the ith placement is deleted
  • Candidate articles obtained after the article to be recommended on the placement position; for any candidate article in at least one candidate article on the i-th placement position, the feature information corresponding to the candidate article includes: the user feature that triggers the user request, the candidate article’s Content features, K style combination features of the candidate articles, and style combination features of the determined articles to be recommended in the 1st to i-1th placements, K is an integer greater than 1.
  • the server can base on user characteristics, content characteristics of candidate articles, K style combination features of candidate articles, and display positions from the 1st display position to the i-1th display
  • the style combination features of the determined articles to be recommended at the positions sort the candidate articles, that is, determine the articles to be recommended on the i-th display position, and then recommend the articles to be recommended on the i-th display position to the user.
  • the method not only considers the user characteristics and the content characteristics of candidate articles, but also considers the style combination characteristics of candidate articles and the style combination characteristics of sorted articles, so that the accuracy of article recommendation can be improved.
  • the K style combination features of the candidate article can be all the style combination features of the candidate article, or part of the style combination features in all the style combination features of the candidate article. Or, that is to say, for any candidate article in at least one candidate article on the i-th placement, the server may select K style combination features from all style combination features of the candidate article.
  • server may select K style combination features from all style combination features of the candidate article in the following practicable manner, but is not limited thereto:
  • Realization method 1 For any style combination feature in all style combination features of candidate articles, the server determines the score of style combination features, and the server selects the style combination features with the top K scores among all style combination features of candidate articles, to combine features as K styles.
  • Realization method 2 For any style combination feature in all the style combination features of the candidate article, the server determines the score of the style combination feature, and the server selects the style combination features with the top P scores among all the style combination features of the candidate article, P is an integer greater than K. Further, the server selects K style combination features from the P style combination features as K style combination features. Wherein, the server may randomly select K style combination features from the P style combination features or select K style combination features according to a certain preset rule, which is not limited in this embodiment of the present application.
  • server may determine the score of the style combination feature in any of the following implementable ways, but is not limited thereto:
  • the server can input user features, content features and style features of candidate articles into the second model for any style feature in the style combination feature, so as to obtain the click-through rate of the candidate article under the style feature, and further Specifically, the server sums the click-through rates of the candidate articles under each style feature in the style combination feature to obtain the score of the style combination feature.
  • Realization method 2 The server can input user features, content features and style features of candidate articles into the second model for any style features in the style combination features, so as to obtain the click-through rate of candidate articles under the style features, and further , the server averages the click-through rates of the candidate articles under each style feature in the style combination feature to obtain the score of the style combination feature.
  • any style feature in the above-mentioned style combination features is a specific style feature of a certain dimension in the style combination feature, for example: the style feature can be: three small pictures are displayed in the candidate article, or, in the candidate article User avatars are not displayed, or article summaries are not displayed in candidate articles, etc.
  • FIG. 13 is a schematic diagram of the second model provided by the embodiment of the present application.
  • the inputs of the second model are user characteristics, content characteristics of candidate articles, and certain style characteristics.
  • the output of the second model is the click-through rate of the candidate article item j under the i -th style feature in the i-th dimension style feature.
  • the server can use the following formula (2) to determine the click-through rate of the candidate article under the style feature:
  • F user represents a user feature, wherein, for all candidate articles, the corresponding user features are the same, and for the same candidate article in different placements, the corresponding user features are also the same.
  • the server can determine the score of the candidate article in a certain style combination feature through the following formula (3), but it is not limited to this:
  • the server can determine the score of a candidate article in a certain style combination feature through the following formula (4), but it is not limited thereto:
  • mean ⁇ means to take the average value.
  • the server calculates the score of the candidate article item j according to the above formula (1), the calculation amount of the server is However, if the server selects K style combination features according to formulas (3) and (4), although the calculation amount of the server is also However, the computational complexity of the second model is much lower than that of the first model. Therefore, the pruning operation adopted by the server, that is, only selecting K style combination features, can reduce the computational complexity and improve the efficiency of article recommendation.
  • the style combination feature selection problem is regarded as a multi-armed lottery problem (Multi-Armed Bandit), and the style combination features with the highest click-through rate are found by exploring as fast as possible.
  • the style combination features obtained in this way are not personalized, that is, for all users, the style combination features of the same candidate article are the same.
  • the server combines user features, content features of candidate articles, style combination features, and The style combination feature is input into the first model to obtain the click-through rate of the candidate article under the style combination feature, and the maximum click-through rate of the candidate article under the K style combination features can be combined with the style combination feature corresponding to the maximum click-through rate It is determined as the style combination feature to be displayed of the candidate article. That is to say, from the perspective of style combination feature selection alone, the final selected style combination features take into account user characteristics, and the style combination features obtained in this way are personalized, that is, for different users, the style combination of the same candidate article The characteristics can be different, so that the sense of user experience can be improved.
  • the server can train the first model, specifically, the first model can be trained in the following practicable ways, but not limited thereto:
  • the server obtains multiple first articles, sorts the multiple first articles according to their respective content features, and determines for any first article in the sorted first articles
  • the style combination feature of the first article is used to train the first model according to a plurality of first articles after the style combination feature has been determined.
  • the server can use n(i) style features for the i-th dimension style features Randomly select style features among the features.
  • the server in order to better train the first model, the server expects all style combination features to appear as much as possible. Therefore, when the server trains the first model, it can randomly select the candidate styles corresponding to the style features to form A wide variety of style combination features, including: possibly rare style combination features.
  • FIG. 14 is a schematic diagram of an article recommendation device provided in an embodiment of the present application.
  • the device may be the server in FIG. 9, but is not limited thereto. As shown in FIG. 14, the device includes:
  • the first obtaining module 1401 is configured to obtain user requests.
  • the first determining module 1402 is used to determine M display positions and N candidate articles in the current webpage according to the user request, where M and N are both integers greater than 1.
  • the second obtaining module 1403 is configured to obtain feature information corresponding to at least one candidate article on the M placements.
  • the second determination module 1404 is configured to determine articles to be recommended on each of the M placements according to feature information corresponding to at least one candidate article on the M placements.
  • the push module 1405 is configured to push the articles to be recommended on the M placements to the terminal device.
  • At least one candidate article on the 1st placement is N candidate articles
  • at least one candidate article on the i+1th placement is at least one candidate article on the ith placement is deleted
  • Candidate articles obtained after the articles to be recommended on the display position, i 1, 2...M.
  • the feature information corresponding to the candidate article includes: the user feature that triggers the user request, the content feature of the candidate article, the K style combination features of the candidate article, and the first The style combination features of the determined articles to be recommended in the 1st to i-1th placements, K is an integer greater than 1.
  • the second determining module 1404 is specifically configured to: for any candidate article in at least one candidate article on the i-th display position, and any style combination in the K style combination features of the candidate article feature, the user features, the content features of candidate articles, the style combination features and the style combination features of the articles to be recommended from the first placement to the i-1th placement are input into the first model to obtain CTR of candidate articles under the style combination feature.
  • the maximum click-through rate of the candidate article under the K style combination features is determined as the score of the candidate article.
  • the candidate article with the highest score on the i-th placement is determined as the article to be recommended on the i-th placement.
  • the device for recommending articles further includes:
  • the third acquiring module 1406 is configured to acquire a plurality of first articles.
  • a sorting module 1407 configured to sort the multiple first articles according to their respective content features.
  • the third determining module 1408 is configured to determine the style combination feature of the first article for any first article in the sorted first articles.
  • the training module 1409 is configured to train the first model according to the plurality of first articles after the determined styles and features are combined.
  • the third determining module 1408 is specifically configured to: for each style feature of the first article, randomly select candidate styles corresponding to the style feature, and form a style combination feature of the first article.
  • the apparatus for article recommendation further includes: a selection module 1410, configured to, for any candidate article in at least one candidate article on the i-th placement, select K out of all style combination features of the candidate article Style combination features.
  • the selection module 1410 is specifically configured to: determine the score of the style combination feature for any style combination feature in all the style combination features of the candidate article. Among all the style combination features of the candidate articles, the top K style combination features are selected as the K style combination features.
  • the selection module 1410 is specifically configured to: for any style feature in the style combination features, input user features, content features and style features of candidate articles into the second model to obtain the style features of candidate articles click through rate. Sum the click-through rates of the candidate articles under each style feature in the style combination feature to obtain the score of the style combination feature.
  • any style combination feature of the candidate article includes at least two of the following: the number of images included in the candidate article, whether to display the user avatar, and whether to display the abstract of the candidate article.
  • the device embodiment and the method embodiment may correspond to each other, and similar descriptions may refer to the method embodiment. To avoid repetition, details are not repeated here.
  • the device shown in FIG. 14 can execute the above-mentioned method embodiments, and the aforementioned and other operations and/or functions of each module in the device are for realizing corresponding processes in each method, and for the sake of brevity, details are not repeated here.
  • the device in the embodiment of the present application is described above from the perspective of functional modules with reference to the accompanying drawings.
  • the functional modules may be implemented in the form of hardware, may also be implemented by instructions in the form of software, and may also be implemented by a combination of hardware and software modules.
  • each step of the method embodiment in the embodiment of the present application can be completed by an integrated logic circuit of hardware in the processor and/or instructions in the form of software, and the steps of the method disclosed in the embodiment of the present application can be directly embodied as hardware
  • the decoding processor is executed, or the combination of hardware and software modules in the decoding processor is used to complete the execution.
  • the software module may be located in a mature storage medium in the field such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, and registers.
  • the storage medium is located in the memory, and the processor reads the information in the memory, and completes the steps in the above method embodiments in combination with its hardware.
  • FIG. 15 is a schematic block diagram of an electronic device 1500 provided by an embodiment of the present application.
  • the electronic device 1500 may include:
  • a memory 1510 and a processor 1520 the memory 1510 is used to store computer programs and transmit the program codes to the processor 1520 .
  • the processor 1520 can invoke and run a computer program from the memory 1510, so as to implement the method in the embodiment of the present application.
  • the processor 1520 can be used to execute the above-mentioned method embodiments according to the instructions in the computer program.
  • the processor 1520 may include but not limited to:
  • DSP Digital Signal Processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field Programmable Gate Array
  • the memory 1510 includes but is not limited to:
  • non-volatile memory can be read-only memory (Read-Only Memory, ROM), programmable read-only memory (Programmable ROM, PROM), erasable programmable read-only memory (Erasable PROM, EPROM), electronically programmable Erase Programmable Read-Only Memory (Electrically EPROM, EEPROM) or Flash.
  • the volatile memory can be Random Access Memory (RAM), which acts as external cache memory.
  • RAM Static Random Access Memory
  • SRAM Static Random Access Memory
  • DRAM Dynamic Random Access Memory
  • Synchronous Dynamic Random Access Memory Synchronous Dynamic Random Access Memory
  • SDRAM double data rate synchronous dynamic random access memory
  • Double Data Rate SDRAM, DDR SDRAM double data rate synchronous dynamic random access memory
  • Enhanced SDRAM, ESDRAM enhanced synchronous dynamic random access memory
  • SLDRAM synchronous connection dynamic random access memory
  • Direct Rambus RAM Direct Rambus RAM
  • the computer program can be divided into one or N modules, and the one or N modules are stored in the memory 1510 and executed by the processor 1520 to complete the application The method that embodiment provides.
  • the one or N modules may be a series of computer program instruction segments capable of accomplishing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.
  • the electronic device may also include:
  • Transceiver 1530 the transceiver 1530 can be connected to the processor 1520 or the memory 1510 .
  • the processor 1520 can control the transceiver 1530 to communicate with other devices, specifically, can send information or data to other devices, or receive information or data sent by other devices.
  • Transceiver 1530 may include a transmitter and a receiver.
  • the transceiver 1530 may further include an antenna, and the number of antennas may be one or N.
  • bus system includes not only a data bus, but also a power bus, a control bus and a status signal bus.
  • the embodiment of the present application also provides a computer storage medium, on which a computer program is stored, and when the computer program is executed by a computer, the computer can execute the method of the above method embodiment.
  • the embodiments of the present application further provide a computer program product including instructions, and when the instructions are executed by a computer, the computer executes the methods of the foregoing method embodiments.
  • the computer program product comprises one or N computer instructions.
  • the computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable device.
  • the computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transferred from a website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) to another website site, computer, server or data center.
  • the computer-readable storage medium may be any available medium that can be accessed by a computer, or a data storage device including a server, a data center, and the like integrated with one or N available media.
  • the available medium may be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a digital video disc (digital video disc, DVD)), or a semiconductor medium (such as a solid state disk (solid state disk, SSD)), etc.
  • a magnetic medium such as a floppy disk, a hard disk, or a magnetic tape
  • an optical medium such as a digital video disc (digital video disc, DVD)
  • a semiconductor medium such as a solid state disk (solid state disk, SSD)
  • modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed by hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be regarded as exceeding the scope of the embodiments of the present application.
  • the disclosed systems, devices and methods may be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the modules is only a logical function division. In actual implementation, there may be other division methods.
  • N modules or components can be combined or can be Integrate into another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or modules may be in electrical, mechanical or other forms.
  • a module described as a separate component may or may not be physically separated, and a component shown as a module may or may not be a physical module, that is, it may be located in one place, or may be distributed to N network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. For example, each functional module in each embodiment of the embodiment of the present application may be integrated into one processing module, each module may exist separately physically, or two or more modules may be integrated into one module.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • User Interface Of Digital Computer (AREA)

