Information push method, device, equipment and storage medium

Abstract

The present application provides an information push method, device, equipment and storage medium, the method comprising: obtaining a relevant information set of a target object; determining a recall information set corresponding to the target object based on the relevant information set; obtaining multiple recall attribute features corresponding to each recall information; fusing the multiple recall attribute features corresponding to each recall information to obtain a fused attribute feature corresponding to the recall information; inputting the obtained fused attribute feature corresponding to each recall information into a click prediction model to perform information click prediction processing to obtain the result of the target object clicking on each recall information; determining a candidate information set corresponding to the target object based on the result of the target object clicking on each recall information; and pushing the candidate information to the target object. The present application improves the accuracy of information push and relieves system pressure at the same time.

Description

Technical Field

[0001] The present application belongs to the field of Internet technology, and specifically relates to an information push method, device, equipment and storage medium. Background Art

[0002] With the development of Internet technology, millions of pieces of information can be pushed to users through the Internet. For example, when a user reads a piece of news through a news software, information similar to the content of the news or information that the user is interested in can be recommended to the user.

[0003] Existing information recommendation methods are all information-centric, that is, information recalls users. For example, every time a news editor edits a piece of news, he will match the news profile with all user profiles one by one, and use the target user whose user profile matches the news profile as the target user of the news, and then push the news to the target user. However, the method of information recalling articles cannot control the user's recall information at a macro level. It can only decide whether to recall based on certain conditions (such as setting a threshold) at the time of recall. At the same time, because information is streaming, the method of information recalling users will cause congestion when the information pool increases, and even cause systemic failures in severe cases. In addition, once the information, such as the news profile, is biased, the target user determined by the news profile will also be biased, which will make it impossible to accurately push news. Summary of the invention

[0004] In order to achieve user-centricity, push information of interest to users, improve the accuracy of information push, and relieve system pressure, the present application proposes an information push method, device, equipment and storage medium.

[0005] On the one hand, the present application proposes an information push method, the method comprising:

[0006] Obtain the target user's portrait information set;

[0007] Based on the portrait information set, determining a recall information set corresponding to the target user;

[0008] Obtain multiple recall attribute features corresponding to each recall information;

[0009] Performing fusion processing on the multiple recall attribute features corresponding to each recall information to obtain a fusion attribute feature corresponding to the recall information;

[0010] Inputting the fused attribute features corresponding to each recalled information into the click prediction model to perform information click prediction processing, and obtaining the result of the target user clicking on each recalled information;

[0011] Determine a candidate information set corresponding to the target user based on the result of the target user clicking each recalled information;

[0012] The candidate information is pushed to the target user.

[0013] On the other hand, the present application proposes an information push device, the device comprising:

[0014] The first acquisition module is used to acquire a portrait information set of a target user;

[0015] A first determination module, configured to determine a recall information set corresponding to the target user based on the portrait information set;

[0016] A second acquisition module is used to acquire multiple recall attribute features corresponding to each recall information;

[0017] A fusion module, used for fusing the multiple recall attribute features corresponding to each recall information to obtain a fused attribute feature corresponding to the recall information;

[0018] A prediction module is used to input the fusion attribute features corresponding to each recalled information into a click prediction model to perform information click prediction processing, and obtain the result of the target user clicking on each recalled information;

[0019] A second determination module, configured to determine a candidate information set corresponding to the target user based on the result of the target user clicking each recalled information;

[0020] The push module is used to push the candidate information to the target user.

[0021] On the other hand, the present application proposes a device, comprising: a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the information push method as described above.

[0022] On the other hand, the present application proposes a computer-readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by a processor to implement the information push method as described above.

