The article recommends methods, apparatus, equipment, and storage media.
By comprehensively considering user characteristics, content features of candidate articles, and style combination features, this technology solves the problem of low article recommendation accuracy in existing technologies and achieves a more efficient article recommendation effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DOUYIN VISION CO LTD
- Filing Date
- 2021-07-22
- Publication Date
- 2026-05-26
Smart Images
Figure CN115687741B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence (AI) technology, and in particular to an article recommendation method, apparatus, device, and storage medium. Background Technology
[0002] With the continuous advancement of artificial intelligence technology, users are increasingly inclined to use intelligent and personalized article recommendation methods when browsing the web.
[0003] A crucial step in article recommendation is article ranking, which determines the display position of each candidate article on the webpage. Currently, servers rank candidate articles based on user characteristics and content features, which leads to low accuracy in article recommendation. Summary of the Invention
[0004] This application provides an article recommendation method, apparatus, device, and storage medium to improve the accuracy of article recommendations.
[0005] Firstly, an article recommendation method is provided, comprising: obtaining a user request; determining M display positions and N candidate articles on the current webpage based on the user request, where M and N are both integers greater than 1; obtaining feature information corresponding to at least one candidate article on each of the M display positions; determining articles to be recommended on each of the M display positions based on the feature information corresponding to at least one candidate article on each of the M display positions; and pushing the articles to be recommended on each of the M display positions to a terminal device; wherein, at least one candidate article on the first display position is N candidate articles, and at least one candidate article on the (i+1)th display position is a candidate article obtained after deleting the articles to be recommended on the i-th display position from at least one candidate article on the i-th display position, i = 1, 2, ..., M; for any candidate article among the at least one candidate article on the i-th display position, the feature information corresponding to the candidate article includes: user features that triggered the user request, content features of the candidate article, K style combination features of the candidate article, and style combination features of the determined articles to be recommended on the first to (i-1)th display positions, where K is an integer greater than 1.
[0006] Secondly, an article recommendation device is provided, comprising: 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 a user request; the first determination module 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 acquisition module is used to acquire feature information corresponding to at least one candidate article in each of the M display positions; the second determination module is used to determine the article to be recommended in each of the M display positions according to the feature information corresponding to at least one candidate article in each of the M display positions; and the push module is used to push the articles to be recommended in each of the M display positions to a terminal device. Articles to be recommended; among them, at least one candidate article in the first display position is N candidate articles, and at least one candidate article in the (i+1)th display position is a candidate article obtained after deleting the candidate article in the ith display position from the one to be recommended article in the ith display position, i = 1, 2, ..., M; for any candidate article in the at least one candidate article in the ith display position, the feature information corresponding to the candidate article includes: the user feature that triggered the user request, the content feature of the candidate article, K style combination features of the candidate article, and the style combination features of the determined articles to be recommended in the first to (i-1)th display positions, where K is an integer greater than 1.
[0007] Thirdly, an electronic device is provided, comprising: a processor and a memory for storing a computer program, the processor for calling and running the computer program stored in the memory, and performing the methods as described in the first aspect or its various implementations.
[0008] Fourthly, a computer-readable storage medium is provided for storing a computer program that causes a computer to perform the methods described in the first aspect or its various implementations.
[0009] Fifthly, a computer program product is provided, including computer program instructions that cause a computer to perform the methods as described in the first aspect or its various implementations.
[0010] Sixthly, a computer program is provided that causes a computer to perform the methods described in the first aspect or its various implementations.
[0011] Through the technical solution of this application embodiment, for the i-th display position, the server can sort the candidate articles based on user characteristics, content characteristics of candidate articles, K style combination characteristics of candidate articles, and style combination characteristics of the determined articles to be recommended from the 1st to the (i-1th)th display positions, that is, determine the articles to be recommended at the i-th display position, and then recommend the articles to be recommended at the i-th display position to the user. This method not only considers user characteristics and content characteristics of candidate articles, but also style combination characteristics of candidate articles and style combination characteristics of sorted articles, thereby improving the accuracy of article recommendation. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0013] Figures 1 to 8 This application provides an illustrated diagram of an article as an embodiment of the present application.
[0014] Figure 9 This is a schematic diagram of an application scenario provided by an embodiment of this application;
[0015] Figure 10 A flowchart illustrating an article recommendation method provided in one embodiment of this application;
[0016] Figure 11 A schematic diagram of the first model provided in the embodiments of this application;
[0017] Figure 12 A flowchart illustrating an article recommendation method provided in another embodiment of this application;
[0018] Figure 13 A schematic diagram of the second model provided in the embodiments of this application;
[0019] Figure 14 A schematic diagram of an article recommendation device provided in an embodiment of this application;
[0020] Figure 15 This is a schematic block diagram of the electronic device 1500 provided in the embodiments of this application. Detailed Implementation
[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0022] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.