Abstract

提供了一种文章推荐方法、装置、设备及存储介质，包括：获取用户请求；根据用户请求确定当前网页中的M个展示位置和N个候选文章；获取M个展示位置上的至少一个候选文章各自对应的特征信息；根据M个展示位置上的至少一个候选文章各自对应的特征信息确定M个展示位置上各自的待推荐文章；向终端设备推送M个展示位置上各自的待推荐文章；其中，针对第i个展示位置上的至少一个候选文章中的任一候选文章，候选文章对应的特征信息包括：触发用户请求的用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，从而可以提高文章推荐精度。

Description

文章推荐方法、装置、设备及存储介质
相关申请的交叉引用
本申请要求于2021年07月22日提交的，申请号为202110832594.2、发明名称为“文章推荐方法、装置、设备及存储介质”的中国专利申请的优先权，该申请的全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及人工智能(Artificial Intelligence，AI)技术领域，尤其涉及一种文章推荐方法、装置、设备及存储介质。
背景技术
随着人工智能技术领域的不断进步，用户在进行网页浏览时，越来越倾向于智能化和个性化的文章推荐方式。
文章推荐方式中的重要环节是文章排序，即确定各个候选文章在网页中的展示位置，目前服务器基于用户特征和候选文章的内容特征对候选文章进行排序，这种方式导致文章推荐精度较低的问题。
发明内容
本申请实施例提供一种文章推荐方法、装置、设备及存储介质，从而提高文章推荐精度。
第一方面，提供一种文章推荐方法，包括：获取用户请求；根据用户请求确定当前网页中的M个展示位置和N个候选文章，M和N均为大于1的整数；获取M个展示位置上的至少一个候选文章各自对应的特征信息；根据M个展示位置上的至少一个候选文章各自对应的特征信息确定M个展示位置上各自的待推荐文章；向终端设备推送M个展示位置上各自的待推荐文章；其中，第1个展示位置上的至少一个候选文章为N个候选文章，第i+1个展示位置上的至少一个候选文章为第i个展示位置上的至少一个候选文章被删除了第i个展示位置上的待推荐文章之后得到的候选文章，i＝1,2……M；针对第i个展示位置上的至少一个候选文章中的任一候选文章，候选文章对应的特征信息包括：触发用户请求的用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，K为大于1的整数。
第二方面，提供一种文章推荐装置，包括：第一获取模块、第一确定模块、第二获取模块、第二确定模块和推送模块，其中，第一获取模块用于获取用户请求；第一确定模块用于根据用户请求确定当前网页中的M个展示位置和N个候选文章，M和N均为大于1的整数；第二获取模块用于获取M个展示位置上的至少一个候选文章各自对应的特征信息；第二确定模块用于根据M个展示位置上的至少一个候选文章各自对应的特征信息确定M个展示位置上各自的待推荐文章；推送模块用于向终端设备推送M个展示位置上各自的待推荐文章；其中，第1个展示位置上的至少一个候选文章为N个候选文章，第i+1个展示位置上的至少一个候选文章为第i个展示位置上的至少一个候选文章被删除了第i个展示位置上的待推荐文章之后得到的候选文章，i＝1,2……M；针对第i个展示位置上的至少一个候选文章中的任一候选文章，候选文章对应的特征信息包括：触发用户请求的用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，K为大于1的整数。
第三方面，提供一种电子设备，包括：处理器和存储器，该存储器用于存储计算机程序，该处理器用于调用并运行该存储器中存储的计算机程序，执行如第一方面或其各实现方式中的方法。
第四方面，提供一种计算机可读存储介质，用于存储计算机程序，计算机程序使得计算机执行如第一方面或其各实现方式中的方法。
第五方面，提供一种计算机程序产品，包括计算机程序指令，该计算机程序指令使得计算机执行如第一方面或其各实现方式中的方法。
第六方面，提供一种计算机程序，计算机程序使得计算机执行如第一方面或其各实现方式中的方法。
通过本申请实施例技术方案，针对第i个展示位置，服务器可以基于用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，对候选文章进行排序，即确定第i个展示位置上的待推荐文章，进而向用户推荐第i个展示位置上的待推荐文章，这种方式不仅考虑了用户特征和候选文章的内容特征，还考虑了候选文章的样式组合特征以及已排序文章的样式组合特征，从而可以提高文章推荐精度。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例， 对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1至图8为本申请实施例提供的文章展示示意图；
图9为本申请实施例提供的一种应用场景示意图；
图10为本申请实施例一实施例提供的一种文章推荐方法的流程图；
图11为本申请实施例提供的第一模型的示意图；
图12为本申请实施例另一实施例提供的一种文章推荐方法的流程图；
图13为本申请实施例提供的第二模型的示意图；
图14为本申请实施例提供的一种文章推荐装置的示意图；
图15是本申请实施例提供的电子设备1500的示意性框图。
具体实施方式
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本发明保护的范围。
需要说明的是，本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或服务器不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
在介绍本申请实施例技术方案之前，下面将对本申请实施例的相关知识进行详细阐述：
应理解的是，候选文章的样式组合特征是对候选文章的至少一个样式组合得到的特征，例如：候选文章的一种样式特征是：显示某数量的图像，或者，候选文章中显示某数量的大图或者小图。候选文章的另一种样式特征是：候选文章中显示或者不显示用户头像。候选文章的再一种样式特征是：候选文章显示或者不显示摘要。基于此，图1至图8为本申请实施例提供的文章展示示意图，如图1所示，该候选文章显示的样式组合特征包括：该候选文章中不显示图像，且不显示用户头像和文章摘要。如图2所示，该候选文章显示的样式组合特征包括：该候选文章中显示一副图像，且不显示 用户头像和文章摘要。如图3所示，该候选文章显示的样式组合特征包括：该候选文章中显示一副大图，且不显示用户头像和文章摘要。如图4所示，该候选文章显示的样式组合特征包括：该候选文章中显示三副小图，且不显示用户头像和文章摘要。如图5所示，该候选文章显示的样式组合特征包括：该候选文章中显示两幅小图，且显示用户头像，而不显示文章摘要。如图6所示，该候选文章显示的样式组合特征包括：该候选文章中显示一副大图，且显示用户头像，而不显示文章摘要。如图7所示，该候选文章显示的样式组合特征包括：该候选文章中不显示图像，且不显示用户头像，而显示文章摘要。如图8所示，该候选文章显示的样式组合特征包括：该候选文章中显示三幅小图，且不显示用户头像，而显示文章摘要。总之，本申请实施例对候选文章的样式特征以及样式组合特征不做限制。
应理解的是，某一候选文章的点击率是指网页上该候选文章被点击的次数与被显示次数之比，即clicks/views，它是一个百分比。反映了网页上该候选文章的受关注程度，其可以用来衡量候选文章的吸引程度。
目前服务器基于用户特征和候选文章的内容特征对候选文章进行排序。在服务器基于用户特征和候选文章的内容特征对候选文章进行排序之后，服务器再确定每个展示位置上的候选文章的展示样式。一方面，这种排序方式并未考虑候选文章的样式组合特征，从而导致文章推荐精度较低的问题。另一方面，目前的文章排序方式未考虑已排序文章的样式组合特征，从而导致文章推荐精度较低的问题。例如，前面连续显示了3篇无图文章，再显示一篇大图文章和连续显示4篇大图文章，前者最后一篇大图文章就更容易被点击。这是因为前者3篇无图文章显示较为单调，而有图文章相对于无图文章信息更为丰富，而后者的三篇文章信息变化不大。