[0023] The present application proposes an information push method, device, equipment and storage medium. According to the target user's portrait information set, a recall information set corresponding to the target user is obtained through different recall methods, that is, through multiple portrait sources and multiple recall reasons. Then, multiple recall attribute features corresponding to each recall information are obtained, and the multiple recall attribute features corresponding to each recall information are fused to obtain the fused attribute features corresponding to the recall information. Then, the fused attribute features corresponding to each recall information are input into a click prediction model to perform information click prediction processing to obtain the result of the target user clicking on each recall information. Finally, based on the result of the target user clicking on each recall information, the corresponding recall information of the target user is determined. Candidate information set, and push the candidate information to the target user, to achieve user-centric information recall, can flexibly process and macro-control the user's recalled information, in line with the product positioning of user personalized reading, at the same time, through multiple portrait sources and multiple recall attribute features, can deeply understand and meet the user's reading needs, recommend information of interest to users, enhance personalization, further, by fusing multiple recall attribute features, based on the fused attribute features, use the prediction model to predict the result of each recalled information being clicked, the prediction result is more accurate and reliable, in addition, user-centric recall information is also very friendly to the expansion of the article pool, and will not bring pressure to the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 It is a schematic diagram of an implementation environment of an information push method provided in an embodiment of the present application.

[0026] Figure 2 It is a flow chart of an information push method provided in an embodiment of the present application.

[0027] Figure 3 It is a flow chart of an embodiment of the present application for fusing multiple recall attribute features corresponding to each recall information to obtain a fused attribute feature corresponding to the recall information.

[0028] Figure 4 It is a flowchart diagram of determining a candidate information set corresponding to the target user based on the result of the target user clicking on each recalled information provided by an embodiment of the present application.

[0029] Figure 5 This is a schematic diagram of a terminal display for pushing sports-related news to users provided by an embodiment of the present application.

[0030] Figure 6 This is an application scenario diagram of an information push method provided in an embodiment of the present application.

[0031] Figure 7 It is a structural diagram of an information push device provided in an embodiment of the present application.

[0032] Figure 8 It is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0034] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0035] See also Figure 1 , Figure 1 Schematic diagram of an implementation environment of an information push method provided in an embodiment of the present application, such as Figure 1 As shown, the implementation environment may include at least a server 01 and a client 02, wherein the server 01 and the client 02 are connected via a wired or wireless network to implement data transmission between the server 01 and the client 02. For example, the transmitted data includes recommended candidate information.

[0036] In the embodiment of this specification, the server 01 may include an independently operated server, or a distributed server, or a server cluster composed of multiple servers. The server 01 may include a network communication unit, a processor, a memory, etc. Specifically, the server 01 may provide background services for the client 02, for example, the background services include multimedia recommendation services.

[0037] In the embodiments of this specification, the client 02 may include physical devices such as smart phones, desktop computers, tablet computers, laptops, digital assistants, smart wearable devices, car-mounted devices, speakers, televisions, robots, etc., and may also include software running in physical devices, such as virtual machines.

[0038] In the embodiment of this specification, through the interaction between the server 01 and the client 02, the client 02 will send an information recommendation request to the server 01, and then the server 01 will provide a multimedia recommendation service and push the recommended candidate information to the client 02 to display the recommended candidate information to the user.

[0039] The following introduces an information push method of this application. Figure 2 It is a flowchart of an information push method provided by an embodiment of the present application. This specification provides method operation steps as described in the embodiment or flowchart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or server product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiment or the accompanying drawings (for example, in a parallel processor or multi-threaded processing environment). Specifically, Figure 2 As shown, the method may include:

[0040] S201. Obtain a portrait information set of the target user.

[0041] The information push method in the embodiments of this specification is based on the user's automatic push. Push means pushing. The so-called automatic push refers to recalling the information that the user is interested in from the interest queue information pool in combination with the user portrait, and selecting the appropriate information from it and finally sending it to the user.

[0042] In the embodiments of this specification, the target user is the user to whom information needs to be pushed. The target user can be determined according to actual needs. For example, the target user may include all registered users of the application. The user whose background record has a user identifier is the registered user. The user identifier can be used to uniquely identify the user, such as an identity identification number (ID). Taking a news application as an example, assuming that it has a total of N1 registered users, these N1 registered users can be used as target users. For another example, the target user may also include some users who meet the user screening conditions among all registered users of the application. The user screening conditions can be set according to actual needs. For example, the user screening conditions may include that the user status is active. Taking a news application as an example, assuming that there are N2 active users among the N1 registered users, these N2 active users can be screened out as target users.