[0023] Before introducing the technical solutions of the embodiments of this application, the relevant knowledge of the embodiments of this application will be described in detail below:
[0024] It should be understood that the style combination feature of a candidate article is a feature obtained by combining at least one style of the candidate article. For example, one style feature of a candidate article is: displaying a certain number of images, or displaying a certain number of large or small images in the candidate article. Another style feature of a candidate article is: displaying or not displaying a user avatar in the candidate article. Yet another style feature of a candidate article is: displaying or not displaying a summary in the candidate article. Based on this, Figures 1 to 8 The article illustration provided for the embodiments of this application is as follows: Figure 1 As shown, the style combination features displayed in this candidate article include: no images are displayed in the candidate article, and no user avatars or article summaries are displayed. For example... Figure 2 As shown, the style combination features displayed in this candidate article include: displaying an image, but not showing the user's avatar or article summary. Figure 3 As shown, the style combination features displayed in this candidate article include: displaying a large image, but not showing the user's avatar or article summary. Figure 4 As shown, the style combination features displayed in this candidate article include: three small images are displayed, and the user avatar and article summary are not shown. Figure 5 As shown, the style combination features displayed in this candidate article include: displaying two small images and the user's avatar, but not the article summary. Figure 6As shown, the style combination features displayed in this candidate article include: displaying a large image and the user's avatar, but not an article summary. For example... Figure 7 As shown, the style combination features displayed in this candidate article include: no images or user avatars are displayed in the candidate article; instead, an article summary is shown. Figure 8 As shown, the style combination features displayed in this candidate article include: displaying three small images and not showing the user's avatar, but instead displaying an article summary. In summary, this application embodiment does not limit the style features and style combination features of the candidate article.
[0025] It should be understood that the click-through rate (CTR) of a candidate article refers to the ratio of the number of times that candidate article is clicked to the number of times it is displayed on a webpage, i.e., clicks / views, which is a percentage. It reflects the level of attention that candidate article receives on the webpage and can be used to measure the attractiveness of the candidate article.
[0026] Currently, the server sorts candidate articles based on user characteristics and content features. After sorting the candidate articles, the server then determines the display style for each article in each display position. On one hand, this sorting method does not consider the style combination features of the candidate articles, leading to low article recommendation accuracy. On the other hand, the current article sorting method does not consider the style combination features of the already sorted articles, resulting in low article recommendation accuracy. For example, if three consecutive articles without images are displayed, and then one large image article is displayed, compared to four consecutive large image articles, the last large image article in the former is more likely to be clicked. This is because the three articles without images in the former are relatively monotonous, while the article with images is more information-rich than the articles without images, whereas the three articles in the latter show little variation in information.
[0027] To address the aforementioned technical issues, embodiments of this application will consider the style combination features of candidate articles and the style combination features of previously ranked articles, thereby improving the accuracy of article recommendation.
[0028] The technical solutions of this application embodiment can be applied to the following scenarios, but are not limited thereto: Figure 9 This is a schematic diagram of an application scenario provided in an embodiment of this application, such as... Figure 9 As shown, the terminal device 910 can communicate with the server 920. The terminal device 910 can obtain and send a user request to the server 920. The user request is used to request browsing a certain webpage. After obtaining the user request, the server 920, which is the webpage browsing server corresponding to the webpage, determines the display position for each candidate article according to the technical solution provided in the embodiments of this application, and finally pushes the articles at each display position to the terminal device 910.
[0029] In some possible implementations, the terminal device may be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, wearable device, etc., but is not limited thereto, and the embodiments of this application do not impose any restrictions on this.
[0030] In some implementations, the server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.
[0031] In some possible implementations, communication between the terminal device and the server, and between servers, can be conducted using cellular mobile networks or Wireless Fidelity (WiFi), etc., and this application embodiment does not limit this.
[0032] The technical solutions of the embodiments of this application will be described in detail below:
[0033] Figure 10 This is a flowchart illustrating an article recommendation method provided in an embodiment of this application. The method can be... Figure 9 The server in this application executes the method recommended in this article, but is not limited to this. The following description uses the server executing the method recommended in this article as an example to illustrate the technical solution of this application's embodiments. Figure 10 As shown, the method includes the following steps:
[0034] S1010: Obtain user request.
[0035] S1020: Determine the M display positions and N candidate articles on the current webpage based on the user's request, where M and N are both integers greater than 1.
[0036] S1030: Obtain the feature information corresponding to at least one candidate article in each of the M display positions. For any candidate article in the at least one candidate article in the i-th display position, the feature information corresponding to the candidate article includes: the user feature that triggered the user request, the content feature of the candidate article, the K style combination features of the candidate article, and the style combination features of the determined articles to be recommended in the 1st to i-1th display positions, i = 1, 2...M, and K is an integer greater than 1.
[0037] S1040: Determine the articles to be recommended for each of the M display positions based on the feature information corresponding to at least one candidate article in each of the M display positions.