针对上述技术问题，本申请实施例将考虑候选文章的样式组合特征以及其之前以排序文章的样式组合特征，从而提高文章推荐精度。
本申请实施例技术方案可以应用于如下场景，但不限于此：图9为本申请实施例提供的一种应用场景示意图，如图9所示，终端设备910可以与服务器920进行通信，其中，终端设备910可以获取并向服务器920发送用户请求，该用户请求用于请求浏览某网页，服务器920即该网页对应的网页浏览服务器获取到该用户请求后，按照本申请实施例提供的技术方案为各个候选文章确定展示位置，并最终向中终端设备910推送各个展示位置上的文章。
在一些可实现方式中，终端设备可以是智能手机、平板电脑、笔记本电脑、台式计算机、智能音箱、可穿戴设备等，但并不限于此，本申请实施例对此不做限制。
在一些可实现方式中，服务器可以是独立的物理服务器，也可以是多个物理服务 器构成的服务器集群或者分布式系统，还可以是提供云计算服务的云服务器。
在一些可实现方式中，终端设备与服务器之间、服务器之间可以采用蜂窝移动网络或者无线保真(Wireless Fidelity，WiFi)等进行通信，本申请实施例对此不做限制。
下面将对本申请实施例技术方案进行详细阐述：
图10为本申请实施例一实施例提供的一种文章推荐方法的流程图，该方法可以由图9中的服务器执行，但不限于此，下面以该服务器执行该文章推荐方法为示例，对本申请实施例技术方案进行说明，如图10所示，该方法包括如下步骤：
S1010：获取用户请求。
S1020：根据用户请求确定当前网页中的M个展示位置和N个候选文章，M和N均为大于1的整数。
S1030：获取M个展示位置上的至少一个候选文章各自对应的特征信息，其中，针对第i个展示位置上的至少一个候选文章中的任一候选文章，候选文章对应的特征信息包括：触发用户请求的用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，i＝1,2……M，K为大于1的整数。
S1040：根据M个展示位置上的至少一个候选文章各自对应的特征信息确定M个展示位置上各自的待推荐文章。
S1050：向终端设备推送M个展示位置上各自的待推荐文章。
应理解的是，该用户请求用于请求浏览某网页，该用户请求可以是用户在终端设备上点击该网页链接所触发的，或者是用户在终端设备输入该网页网址，并按下回车键所触发的，本申请实施例对如何触发或者获取用户请求不做限制。
应理解的是，上述M个展示位置是当前网页中可以展示候选文章的位置，该当前网页是用户请求浏览的网页。
应理解的是，N个候选文章可以是同一主题的N个候选文章，例如：N个候选文章都是关于某人物的介绍，或者，N个候选文章可以不是同一主题的N个候选文章，例如：当用户点击某浏览器链接时，这时终端设备所显示的网页中包括：各种热门主题的文章等。
在一些可实现方式中，M小于或等于N。即候选文章的数量大于或等于当前网页中可显示候选文章的展示位置。或者，M大于N。即候选文章的数量小于当前网页中可显示候选文章的展示位置。总之，本申请实施例对M和N的大小关系不做限制。
应理解的是，第1个展示位置上的至少一个候选文章为N个候选文章，第i+1个展示位置上的至少一个候选文章为第i个展示位置上的至少一个候选文章被删除了第i 个展示位置上的待推荐文章之后得到的候选文章，i＝1,2……M。具体的，服务器可以对上述M个展示位置进行排序，使得M个展示位置都具有唯一的索引。对于第1个展示位置而言，该位置上的候选文章为上述N个候选文章，基于此，服务器可以先确定该第1个展示位置上的待推荐文章，进一步地，对于第2个展示位置而言，该位置上的候选文章为上述N个候选文章被删除了第1个展示位置上的待推荐文章之后的候选文章，基于此，服务器可以确定该第2个展示位置上的待推荐文章，以此类推，服务器可以确定每个展示位置上的候选文章，并在每个展示位置上的候选文章中确定每个展示位置上的待推荐文章。
应理解的是，服务器可以采用如下任一种可实现方式对M个展示位置进行排序，但不限于此：
可实现方式一，服务器可以按照M个展示位置的优先级对M个展示位置进行排序，其中，优先级越高的展示位置，其排序越靠前，相反，优先级越低的展示位置，其排序越靠后。也就是说，优先级越高的展示位置，其对应的索引越小，相反，优先级越低的展示位置，其对应的索引越大。M个展示位置的优先级与M个展示位置对用户的吸引度有关，例如：当前网页中最中心的展示位置的吸引度最高，因此，该展示位置的优先级最高，而当前网页中最左上角的展示位置的吸引度次之，因此，该展示位置的优先级相对于最中心展示位置的优先级次之，以此类推。
可实现方式二，服务器可以按照M个展示位置的优先级对M个展示位置进行排序，其中，优先级越高的展示位置，其排序越靠前，相反，优先级越低的展示位置，其排序越靠后。也就是说，优先级越高的展示位置，其对应的索引越小，相反，优先级越低的展示位置，其对应的索引越大。服务器可以按照从左至右、从上至下的顺序确定M个展示位置的优先级，例如：当前网页中最左上角的展示位置的优先级最高，该最左上角的展示位置的下一个展示位置的优先级次之，以此类推。
可实现方式三，服务器可以按照M个展示位置对用户的吸引度对M个展示位置进行排序。例如：当前网页中最中心的展示位置的吸引度最高，因此，该展示位置排在第一位，而当前网页中最左上角的展示位置的吸引度次之，因此，该展示位置排在第二位，以此类推。
可实现方式四，服务器可以按照从左至右、从上至下的顺序对M个展示位置进行排序，例如：当前网页中最左上角的展示位置排在第一位，该最左上角的展示位置的下一个展示位置排在第二位，以此类推。
应理解的是，上述M个展示位置，任意排序相邻的两个展示位置，它们在当前网页上可以相邻或者不相邻，本申请实施例对此不做限制。
如上所述，针对第i个展示位置上的至少一个候选文章中的任一候选文章，候选文章对应的特征信息包括：触发用户请求的用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，i＝1,2……M，K为大于1的整数。
在一些可实现方式中，触发用户请求的用户特征包括以下至少一项，但不限于此：该用户的年龄、性别、喜好、学历等。
在一些可实现方式中，候选文章的内容特征包括以下至少一项，但不限于此：该候选文章所属类别、该候选文章的关键词、作者、该候选文章在其所属类别的点击率等。
在一些可实现方式中，候选文章所属类别可以按照文章的表达方式划分，包括如下几项，但不限于此：记叙文、说明文、议论文等。或者，候选文章所属类别可以按照文学作品划分，包括如下几项，但不限于此：诗歌、小说、散文(抒情散文、叙事散文)、剧本等。候选文章所属类别也可以按照表达方式、修辞手法、表现手法等划分，总之，本申请实施例对候选文章的类别划分标准不做限制。
在一些可实现方式中，上述候选文章的K个样式组合特征可以是该候选文章的所有样式组合特征，也可以是该候选文章的所有样式组合特征中的部分样式组合特征，本申请实施例对此不做限制。
示例性的，假设K＝2，针对候选文章item，服务器确定该候选文章item的2个样式组合特征分别是：该候选文章中显示一副大图，且显示用户头像，而不显示文章摘要。以及，该候选文章中显示三幅小图，且不显示用户头像，而显示文章摘要。
应理解的是，第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征包括：第1个展示位置上已确定的待推荐文章的样式组合特征、第2个展示位置上已确定的待推荐文章的样式组合特征……第i-1个展示位置上的已确定的待推荐文章的样式组合特征。例如：i＝3，即服务器当前正在确定第3个展示位置上的待推荐文章，那么第1个展示位置至第2个展示位置上的已确定的待推荐文章的样式组合特征包括：第1个展示位置上已确定的待推荐文章的样式组合特征和第2个展示位置上已确定的待推荐文章的样式组合特征，例如：第1个展示位置上已确定的待推荐文章的样式组合特征为：该待推荐文章中显示一副大图，且显示用户头像，而不显示文章摘要。第2个展示位置上已确定的待推荐文章的样式组合特征为：该待推荐文章中显示两幅小图，且显示用户头像，显示文章摘要。