[0043] In a feasible embodiment, after determining the target users, the step of dividing or grouping the target users may be included. The step of dividing or grouping the target users may be as follows: the determined target users may be hashed into N buckets, and each bucket may be assigned M users. For example, taking 550 million target users as an example, these 550 million users may be hashed into 100 buckets, each bucket having approximately 5.5 million users. Each bucket may have a task to recall these users, and the task is started every half an hour. These tasks are independent of each other, which reduces system risks. The single execution time of each task is 5-7 minutes, which is sufficient to meet the performance requirements and reserves space for future expansion of the information pool and larger-scale user growth.

[0044] In the embodiments of the present specification, the portrait is user preference description information abstracted from the historical user behavior data of the target user, which can be used to characterize the target user's preference for each predetermined information category and each predetermined information tag.

[0045] In some embodiments, the portraits used for automated push may include at least push portraits, news portraits, seven-day portraits, video portraits, external portraits, WeChat plug-in portraits, etc. Push portraits refer to portraits generated through user push click behavior; news portraits refer to portraits accumulated by user clicks on news channels; seven-day portraits refer to portraits generated by users through push clicks and news clicks in the past seven days; video portraits refer to portraits accumulated by users clicking on video channels; external portraits refer to portraits introduced from outside the current application. Taking a news application (application, app) as an example, portraits introduced from various channels outside the news app, such as browsers, servers, etc., are collectively referred to as external portraits; WeChat plug-in portraits refer to portraits accumulated through WeChat plug-in clicks. In actual applications, the importance of each portrait varies, and the importance of each portrait will also serve as an important basis for selecting recall information.

[0046] In a feasible embodiment, when loading the portrait, the user's experimental information (such as the user's experimental number, experimental parameters) will also be loaded to perform iterative optimization of the experiment. The experimental information will be used to verify the accuracy of the information push method provided in the embodiment of this specification.

[0047] In a feasible embodiment, multiple portrait information such as the target user's push portrait, news portrait, seven-day portrait, video portrait, external portrait, WeChat plug-in portrait, etc. can be obtained, and the obtained multiple portrait information can be merged together to form a portrait information set of the target user.

[0048] S203. Based on the portrait information set, determine the recall information set corresponding to the target user.

[0049] In the embodiment of this specification, recall may specifically be to trigger as many correct results as possible from the full information set and return the results.

[0050] In the embodiment of this specification, determining the recall information set corresponding to the target user based on the portrait information set may include:

[0051] According to different recall reasons, the portrait information in the portrait information set and the information to be pushed in the information pool are matched for similarity to obtain multiple matching recall information.

[0052] The plurality of matching recall information are determined as a recall information set corresponding to the target user.

[0053] In a feasible embodiment, the portrait information in the portrait information set can be matched with the information to be pushed in the information pool according to different recall reasons, such as primary category recall, secondary category recall, tag recall, topic recall, ItemCF recall, related news recall, popular recall, etc., to obtain multiple matching recall features. Among them, primary category recall and secondary category recall are a classification method based on directory structure. The primary category can include "current affairs", "science and technology", "sports", "entertainment", "education", "travel", "food", "health", etc. The secondary category is a subdivision of the primary category, such as the primary category "sports" can be subdivided into secondary categories such as "football", "basketball", "swimming", and "diving". Tag means tag, tag recall is a recall method based on tag classification, topic recall is a recall method based on subject classification, and ItemCF recall, referred to as ICF recall, is a collaborative filtering recall method based on items.