[0038] S1050: Pushes articles to be recommended to M display positions to the terminal device.
[0039] It should be understood that the user request is used to request browsing a webpage. The user request may be triggered by the user clicking the webpage link on the terminal device, or by the user entering the webpage URL on the terminal device and pressing the Enter key. This application embodiment does not limit how the user request is triggered or obtained.
[0040] It should be understood that the above M display locations are the locations on the current webpage where candidate articles can be displayed, and the current webpage is the webpage that the user requests to browse.
[0041] It should be understood that N candidate articles can be N candidate articles on the same topic, for example: N candidate articles are all about a certain person, or N candidate articles can be N candidate articles on different topics, for example: when a user clicks a link in a browser, the webpage displayed on the terminal device includes articles on various popular topics.
[0042] In some implementations, M is less than or equal to N. That is, the number of candidate articles is greater than or equal to the display position of the candidate articles on the current webpage. Alternatively, M is greater than N. That is, the number of candidate articles is less than the display position of the candidate articles on the current webpage. In summary, the embodiments of this application do not limit the relationship between M and N.
[0043] It should be understood that at least one candidate article in the first display position is one of N candidate articles, and at least one candidate article in the (i+1)th display position is a candidate article obtained after deleting the recommended article in the ith display position from at least one candidate article in the ith display position, where i = 1, 2, ..., M. Specifically, the server can sort the above M display positions so that each of the M display positions has a unique index. For the first display position, the candidate articles in this position are the above N candidate articles. Based on this, the server can first determine the recommended article in the first display position. Furthermore, for the second display position, the candidate articles in this position are the candidate articles obtained after deleting the recommended article in the first display position from the above N candidate articles. Based on this, the server can determine the recommended article in the second display position, and so on. The server can determine the candidate articles in each display position and determine the recommended article in each display position from the candidate articles in each display position.
[0044] It should be understood that the server can sort the M display positions using any of the following feasible methods, but is not limited to these:
[0045] One possible approach is for the server to sort the M display positions according to their priority. The higher the priority of a display position, the higher it appears in the ranking, and vice versa. In other words, a higher priority display position has a smaller index, and vice versa. The priority of the M display positions is related to their attractiveness to the user. For example, the central display position on the current webpage has the highest attractiveness and therefore the highest priority, while the top-left corner display position has the next highest attractiveness and therefore the lowest priority, and so on.
[0046] In the second possible implementation, the server can sort the M display positions according to their priorities. The higher the priority of a display position, the earlier it appears in the ranking, and vice versa. In other words, a higher priority display position has a smaller index, and vice versa. The server can determine the priority of the M display positions from left to right and from top to bottom. For example, the top-left corner display position on the current webpage has the highest priority, the next highest priority position is next, and so on.
[0047] The third possible approach is for the server to sort the M display positions according to their attractiveness to users. For example, the most central display position on the current webpage has the highest attractiveness, so it is ranked first; the top left corner display position has the second highest attractiveness, so it is ranked second, and so on.
[0048] The fourth possible approach is for the server to sort the M display positions from left to right and from top to bottom. For example, the top left corner of the current webpage is ranked first, the next display position after the top left corner is ranked second, and so on.
[0049] It should be understood that any two adjacent display positions among the above M display positions can be adjacent or not adjacent on the current webpage, and this application embodiment does not impose any restrictions on this.
[0050] As described above, for any candidate article among at least one candidate article in the i-th display position, the feature information corresponding to the candidate article includes: user features that triggered the user request, content features of the candidate article, K style combination features of the candidate article, and style combination features of the determined articles to be recommended in the 1st to i-1th display positions, i = 1, 2, ..., M, where K is an integer greater than 1.
[0051] In some possible implementations, the user characteristics that trigger a user request include at least one of the following, but are not limited to: the user's age, gender, preferences, education level, etc.
[0052] In some possible implementations, the content characteristics of candidate articles include at least one of the following, but are not limited to: the category to which the candidate article belongs, the keywords of the candidate article, the author, and the click-through rate of the candidate article in its category.
[0053] In some possible implementations, candidate articles can be categorized according to their mode of expression, including, but not limited to, narrative, expository, and argumentative texts. Alternatively, candidate articles can be categorized according to literary works, including, but not limited to, poetry, novels, essays (lyrical and narrative), and plays. Candidate articles can also be categorized according to their mode of expression, rhetorical devices, and stylistic features. In short, the embodiments of this application do not limit the criteria for categorizing candidate articles.
[0054] In some possible implementations, the K style combination features of the candidate article can be all style combination features of the candidate article, or only some style combination features among all style combination features of the candidate article. This application embodiment does not limit this.
[0055] For example, assuming K=2, for a candidate article item, the server determines two style combination features of the candidate article item as follows: the candidate article displays a large image and the user's avatar, but does not display an article summary; and the candidate article displays three small images, but does not display the user's avatar, but displays an article summary.