应理解的是，服务器可以通过如下任一可实现方式确定M个展示位置上各自的待推荐文章，但不限于此：
可实现方式一：针对第i个展示位置上的至少一个候选文章中的任一候选文章，以及候选文章的K个样式组合特征中的任一样式组合特征，服务器将用户特征、候选文章的内容特征、样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征输入至第一模型中，以得到候选文章在样式组合特征下的点击率。服务器将候选文章在K个样式组合特征下的最大点击率，确定为候选文章的得分。服务器将第i个展示位置上得分最高的候选文章，确定为第i个展示位置上的待推荐文章。
可实现方式二：针对第i个展示位置上的至少一个候选文章中的任一候选文章，以及候选文章的K个样式组合特征中的任一样式组合特征，服务器将用户特征、候选文章的内容特征、样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征输入至第一模型中，以得到候选文章在样式组合特征下的点击率。服务器将候选文章在K个样式组合特征下的平均点击率，确定为候选文章的得分。服务器将第i个展示位置上得分最高的候选文章，确定为第i个展示位置上的待推荐文章。
示例性的，图11为本申请实施例提供的第一模型的示意图，如图11所示，该第一模型的输入是用户特征、候选文章的内容特征、某一样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征。该第一模型的输出是候选文章在该某一样式组合特征下的点击率。
针对可实现方式一进行说明：
假设在一次用户请求中，服务器确定有M个展示位置，服务器可以针对第1个展示位置到第M个展示位置依次从候选文章中选取一个待推荐文章填入对应位置。设当前需要确定第i个展示位置上的待推荐文章，该展示位置对应的候选文章有C个，假设这C个候选文章构成的候选集合为{item 1,item 2……item C}，令score(item j)为候选文章item j的得分。在排序过程中，服务器可以选取分数最大的候选文章，排在第i个展示位置上。其中，候选文章item j的得分可以按照如下公式(1)计算：
Figure PCTCN2022106208-appb-000001
Model multi()表示第一模型。
max{}表示取最大值。
F user表示用户特征，其中，对于所有候选文章而言，其对应的用户特征都相同，另外，针对不同展示位置上的同一候选文章，其对应的用户特征也是相同的。
Figure PCTCN2022106208-appb-000002
表示候选文章item j的内容特征，其中，针对不同的候选文章，它们对应的内容特征不同，但是针对不同展示位置上的同一候选文章，其对应的内容特征是相同的。
Figure PCTCN2022106208-appb-000003
表示候选文章item j的一个样式特征组合，例如：(1,j 1)表示在第1个样式维度中选择第j 1种样式，(2,j 2)表示在第2个样式维度中选择第j 2种样式……(N,j N)表示在第N个样式维度中选择第j N种样式。其中，候选文章item j包括K个样式特征组合。
Figure PCTCN2022106208-appb-000004
表示第1个展示位置，第2个展示位置……第i-1个展示位置上已确定的待推荐文章的样式组合特征。
针对可实现方式二进行说明：
假设在一次用户请求中，服务器确定有M个展示位置，服务器可以针对第1个展示位置到第M个展示位置依次从候选文章中选取一个待推荐文章填入对应位置。设当前需要确定第i个展示位置上的待推荐文章，该展示位置对应的候选文章有C个，假设这C个候选文章构成的候选集合为{item 1,item 2……item C}，令score(item j)为候选文章item j的得分。在排序过程中，服务器可以选取分数最大的候选文章，排在第i个展示位置上。其中，候选文章item j的得分可以按照如下公式(2)计算：
Figure PCTCN2022106208-appb-000005
Model multi()表示第一模型。
mean{}表示取平均值。
F user表示用户特征，其中，对于所有候选文章而言，其对应的用户特征都相同，另外，针对不同展示位置上的同一候选文章，其对应的用户特征也是相同的。
Figure PCTCN2022106208-appb-000006
表示候选文章itemj的内容特征，其中，针对不同的候选文章，它们对应的内容特征不同，但是针对不同展示位置上的同一候选文章，其对应的内容特征是相同的。
Figure PCTCN2022106208-appb-000007
表示候选文章item j的一个样式特征组合，例如：(1,j 1)表示在第1个样式维度中选择第j 1种样式，(2,j 2)表示在第2个样式维度中选择第j 2种样式……(N,j N)表示在第N个样式维度中选择第j N种样式。其中，候选文章itemj包括K个样式特征组合。
Figure PCTCN2022106208-appb-000008
表示第1个展示位置，第2个展示位置……第i-1个展示位置上已确定的待推荐文章的样式组合特征。
进一步地，服务器确定了M个展示位置上各自的待推荐文章之后，可以向终端设备推送M个展示位置上各自对应的待推荐文章，以使终端设备可以在当前网页上在上述M个展示位置上展示各自对应的待推荐文章。
在一些可实现方式中，针对M个展示位置中的任一个展示位置，该展示位置上的待推荐文章可以以目标样式组合特征展示，该目标样式组合特征是服务器在确定该待推荐文章的得分时，点击率最大的样式组合特征，当然，该目标样式组合特征是服务器在确定该待推荐文章的得分时，点击率次大的样式组合特征。需要说明的是，该目标样式组合特征的确定方式不限于此。
应理解的是，图10所示的文章推荐方法从执行代码的角度来描述，具体的，图12为本申请实施例另一实施例提供的一种文章推荐方法的流程图，如图12所示，该方法包括如下步骤：
S1：获取用户请求；
S2：根据用户请求确定当前网页中的M个展示位置和N个候选文章，M和N均为大于1的整数，令i＝1，i表示展示位置的索引；
S3：获取第i个展示位置上的至少一个候选文章各自对应的特征信息；
S4：根据第i个展示位置上的至少一个候选文章各自对应的特征信息确定第i个展示位置上的待推荐文章；
S5：令i＝i+1，判断i是否小于或等于M，若i小于或等于M，则继续执行S3，否则，执行S6；
S6：向终端设备推送M个展示位置上各自的待推荐文章。
其中，第1个展示位置上的至少一个候选文章为N个候选文章，第i+1个展示位置上的至少一个候选文章为第i个展示位置上的至少一个候选文章被删除了第i个展示位置上的待推荐文章之后得到的候选文章；针对第i个展示位置上的至少一个候选文章中的任一候选文章，候选文章对应的特征信息包括：触发用户请求的用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，K为大于1的整数。
综上，在本申请实施例中，针对第i个展示位置，服务器可以基于用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，对候选文章进行排序，即确定第i个展示位 置上的待推荐文章，进而向用户推荐第i个展示位置上的待推荐文章，这种方式不仅考虑了用户特征和候选文章的内容特征，还考虑了候选文章的样式组合特征以及已排序文章的样式组合特征，从而可以提高文章推荐精度。
如上所述，针对任一候选文章，该候选文章的K个样式组合特征可以是该候选文章的所有样式组合特征，也可以是该候选文章的所有样式组合特征中的部分样式组合特征，对于后者，也就是说，针对第i个展示位置上的至少一个候选文章中的任一候选文章，服务器可以在该候选文章的所有样式组合特征中选择K个样式组合特征。
应理解的是，服务器可以通过如下可实现方式在候选文章的所有样式组合特征中选择K个样式组合特征，但不限于此：
可实现方式一：针对候选文章的所有样式组合特征中的任一样式组合特征，服务器确定样式组合特征的得分，服务器在候选文章的所有样式组合特征中选择得分在前K个的样式组合特征，以作为K个样式组合特征。