[0056] It should be understood that the style combination features of the determined articles to be recommended at the first to the (i-1)th display positions include: the style combination features of the determined articles to be recommended at the first display position, the style combination features of the determined articles to be recommended at the second display position, and so on, up to the (i-1)th display position. For example, if i = 3, meaning the server is currently determining the article to be recommended at the third display position, then the style combination features of the determined articles to be recommended at the first to the second display positions include: the style combination features of the determined articles to be recommended at the first display position and the style combination features of the determined articles to be recommended at the second display position. For example, the style combination feature of the determined article to be recommended at the first display position is: the article to be recommended displays a large image and the user's avatar, but not an article summary. The style combination feature of the determined article to be recommended at the second display position is: the article to be recommended displays two small images, the user's avatar, and an article summary.
[0057] It should be understood that the server can determine the articles to be recommended for each of the M display positions in any of the following feasible ways, but is not limited to these:
[0058] One possible approach is as follows: For any candidate article among at least one candidate article in the i-th display position, and any style combination feature among K style combination features of the candidate article, the server inputs the user features, the content features of the candidate article, the style combination features, and the style combination features of the determined articles to be recommended in the 1st to (i-1th)th display positions into the first model to obtain the click-through rate (CTR) of the candidate article under the style combination features. The server determines the maximum CTR 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 in the i-th display position as the article to be recommended in the i-th display position.
[0059] Option 2: For any candidate article among at least one candidate article in the i-th display position, and any style combination feature among K style combination features of the candidate article, the server inputs the user features, the content features of the candidate article, the style combination features, and the style combination features of the determined articles to be recommended in the 1st to (i-1th)th display positions into the first model to obtain the click-through rate of the candidate article under the style combination features. The server determines the score of the candidate article as the average click-through rate of the candidate article under the K style combination features. The server determines the candidate article with the highest score in the i-th display position as the article to be recommended in the i-th display position.
[0060] For example, Figure 11 A schematic diagram of the first model provided for an embodiment of this application, as shown below. Figure 11 As shown, the input to the first model consists of user features, content features of candidate articles, a certain style combination feature, and the style combination features of the identified articles to be recommended at the first to (i-1)th display positions. The output of the first model is the click-through rate of the candidate articles under that certain style combination feature.
[0061] The following explanation addresses the first feasible method:
[0062] Suppose that in a single user request, the server determines there are M display positions. The server can sequentially select one article from the candidate articles to fill the corresponding position from the 1st to the Mth display position. Let's say we need to determine the article to be recommended for the i-th display position, and there are C candidate articles for this position. Let the candidate set of these C candidate articles be {item1, item2, ..., item...} C}, let score(item j ) as candidate article item j The score. During the sorting process, the server can select the candidate article with the highest score and place it in the i-th display position. The candidate article item... j The score can be calculated according to the following formula (1):
[0063]
[0064] Model multi ( ) indicates the first model.
[0065] `max{}` means taking the maximum value.
[0066] F user This represents user characteristics. The corresponding user characteristics are the same for all candidate articles. In addition, the corresponding user characteristics are also the same for the same candidate article displayed in different positions.
[0067] Indicates candidate article item j The content features are as follows: different candidate articles have different content features, but the same candidate article in different display positions has the same content features.
[0068] Indicates candidate article item j A combination of style features, for example: (1,j1) represents selecting the j1th style in the first style dimension, (2,j2) represents selecting the j2th style in the second style dimension, ... (N,j N ) indicates selecting the j-th style in the N-th style dimension. NThere are several styles. Among them, candidate article items... j It includes K style feature combinations.
[0069] This represents the style combination features of the articles to be recommended that have been determined for the first display position, the second display position, ..., the (i-1)th display position.
[0070] The second feasible method will be explained as follows:
[0071] Suppose that in a single user request, the server determines there are M display positions. The server can sequentially select one article from the candidate articles to fill the corresponding position from the 1st to the Mth display position. Let's say we need to determine the article to be recommended for the i-th display position, and there are C candidate articles for this position. Let the candidate set of these C candidate articles be {item1, item2, ..., item...} C}, let score(item j ) as candidate article item j The score. During the sorting process, the server can select the candidate article with the highest score and place it in the i-th display position. The candidate article item... j The score can be calculated according to the following formula (2):
[0072]
[0073] Model multi ( ) indicates the first model.
[0074] The mean{} represents taking the average value.
[0075] F user This represents user characteristics. The corresponding user characteristics are the same for all candidate articles. In addition, the corresponding user characteristics are also the same for the same candidate article displayed in different positions.
[0076] Indicates candidate article item j The content features are as follows: different candidate articles have different content features, but the same candidate article in different display positions has the same content features.
[0077] Indicates candidate article item j A combination of style features, for example: (1,j1) represents selecting the j1th style in the first style dimension, (2,j2) represents selecting the j2th style in the second style dimension, ... (N,j N ) indicates selecting the j-th style in the N-th style dimension.N There are several styles. Among them, candidate article items... j It includes K style feature combinations.
[0078] This represents the style combination features of the articles to be recommended that have been determined for the first display position, the second display position, ..., the (i-1)th display position.
[0079] Furthermore, after the server determines the articles to be recommended in each of the M display positions, it can push the corresponding articles to be recommended in each of the M display positions to the terminal device, so that the terminal device can display the corresponding articles to be recommended in the aforementioned M display positions on the current webpage.
[0080] In some feasible implementations, for any one of the M display positions, the article to be recommended in that position can be displayed using a target style combination feature. This target style combination feature is the style combination feature with the highest click-through rate when the server determines the score of the article to be recommended. Of course, this target style combination feature is also the style combination feature with the second highest click-through rate when the server determines the score of the article to be recommended. It should be noted that the method of determining this target style combination feature is not limited to this.
[0081] It should be understood that, Figure 10 The article recommendation method shown is described from the perspective of the executed code. Specifically, Figure 12 A flowchart illustrating an article recommendation method provided in another embodiment of this application is shown below. Figure 12 As shown, the method includes the following steps:
[0082] S1: Obtain user request;
[0083] S2: Determine the M display positions and N candidate articles on the current webpage based on the user request, where M and N are both integers greater than 1. Let i = 1, where i represents the index of the display position;
[0084] S3: Obtain the feature information corresponding to at least one candidate article at the i-th display position;
[0085] S4: Determine the article to be recommended in the i-th display position based on the feature information corresponding to at least one candidate article in the i-th display position;
[0086] S5: Let i = i + 1, and determine whether i is less than or equal to M. If i is less than or equal to M, continue to execute S3; otherwise, execute S6.
[0087] S6: Push the recommended articles to M display positions to the terminal device.
[0088] Among them, at least one candidate article in the first display position is N candidate articles, and at least one candidate article in the (i+1)th display position is a candidate article obtained after deleting the recommended article in the ith display position from at least one candidate article in the ith display position; for any candidate article in the at least one candidate article in the ith display position, the feature information corresponding to the candidate article includes: the user feature that triggered the user request, the content feature of the candidate article, the K style combination features of the candidate article, and the style combination features of the determined recommended articles in the first to (i-1)th display positions, where K is an integer greater than 1.
[0089] In summary, in this embodiment, for the i-th display position, the server can sort the candidate articles based on user characteristics, content characteristics of candidate articles, K style combination characteristics of candidate articles, and style combination characteristics of the determined articles to be recommended from the first to the (i-1)-th display positions, thus determining the articles to be recommended at the i-th display position, and then recommending the articles to be recommended at the i-th display position to the user. This approach not only considers user characteristics and content characteristics of candidate articles, but also style combination characteristics of candidate articles and style combination characteristics of sorted articles, thereby improving the accuracy of article recommendation.
[0090] As described above, for any candidate article, the K style combination features of the candidate article can be all style combination features of the candidate article, or it can be a partial style combination feature among all style combination features of the candidate article. For the latter, that is, for any candidate article among at least one candidate article at the i-th display position, the server can select K style combination features from all style combination features of the candidate article.
[0091] It should be understood that the server can select K style combination features from all style combination features of candidate articles in the following ways, but is not limited to:
[0092] One possible approach: For any style combination feature among all style combination features of the candidate article, the server determines the score of the style combination feature, and selects the top K style combination features with the highest scores from all style combination features of the candidate article as the K style combination features.
[0093] Option 2: For any style combination feature among all style combination features of the candidate article, the server determines the score of the style combination feature. The server selects the top P style combination features with the highest scores from all style combination features of the candidate article, where P is an integer greater than K. Further, the server selects K style combination features from these P style combination features to form the K style combination features. The server can randomly select K style combination features from these P style combination features or select K style combination features according to a certain preset rule. This embodiment does not limit the preset rule.
[0094] It should be understood that the server can determine the score of the style combination feature in any of the following ways, but is not limited to:
[0095] One possible approach is for the server to input user features, content features of candidate articles, and style features into the second model for any style feature in the style combination features to obtain the click-through rate of the candidate article under that style feature. Furthermore, the server sums the click-through rates of the candidate article under each style feature in the style combination features to obtain the score of the style combination features.
[0096] Option 2: The server can input user features, content features of candidate articles, and style features into the second model for any style feature in the style combination features to obtain the click-through rate of the candidate article under the style feature. Furthermore, the server calculates the average of the click-through rates of the candidate article under each style feature in the style combination features to obtain the score of the style combination features.
[0097] It should be understood that any one of the style features in the above style combination features is a specific style feature of a certain dimension of the style combination features. For example, the style feature could be: displaying three small images in the candidate article, or not displaying the user avatar in the candidate article, or not displaying the article summary in the candidate article, etc.