可实现方式二：针对候选文章的所有样式组合特征中的任一样式组合特征，服务器确定样式组合特征的得分，服务器在候选文章的所有样式组合特征中选择得分在前P个的样式组合特征，P为大于K的整数，进一步地，服务器在这P个样式组合特征中选择K个样式组合特征，以作为K个样式组合特征。其中，服务器可以在这P个样式组合特征中随机选择K个样式组合特征或者按照一定的预设规则选择K个样式组合特征，本申请实施例对该预设规则不做限制。
应理解的是，服务器可以通过如下任一可实现方式确定样式组合特征的得分，但不限于此：
可实现方式一：服务器可以针对样式组合特征中的任一样式特征，将用户特征、候选文章的内容特征和样式特征输入至第二模型，以得到候选文章在该样式特征下的点击率，进一步地，服务器对候选文章在样式组合特征中各个样式特征下的点击率求和，以得到样式组合特征的得分。
可实现方式二：服务器可以针对样式组合特征中的任一样式特征，将用户特征、候选文章的内容特征和样式特征输入至第二模型，以得到候选文章在样式特征下的点击率，进一步地，服务器对候选文章在样式组合特征中各个样式特征下的点击率求平均值，以得到样式组合特征的得分。
应理解的是，上述样式组合特征中的任一样式特征是该样式组合特征中某一维具体样式特征，例如：该样式特征可以是：候选文章中显示三幅小图，或者，候选文章中不显示用户头像，或者，候选文章中不显示文章摘要等。
示例性的，图13为本申请实施例提供的第二模型的示意图，如图13所示，该第 二模型的输入是用户特征、候选文章的内容特征、和某样式特征。该第二模型的输出是候选文章item j在第i维样式特征中第j i个样式特征下的点击率。
针对上述可实现方式一和可实现方式二，服务器可以通过如下公式(2)确定候选文章在该样式特征下的点击率：
Figure PCTCN2022106208-appb-000009
Figure PCTCN2022106208-appb-000010
表示候选文章在该样式特征下的点击率。
Figure PCTCN2022106208-appb-000011
表示第一模型。
F user表示用户特征，其中，对于所有候选文章而言，其对应的用户特征都相同，另外，针对不同展示位置上的同一候选文章，其对应的用户特征也是相同的。
Figure PCTCN2022106208-appb-000012
表示候选文章item j的内容特征，其中，针对不同的候选文章，它们对应的内容特征不同，但是针对不同展示位置上的同一候选文章，其对应的内容特征是相同的。
针对上述可实现方式一，服务器可以通过如下公式(3)确定候选文章在某样式组合特征的得分，但不限于此：
Figure PCTCN2022106208-appb-000013
针对上述可实现方式二，服务器可以通过如下公式(4)确定候选文章在某样式组合特征的得分，但不限于此：
Figure PCTCN2022106208-appb-000014
mean{}表示取平均值。
应理解的是，假设目前存在N维样式特征，而第1维样式特征包括n(1)种样式特征，第2维样式特征包括n(2)种样式特征……第N维样式特征包括n(N)种样式特征，那么如果服务器按照上述公式(1)计算候选文章item j的得分，则服务器的计算量是
Figure PCTCN2022106208-appb-000015
而如果服务器按照公式(3)和(4)选择K个样式组合特征，虽然服务器的计算量也是
Figure PCTCN2022106208-appb-000016
但是第二模型相对于第一模型的计算复杂度低很多，因 此，通过服务器采用的减枝操作，即在只选择K个样式组合特征，可以降低计算复杂度，进而可以提高文章推荐效率。
此外，现有技术中，将样式组合特征选择问题看作一个多臂抽奖问题(Multi-Armed Bandit)，通过尽可能快速的探索，寻找点击率最高的样式组合特征。然而通过这种方式得到的样式组合特征不具有个性化，即对于所有的用户，同一候选文章的样式组合特征相同。然而，当采用本申请实施例技术方案时，由于服务器将用户特征、候选文章的内容特征、样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征输入至第一模型中，以得到候选文章在所述样式组合特征下的点击率，将候选文章在K个样式组合特征下的最大点击率，可以将最大点击率对应的样式组合特征确定为该候选文章的待展示的样式组合特征。也就是说，单从样式组合特征选择角度来讲，最终选择的样式组合特征考虑了用户特征，通过这种方式得到的样式组合特征具有个性化，即对于不同的用户，同一候选文章的样式组合特征可以不同，从而可以提高用户体验感。
应理解的是，为了提高文章推荐精度，服务器可以对第一模型进行训练，具体可通过如下可实现方式训练第一模型，但不限于此：
在一些可实现方式中，服务器获取多个第一文章，根据多个第一文章各自的内容特征对多个第一文章排序，针对经过排序后的第一文章中的任一第一文章，确定第一文章的样式组合特征，根据已确定样式组合特征后的多个第一文章训练第一模型。
其中，针对第一文章的每一个样式特征，在样式特征对应的候选样式中随机选择，并组成第一文章的样式组合特征。或者，针对第一文章的每一个样式特征，在样式特征对应的候选样式中按照预设规则选择，并组成第一文章的样式组合特征。本申请实施例对如何在样式特征对应的候选样式中选择不做限制。
示例性的，对应任一候选文章，假设目前存在N维样式特征，而第i维样式特征包括n(i)种样式特征，服务器针对第i维样式特征，其可以在n(i)种样式特征中随机选择样式特征。
应理解的是，为了更好的训练第一模型，服务器期望所有的样式组合特征尽可能的出现，因此，服务器在训练第一模型时，可以在样式特征对应的候选样式中随机选择，从而组成多种多样的样式组合特征，包括：可能极少出现的样式组合特征。
图14为本申请实施例提供的一种文章推荐装置的示意图，该装置可以是图9中的服务器，但不限于此，如图14所示，该装置包括：
第一获取模块1401，用于获取用户请求。
第一确定模块1402，用于根据用户请求确定当前网页中的M个展示位置和N个候 选文章，M和N均为大于1的整数。
第二获取模块1403，用于获取M个展示位置上的至少一个候选文章各自对应的特征信息。
第二确定模块1404，用于根据M个展示位置上的至少一个候选文章各自对应的特征信息确定M个展示位置上各自的待推荐文章。
推送模块1405，用于向终端设备推送M个展示位置上各自的待推荐文章。
其中，第1个展示位置上的至少一个候选文章为N个候选文章，第i+1个展示位置上的至少一个候选文章为第i个展示位置上的至少一个候选文章被删除了第i个展示位置上的待推荐文章之后得到的候选文章，i＝1,2……M。
针对第i个展示位置上的至少一个候选文章中的任一候选文章，候选文章对应的特征信息包括：触发用户请求的用户特征、候选文章的内容特征、候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，K为大于1的整数。
在一些可实现方式中，第二确定模块1404具体用于：针对第i个展示位置上的至少一个候选文章中的任一候选文章，以及候选文章的K个样式组合特征中的任一样式组合特征，将用户特征、候选文章的内容特征、样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征输入至第一模型中，以得到候选文章在样式组合特征下的点击率。将候选文章在K个样式组合特征下的最大点击率，确定为候选文章的得分。将第i个展示位置上得分最高的候选文章，确定为第i个展示位置上的待推荐文章。
在一些可实现方式中，文章推荐装置还包括：
第三获取模块1406，用于获取多个第一文章。
排序模块1407，用于根据多个第一文章各自的内容特征对多个第一文章排序。
第三确定模块1408，用于针对经过排序后的第一文章中的任一第一文章，确定第一文章的样式组合特征。
训练模块1409，用于根据已确定样式组合特征后的多个第一文章训练第一模型。
在一些可实现方式中，第三确定模块1408具体用于：针对第一文章的每一个样式特征，在样式特征对应的候选样式中随机选择，并组成第一文章的样式组合特征。