[0098] For example, Figure 13 A schematic diagram of the second model provided in the embodiments of this application, as shown below. Figure 13 As shown, the input to the second model is user features, content features of candidate articles, and a certain style feature. The output of the second model is the candidate article item. j In the i-th style feature, the j-th i Click-through rate under each style feature.
[0099] For the two feasible methods mentioned above, the server can determine the click-through rate of the candidate article under this style feature using the following formula (2):
[0100]
[0101] This indicates the click-through rate of the candidate article under this style feature.
[0102] This represents the first model.
[0103] F user This represents user characteristics. The corresponding user characteristics are the same for all candidate articles. In addition, the corresponding user characteristics are also the same for the same candidate article displayed in different positions.
[0104] Indicates candidate article item j The content features are as follows: different candidate articles have different content features, but the same candidate article in different display positions has the same content features.
[0105] Regarding the first feasible method mentioned above, the server can determine the score of a candidate article in a certain style combination feature using the following formula (3), but is not limited to this:
[0106]
[0107] Regarding the second feasible method mentioned above, the server can determine the score of a candidate article in a certain style combination feature using the following formula (4), but is not limited to this:
[0108]
[0109] The mean{} represents taking the average value.
[0110] It should be understood that, assuming there are currently N-dimensional style features, where the first-dimensional style feature includes n(1) style features, the second-dimensional style feature includes n(2) style features, and so on, the Nth-dimensional style feature includes n(N) style features, then if the server calculates the candidate article item according to the above formula (1)... j The score indicates that the server's computational load is... However, if the server selects K style combination features according to formulas (3) and (4), although the server's computational load is also... However, the second model has a much lower computational complexity than the first model. Therefore, by using a pruning operation on the server, that is, selecting only K style combination features, the computational complexity can be reduced, thereby improving the efficiency of article recommendation.
[0111] Furthermore, in existing technologies, the style combination feature selection problem is viewed as a multi-armed lottery problem, seeking the style combination feature with the highest click-through rate as quickly as possible. However, the style combination features obtained in this way lack personalization; that is, the style combination features of the same candidate article are the same for all users. However, when adopting the technical solution of this application, the server inputs user characteristics, candidate article content characteristics, style combination features, and the determined style combination features of the articles to be recommended from the first display position to the (i-1)th display position into the first model to obtain the click-through rate of the candidate article under the stated style combination features. The maximum click-through rate of the candidate article under K style combination features can be used to determine the style combination feature corresponding to the maximum click-through rate as the style combination feature to be displayed for that candidate article. In other words, from the perspective of style combination feature selection alone, the final selected style combination feature takes user characteristics into account. The style combination features obtained in this way are personalized; that is, the style combination features of the same candidate article can be different for different users, thereby improving the user experience.
[0112] It should be understood that, in order to improve the accuracy of article recommendations, the server can train the first model, specifically through the following possible methods, but not limited to:
[0113] In some implementation methods, the server obtains multiple first articles, sorts the multiple first articles according to their respective content features, determines the style combination features of any first article in the sorted first articles, and trains a first model based on the multiple first articles with the determined style combination features.
[0114] Specifically, for each style feature of the first article, a random selection is made from the candidate styles corresponding to the style feature to form the style combination feature of the first article. Alternatively, for each style feature of the first article, a selection is made from the candidate styles corresponding to the style feature according to a preset rule to form the style combination feature of the first article. This application embodiment does not restrict how the selection is made from the candidate styles corresponding to the style features.
[0115] For example, for any candidate article, suppose there are N-dimensional style features, and the i-th style feature includes n(i) style features. For the i-th style feature, the server can randomly select a style feature from the n(i) style features.
[0116] It should be understood that, in order to better train the first model, the server expects all style combination features to appear as much as possible. Therefore, when training the first model, the server can randomly select from the candidate styles corresponding to the style features to form a variety of style combination features, including style combination features that may appear very rarely.
[0117] Figure 14 This is a schematic diagram of an article recommendation device provided in an embodiment of this application. The device may be... Figure 9 Servers in, but not limited to, such as Figure 14 As shown, the device includes:
[0118] The first acquisition module 1401 is used to acquire user requests.
[0119] The first determining module 1402 is used to determine M display positions and N candidate articles on the current webpage according to the user request, where M and N are both integers greater than 1.
[0120] The second acquisition module 1403 is used to acquire feature information corresponding to at least one candidate article in each of the M display positions.
[0121] The second determining module 1404 is used to determine the article to be recommended in each of the M display positions based on the feature information corresponding to at least one candidate article in each of the M display positions.
[0122] The push module 1405 is used to push articles to be recommended to M display positions to the terminal device.
[0123] Among them, at least one candidate article in the first display position is N candidate articles, and at least one candidate article in the (i+1)th display position is a candidate article obtained after deleting the article to be recommended in the i-th display position from at least one candidate article in the i-th display position, i = 1, 2, ..., M.