在一些可实现方式中，文章推荐装置还包括：选择模块1410，用于针对第i个展示位置上的至少一个候选文章中的任一候选文章，在候选文章的所有样式组合特征中选择K个样式组合特征。
在一些可实现方式中，选择模块1410具体用于：针对候选文章的所有样式组合特 征中的任一样式组合特征，确定样式组合特征的得分。在候选文章的所有样式组合特征中选择得分在前K个的样式组合特征，以作为K个样式组合特征。
在一些可实现方式中，选择模块1410具体用于：针对样式组合特征中的任一样式特征，将用户特征、候选文章的内容特征和样式特征输入至第二模型，以得到候选文章在样式特征下的点击率。对候选文章在样式组合特征中各个样式特征下的点击率求和，以得到样式组合特征的得分。
在一些可实现方式中，候选文章的任一样式组合特征包括以下至少两项：候选文章包括的图像数量、是否显示用户头像、是否显示候选文章的摘要。
应理解的是，装置实施例与方法实施例可以相互对应，类似的描述可以参照方法实施例。为避免重复，此处不再赘述。具体地，图14所示的装置可以执行上述方法实施例，并且装置中的各个模块的前述和其它操作和/或功能分别为了实现各个方法中的相应流程，为了简洁，在此不再赘述。
上文中结合附图从功能模块的角度描述了本申请实施例的装置。应理解，该功能模块可以通过硬件形式实现，也可以通过软件形式的指令实现，还可以通过硬件和软件模块组合实现。具体地，本申请实施例中的方法实施例的各步骤可以通过处理器中的硬件的集成逻辑电路和/或软件形式的指令完成，结合本申请实施例公开的方法的步骤可以直接体现为硬件译码处理器执行完成，或者用译码处理器中的硬件及软件模块组合执行完成。可选地，软件模块可以位于随机存储器，闪存、只读存储器、可编程只读存储器、电可擦写可编程存储器、寄存器等本领域的成熟的存储介质中。该存储介质位于存储器，处理器读取存储器中的信息，结合其硬件完成上述方法实施例中的步骤。
图15是本申请实施例提供的电子设备1500的示意性框图。
如图15所示，该电子设备1500可包括：
存储器1510和处理器1520，该存储器1510用于存储计算机程序，并将该程序代码传输给该处理器1520。换言之，该处理器1520可以从存储器1510中调用并运行计算机程序，以实现本申请实施例中的方法。
例如，该处理器1520可用于根据该计算机程序中的指令执行上述方法实施例。
在本申请实施例的一些实施例中，该处理器1520可以包括但不限于：
通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等等。
在本申请实施例的一些实施例中，该存储器1510包括但不限于：
易失性存储器和/或非易失性存储器。其中，非易失性存储器可以是只读存储器(Read-Only Memory，ROM)、可编程只读存储器(Programmable ROM，PROM)、可擦除可编程只读存储器(Erasable PROM，EPROM)、电可擦除可编程只读存储器(Electrically EPROM，EEPROM)或闪存。易失性存储器可以是随机存取存储器(Random Access Memory，RAM)，其用作外部高速缓存。通过示例性但不是限制性说明，许多形式的RAM可用，例如静态随机存取存储器(Static RAM，SRAM)、动态随机存取存储器(Dynamic RAM，DRAM)、同步动态随机存取存储器(Synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(Double Data Rate SDRAM，DDR SDRAM)、增强型同步动态随机存取存储器(Enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(synch link DRAM，SLDRAM)和直接内存总线随机存取存储器(Direct Rambus RAM，DR RAM)。
在本申请实施例的一些实施例中，该计算机程序可以被分割成一个或N个模块，该一个或者N个模块被存储在该存储器1510中，并由该处理器1520执行，以完成本申请实施例提供的方法。该一个或N个模块可以是能够完成特定功能的一系列计算机程序指令段，该指令段用于描述该计算机程序在该电子设备中的执行过程。
如图15所示，该电子设备还可包括：
收发器1530，该收发器1530可连接至该处理器1520或存储器1510。
其中，处理器1520可以控制该收发器1530与其他设备进行通信，具体地，可以向其他设备发送信息或数据，或接收其他设备发送的信息或数据。收发器1530可以包括发射机和接收机。收发器1530还可以进一步包括天线，天线的数量可以为一个或N个。
应当理解，该车流量控制设备中的各个组件通过总线系统相连，其中，总线系统除包括数据总线之外，还包括电源总线、控制总线和状态信号总线。
本申请实施例还提供了一种计算机存储介质，其上存储有计算机程序，该计算机程序被计算机执行时使得该计算机能够执行上述方法实施例的方法。或者说，本申请实施例还提供一种包含指令的计算机程序产品，该指令被计算机执行时使得计算机执行上述方法实施例的方法。
当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。该计算机程序产品包括一个或N个计算机指令。在计算机上加载和执行该计算机程序指令时，全部或部分地产生按照本申请实施例该的流程或功能。该计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。该计算机指令可以存储在计算机可 读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，该计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(digital subscriber line，DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。该计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或N个可用介质集成的服务器、数据中心等数据存储设备。该可用介质可以是磁性介质(例如，软盘、硬盘、磁带)、光介质(例如数字视频光盘(digital video disc，DVD))、或者半导体介质(例如固态硬盘(solid state disk，SSD))等。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的模块及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请实施例的范围。
在本申请实施例所提供的几个实施例中，应该理解到，所揭露的系统、装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，该模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如N个模块或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或模块的间接耦合或通信连接，可以是电性，机械或其它的形式。
作为分离部件说明的模块可以是或者也可以不是物理上分开的，作为模块显示的部件可以是或者也可以不是物理模块，即可以位于一个地方，或者也可以分布到N个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。例如，在本申请实施例各个实施例中的各功能模块可以集成在一个处理模块中，也可以是各个模块单独物理存在，也可以两个或两个以上模块集成在一个模块中。
以上该，仅为本申请实施例的具体实施方式，但本申请实施例的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本申请实施例揭露的技术范围内，可轻易想到变化或替换，都应涵盖在本申请实施例的保护范围之内。因此，本申请实施例的保护范围应以该权利要求的保护范围为准。

Claims (11)

  1. 一种文章推荐方法，其特征在于，包括：
    获取用户请求；
    根据所述用户请求确定当前网页中的M个展示位置和N个候选文章，M和N均为大于1的整数；
    获取所述M个展示位置上的至少一个候选文章各自对应的特征信息；
    根据所述M个展示位置上的至少一个候选文章各自对应的特征信息确定所述M个展示位置上各自的待推荐文章；
    向终端设备推送所述M个展示位置上各自的待推荐文章；
    其中，第1个展示位置上的至少一个候选文章为所述N个候选文章，第i+1个展示位置上的至少一个候选文章为第i个展示位置上的至少一个候选文章被删除了第i个展示位置上的待推荐文章之后得到的候选文章，i＝1,2……M；
    针对所述第i个展示位置上的至少一个候选文章中的任一候选文章，所述候选文章对应的特征信息包括：触发所述用户请求的用户特征、所述候选文章的内容特征、所述候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，K为大于1的整数。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述M个展示位置上的至少一个候选文章各自对应的特征信息确定所述M个展示位置上各自的待推荐文章，包括：
    针对所述第i个展示位置上的至少一个候选文章中的任一候选文章，以及所述候选文章的K个样式组合特征中的任一样式组合特征，将所述用户特征、所述候选文章的内容特征、所述样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征输入至第一模型中，以得到所述候选文章在所述样式组合特征下的点击率；
    将所述候选文章在K个样式组合特征下的最大点击率，确定为所述候选文章的得分；
    将所述第i个展示位置上得分最高的候选文章，确定为所述第i个展示位置上的待推荐文章。
  3. 根据权利要求2所述的方法，其特征在于，还包括：
    获取多个第一文章；
    根据所述多个第一文章各自的内容特征对所述多个第一文章排序；
    针对经过排序后的所述第一文章中的任一第一文章，确定所述第一文章的样式组合特征；
    根据已确定样式组合特征后的所述多个第一文章训练所述第一模型。
  4. 根据权利要求3所述的方法，其特征在于，所述确定所述第一文章的样式组合特征，包括：
    针对所述第一文章的每一个样式特征，在所述样式特征对应的候选样式中随机选择，并组成所述第一文章的样式组合特征。
  5. 根据权利要求2-4任一项所述的方法，其特征在于，还包括：
    针对所述第i个展示位置上的至少一个候选文章中的任一候选文章，在所述候选文章的所有样式组合特征中选择所述K个样式组合特征。
  6. 根据权利要求5所述的方法，其特征在于，所述在所述候选文章的所有样式组合特征中选择所述K个样式组合特征，包括：
    针对所述候选文章的所有样式组合特征中的任一样式组合特征，确定所述样式组合特征的得分；
    在所述候选文章的所有样式组合特征中选择得分在前K个的样式组合特征，以作为所述K个样式组合特征。
  7. 根据权利要求6所述的方法，其特征在于，所述确定所述样式组合特征的得分，包括：
    针对所述样式组合特征中的任一样式特征，将所述用户特征、所述候选文章的内容特征和所述样式特征输入至第二模型，以得到所述候选文章在所述样式特征下的点击率；
    对所述候选文章在所述样式组合特征中各个样式特征下的点击率求和，以得到所述样式组合特征的得分。
  8. 根据权利要求1-4任一项所述的方法，其特征在于，所述候选文章的任一样式组合特征包括以下至少两项：所述候选文章包括的图像数量、是否显示用户头像、是否显示所述候选文章的摘要。
  9. 一种文章推荐装置，其特征在于，包括：
    第一获取模块，用于获取用户请求；
    第一确定模块，用于根据所述用户请求确定当前网页中的M个展示位置和N个候选文章，M和N均为大于1的整数；
    第二获取模块，用于获取所述M个展示位置上的至少一个候选文章各自对应的特征信息；
    第二确定模块，用于根据所述M个展示位置上的至少一个候选文章各自对应的特征信息确定所述M个展示位置上各自的待推荐文章；
    推送模块，用于向终端设备推送所述M个展示位置上各自的待推荐文章；
    其中，第1个展示位置上的至少一个候选文章为所述N个候选文章，第i+1个展示位置上的至少一个候选文章为第i个展示位置上的至少一个候选文章被删除了第i个展示位置上的待推荐文章之后得到的候选文章，i＝1,2……M；
    针对所述第i个展示位置上的至少一个候选文章中的任一候选文章，所述候选文章对应的特征信息包括：触发所述用户请求的用户特征、所述候选文章的内容特征、所述候选文章的K个样式组合特征和第1个展示位置至第i-1个展示位置上的已确定的待推荐文章的样式组合特征，K为大于1的整数。
  10. 一种电子设备，其特征在于，包括：
    处理器和存储器，所述存储器用于存储计算机程序，所述处理器用于调用并运行所述存储器中存储的计算机程序，以执行权利要求1至8中任一项所述的方法。
  11. 一种计算机可读存储介质，其特征在于，用于存储计算机程序，所述计算机程序使得计算机执行如权利要求1至8中任一项所述的方法。
PCT/CN2022/106208 2021-07-22 2022-07-18 文章推荐方法、装置、设备及存储介质 Ceased WO2023001093A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110832594.2A CN115687741A (zh) 2021-07-22 2021-07-22 文章推荐方法、装置、设备及存储介质
CN202110832594.2 2021-07-22