[0124] For any candidate article among at least one candidate article in the i-th display position, the feature information corresponding to the candidate article includes: user features that triggered the user request, content features of the candidate article, K style combination features of the candidate article, and style combination features of the determined articles to be recommended in the first to (i-1)-th display positions, where K is an integer greater than 1.
[0125] In some implementations, the second determining module 1404 is specifically used to: for any candidate article among at least one candidate article at the i-th display position, and any style combination feature among K style combination features of the candidate article, input the user features, the content features of the candidate article, the style combination feature, and the style combination features of the determined articles to be recommended at the 1st to (i-1th)th display positions into the first model to obtain the click-through rate of the candidate article 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 at the i-th display position is determined as the article to be recommended at the i-th display position.
[0126] In some possible implementations, the article recommendation device also includes:
[0127] The third acquisition module 1406 is used to acquire multiple first articles.
[0128] The sorting module 1407 is used to sort multiple first articles according to their respective content characteristics.
[0129] The third determining module 1408 is used to determine the style combination features of any first article in the sorted first articles.
[0130] Training module 1409 is used to train the first model based on multiple first articles with determined style combination features.
[0131] In some possible implementations, the third determining module 1408 is specifically used to: for each style feature of the first article, randomly select from the candidate styles corresponding to the style feature, and form a style combination feature of the first article.
[0132] In some implementations, the article recommendation device further includes: a selection module 1410, for selecting K style combination features from all style combination features of any candidate article among at least one candidate article at the i-th display position.
[0133] In some implementations, the selection module 1410 is specifically used to: determine the score of any style combination feature among all style combination features of the candidate article; and select the top K style combination features with the highest scores from all style combination features of the candidate article as the K style combination features.
[0134] In some implementations, the selection module 1410 is specifically used to: for any style feature in the style combination features, input the user features, the content features of the candidate article, and the style features into the second model to obtain the click-through rate of the candidate article under the style features. The click-through rates of the candidate article under each style feature in the style combination features are summed to obtain the score of the style combination features.
[0135] In some implementations, any combination of style features of the candidate article includes at least two of the following: the number of images included in the candidate article, whether a user avatar is displayed, and whether a summary of the candidate article is displayed.
[0136] It should be understood that the device embodiments and method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, further details will not be provided here. Specifically, Figure 14 The apparatus shown can execute the above-described method embodiments, and the aforementioned and other operations and / or functions of each module in the apparatus are respectively for implementing the corresponding processes in each method. For the sake of brevity, they will not be described in detail here.
[0137] The apparatus of this application embodiment has been described above from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that this functional module can be implemented in hardware, in software instructions, or in a combination of hardware and software modules. Specifically, the steps of the method embodiments in this application can be completed by integrated logic circuits in the processor's hardware and / or by software instructions. The steps of the method disclosed in this application embodiment can be directly embodied as being executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. Optionally, the software module can reside in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps in the above method embodiments.
[0138] Figure 15 This is a schematic block diagram of the electronic device 1500 provided in the embodiments of this application.
[0139] like Figure 15 As shown, the electronic device 1500 may include:
[0140] The system includes a memory 1510 and a processor 1520. The memory 1510 stores computer programs and transfers the program code to the processor 1520. In other words, the processor 1520 can retrieve and run the computer program from the memory 1510 to implement the methods described in the embodiments of this application.
[0141] For example, the processor 1520 can be used to execute the above-described method embodiments according to instructions in the computer program.
[0142] In some embodiments of this application, the processor 1520 may include, but is not limited to:
[0143] General-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
[0144] In some embodiments of this application, the memory 1510 includes, but is not limited to:
[0145] Volatile memory and / or non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).
[0146] In some embodiments of this application, the computer program may be divided into one or N modules, which are stored in the memory 1510 and executed by the processor 1520 to complete the method provided in this application. The one or N modules may be a series of computer program instruction segments capable of performing a specific function, and these instruction segments describe the execution process of the computer program in the electronic device.
[0147] like Figure 15 As shown, the electronic device may also include:
[0148] Transceiver 1530, which can be connected to processor 1520 or memory 1510.
[0149] The processor 1520 can control the transceiver 1530 to communicate with other devices; specifically, it can send information or data to other devices or receive information or data sent by other devices. The transceiver 1530 may include a transmitter and a receiver. The transceiver 1530 may further include antennas, which may be one or N.
[0150] It should be understood that the various components in the traffic flow control device are connected through a bus system, which includes a data bus, a power bus, a control bus, and a status signal bus.
[0151] This application also provides a computer storage medium storing a computer program thereon, which, when executed by a computer, enables the computer to perform the methods of the above-described method embodiments. Alternatively, this application also provides a computer program product containing instructions that, when executed by a computer, cause the computer to perform the methods of the above-described method embodiments.
[0152] When implemented using software, it can be implemented entirely or partially as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)).
[0153] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the embodiments of this application.
[0154] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, N modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or modules may be electrical, mechanical, or other forms.