Publications (1)

Publication Number Publication Date
WO2023001093A1 true WO2023001093A1 (zh) 2023-01-26

Family

ID=84978921

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/106208 Ceased WO2023001093A1 (zh) 2021-07-22 2022-07-18 文章推荐方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN115687741A (zh)
WO (1) WO2023001093A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116186412A (zh) * 2023-03-08 2023-05-30 上海哔哩哔哩科技有限公司 推荐信息展示方法及装置

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110313985A1 (en) * 2010-06-21 2011-12-22 Jianwei Dian User adjustable ranking and presentation of search results
CN108776676A (zh) * 2018-02-02 2018-11-09 腾讯科技（深圳）有限公司 信息推荐方法、装置、计算机可读介质及电子设备
CN109255070A (zh) * 2018-08-01 2019-01-22 百度在线网络技术（北京）有限公司 推荐信息处理方法、装置、计算机设备及存储介质
CN109582903A (zh) * 2018-11-29 2019-04-05 广州市百果园信息技术有限公司 一种信息展示的方法、装置、设备和存储介质
CN111651692A (zh) * 2020-06-02 2020-09-11 腾讯科技（北京）有限公司 基于人工智能的信息推荐方法、装置及电子设备
CN112398947A (zh) * 2020-11-18 2021-02-23 腾讯科技（深圳）有限公司 一种信息推送方法、装置、设备及计算机可读存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107171932B (zh) * 2017-04-27 2021-06-08 腾讯科技（深圳）有限公司 一种图片风格转换方法、装置及系统
CN107920132B (zh) * 2017-12-08 2018-12-21 百度在线网络技术（北京）有限公司 信息推送方法、信息展示方法和装置
RU2731335C2 (ru) * 2018-10-09 2020-09-01 Общество С Ограниченной Ответственностью "Яндекс" Способ и система для формирования рекомендаций цифрового контента
CN110969479B (zh) * 2019-11-07 2023-07-28 连尚（新昌）网络科技有限公司 一种展示推广资源的方法、设备和计算机存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110313985A1 (en) * 2010-06-21 2011-12-22 Jianwei Dian User adjustable ranking and presentation of search results
CN108776676A (zh) * 2018-02-02 2018-11-09 腾讯科技（深圳）有限公司 信息推荐方法、装置、计算机可读介质及电子设备
CN109255070A (zh) * 2018-08-01 2019-01-22 百度在线网络技术（北京）有限公司 推荐信息处理方法、装置、计算机设备及存储介质
CN109582903A (zh) * 2018-11-29 2019-04-05 广州市百果园信息技术有限公司 一种信息展示的方法、装置、设备和存储介质
CN111651692A (zh) * 2020-06-02 2020-09-11 腾讯科技（北京）有限公司 基于人工智能的信息推荐方法、装置及电子设备
CN112398947A (zh) * 2020-11-18 2021-02-23 腾讯科技（深圳）有限公司 一种信息推送方法、装置、设备及计算机可读存储介质

Also Published As

Publication number Publication date
CN115687741A (zh) 2023-02-03

Similar Documents

Publication Publication Date Title
US12386845B2 (en) Method and system for presenting a search result in a search result card
US10650009B2 (en) Generating news headlines on online social networks
US9355168B1 (en) Topic based user profiles
WO2018090793A1 (zh) 多媒体推荐方法和装置
CN109583970A (zh) 广告投放方法、装置、计算机设备及存储介质
CN109063104B (zh) 推荐信息的刷新方法、装置、存储介质和终端设备
CN114281976B (zh) 一种模型训练方法、装置、电子设备及存储介质
CN108804532A (zh) 一种查询意图的挖掘和查询意图的识别方法、装置
US10185983B2 (en) Least-ask: conversational recommender system with minimized user interaction
CN111552884B (zh) 用于内容推荐的方法和设备
US10353951B1 (en) Search query refinement based on user image selections
CN112052297B (zh) 信息生成方法、装置、电子设备和计算机可读介质
WO2024131762A1 (zh) 一种推荐方法及相关设备
CN106503059A (zh) 展示页面推送方法及装置
CN114611023B (zh) 搜索结果的显示方法、装置、设备、介质及程序产品
JP2022530725A (ja) 情報プッシュ方法、装置、機器及び記憶媒体
WO2023001093A1 (zh) 文章推荐方法、装置、设备及存储介质
US20240152512A1 (en) Machine learning for dynamic information retrieval in a cold start setting
CN115375394B (zh) 推荐方法、推荐装置、电子设备及计算机可读存储介质
CN116975427A (zh) 确定对象对资源信息的感兴趣程度的方法、装置和计算设备
US20180090021A1 (en) Ranking courses for a presentation
CN108335008A (zh) 网络信息处理方法和装置、存储介质及电子装置
CN111191056A (zh) 一种多媒体推荐方法及装置
CN116010689A (zh) 推荐模型的训练方法、装置、电子设备及存储介质
CN116777529B (zh) 一种对象推荐的方法、装置、设备、存储介质及程序产品

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22845258

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 240524)

122 Ep: pct application non-entry in european phase

Ref document number: 22845258

Country of ref document: EP

Kind code of ref document: A1