[0155] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across N network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. For example, the functional modules in the various embodiments of this application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
[0156] The above description is merely a specific implementation of the embodiments of this application, but the protection scope of the embodiments of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the embodiments of this application should be included within the protection scope of the embodiments of this application. Therefore, the protection scope of the embodiments of this application should be determined by the protection scope of the claims.
Claims
1. An article recommendation method characterized by comprising: include: Get user request; Based on the user request, determine M display positions and N candidate articles on the current webpage, where M and N are both integers greater than 1; Obtain feature information corresponding to at least one candidate article at each of the M display locations; The article to be recommended for each of the M display positions is determined based on the feature information corresponding to at least one candidate article in each of the M display positions; Push the recommended articles to the terminal devices in the M display locations; Among them, at least one candidate article in the first display position is the N candidate articles, and at least one candidate article in the (i+1)th display position is the candidate article obtained after deleting the recommended article in the i-th display position from at least one candidate article in the i-th display position, i=1,2……M-1; For any candidate article among at least one candidate article at the i-th display position, the feature information corresponding to the candidate article includes: user features that triggered the user request, content features of the candidate article, K style combination features of the candidate article, and style combination features of the determined articles to be recommended at the 1st to (i-1th)th display positions, where K is an integer greater than 1; The step of determining the article to be recommended for each of the M display positions based on the feature information corresponding to at least one candidate article for each of the M display positions includes: For any candidate article among at least one candidate article in the i-th display position, and any style combination feature among K style combination features of the candidate article, the user feature, the content feature of the candidate article, the any style combination feature, and the style combination features of the determined articles to be recommended in the 1st to (i-1th)th display positions are input into the first model to obtain the click-through rate of the candidate article 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 in the i-th display position is determined as the article to be recommended in the i-th display position.
2. The method of claim 1, wherein, Also includes: Obtain multiple first articles; The multiple first articles are sorted according to their respective content characteristics; For any first article in the sorted first articles, determine the style combination features of the first article; The first model is trained based on the multiple first articles whose style combination features have been determined.
3. The method according to claim 2, characterized in that, Determining the style combination features of the first article includes: For each style feature of the first article, a random selection is made from the candidate styles corresponding to the style feature, and these styles are combined to form the style combination features of the first article.
4. The method according to any one of claims 1-3, characterized in that, Also includes: For any candidate article among at least one candidate article at the i-th display position, select the K style combination features from all style combination features of the candidate article.
5. The method according to claim 4, characterized in that, The step of selecting the K style combination features from all style combination features of the candidate articles includes: For any style combination feature among all style combination features of the candidate article, determine the score of the style combination feature; Select the K style combination features with the highest scores from all style combination features of the candidate articles, and use them as the K style combination features.
6. The method according to claim 5, characterized in that, The determination of the score for the style combination feature includes: For any style feature in the style combination features, the user feature, the content feature of the candidate article, and the style feature are input into the second model to obtain the click-through rate of the candidate article under the style feature; The click-through rates of the candidate articles under each style feature in the style combination feature are summed to obtain the score of the style combination feature.
7. The method according to any one of claims 1-3, characterized in that, The candidate article's style combination features include at least two of the following: the number of images included in the candidate article, whether the user's avatar is displayed, and whether the candidate article's summary is displayed.
8. An article recommendation device, characterized in that, include: The first acquisition module is used to acquire user requests; The first determining module 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 acquisition module is used to acquire feature information corresponding to at least one candidate article at each of the M display positions; The second determining module is used to determine the article to be recommended for each of the M display positions based on the feature information corresponding to at least one candidate article for each of the M display positions; The push module is used to push the articles to be recommended in the M display positions to the terminal device; Among them, at least one candidate article in the first display position is the N candidate articles, and at least one candidate article in the (i+1)th display position is the candidate article obtained after deleting the recommended article in the i-th display position from at least one candidate article in the i-th display position, i=1,2……M-1; For any candidate article among at least one candidate article at the i-th display position, the feature information corresponding to the candidate article includes: user features that triggered the user request, content features of the candidate article, K style combination features of the candidate article, and style combination features of the determined articles to be recommended at the 1st to (i-1th)th display positions, where K is an integer greater than 1; The second determining module is further configured to, for any candidate article among at least one candidate article in the i-th display position, and any style combination feature among K style combination features of the candidate article, input the user feature, the content feature of the candidate article, the any style combination feature, and the style combination features of the determined articles to be recommended in the first display position to obtain the click-through rate of the candidate article under the style combination feature; determine the maximum click-through rate of the candidate article under the K style combination features as the score of the candidate article; and determine the candidate article with the highest score in the i-th display position as the article to be recommended in the i-th display position.
9. An electronic device, characterized in that, include: A processor and a memory, the memory being used to store a computer program, the processor being used to invoke and run the computer program stored in the memory to perform the method of any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, Used to store a computer program that causes a computer to perform the method as described in any one of claims 1 to 7.