[0054] In a feasible embodiment, for example, through the primary category recall, the portrait information is matched with the primary title (sports, science and technology, etc.) of the information to be pushed in the information pool for similarity, and the first matching recall information is obtained; through the secondary category recall, the portrait information is matched with the secondary title (football, basketball, etc.) of the information to be pushed in the information pool for similarity, and the second matching recall information is obtained; through the tag recall, the tag in the portrait information is matched with the tag in the information to be pushed in the information pool for similarity, and the third matching recall information is obtained; through the ItemCF recall, the portrait information is matched with the information to be pushed in the information pool for similarity, and the fourth matching recall information is obtained, etc. Then, the first matching recall information, the second matching recall information, the third matching recall information, the fourth matching recall information, etc. obtained through different recall reasons are merged to obtain the recall information set corresponding to the target user. By recalling information through multiple portrait sources and multiple recall reasons, the recall methods are more diversified. Various recall methods in the feed flow can be followed and personalized recall can be performed in combination with specific business scenarios of push. The so-called feed flow refers to the information flow that is continuously updated and presented to users. At the same time, it can deeply understand and meet users' reading needs and enhance personalization. In addition, the above-mentioned recall reasons are based on user dimensions, and can flexibly handle and macro-control users' recalled articles, which is in line with the product positioning of personalized user reading.

[0055] In a feasible embodiment, the information to be pushed includes but is not limited to text information, video information, audio information, picture information, etc. The embodiments of this specification do not limit the type of the information to be pushed. Accordingly, different types of candidate information can correspond to different application scenarios. For example, text information can correspond to news reading scenarios, video information can correspond to user on-demand film and television program scenarios, audio information can correspond to user on-demand song scenarios, and picture information can correspond to user browsing picture scenarios. Therefore, the information recommendation method provided in this embodiment can be applied to different application scenarios according to different types of candidate information, and has a wider range of applications.

[0056] In a feasible embodiment, the information in the automatic push information pool is generally information that is not very real-time. The so-called information that is not very real-time specifically refers to information that is not being broadcast live. For example, a live basketball game can be considered to be information that is very real-time. Every time an editor publishes a piece of information, the information will enter the information pool. Each piece of information has an expiration time. Once the information expires, it will no longer be sent and will not be recalled. Each recall reason can set its own information pool.

[0057] In a feasible embodiment, after determining the recall information set corresponding to the target user based on the portrait information set, the method may further include:

[0058] Each recalled information is filtered for duplicate removal or region filtering.

[0059] Filtering the recall information can facilitate the selection of high-quality recall results, effectively improve the accuracy of information push, and reduce the probability of erroneous information push.

[0060] S205. Obtain multiple recall attribute features corresponding to each recall information.

[0061] In the embodiments of the present specification, when recalling the recall information through different recall reasons, the system will automatically record multiple recall attribute features corresponding to each recall information. Among them, the multiple recall attribute features corresponding to each recall information include multiple recall reasons corresponding to each recall information, multiple recall portraits corresponding to each recall information, and multiple recall scores corresponding to each recall information. The recall portrait represents at least one portrait information in the portrait information set, the recall reason represents the way of obtaining the recall information, and the recall score represents the matching degree between the portrait information of the target user and the information to be pushed.

[0062] In a feasible embodiment, taking the information as news information and the scene as a news reading scene as an example, if N news articles are obtained through S203, article 1 is based on push portrait, news portrait and seven-day portrait, and is obtained through the first-level recall method, and the matching score of article 1 through the first-level recall is 90%, then when the article 1 is obtained, the system will automatically record the recall reason of article 1 as the first-level recall, the recall portrait is the push portrait, the news portrait and the seven-day portrait, and the recall score is 90%, and article 2 is based on push portrait, external portrait and WeChat plug-in portrait, and is obtained through the tag recall method, and the matching score of article 2 through tag recall is 95%, then when the article 2 is obtained, the system will automatically record the recall reason of article 2 as tag recall, the recall portrait is the push portrait, the external portrait and the WeChat plug-in portrait, and the recall score is 95%, and so on, you can get multiple recall attribute features corresponding to each recall information.

[0063] In actual applications, it is possible that an article will be recalled multiple times. Taking Article 1 as an example, in addition to being recalled by the above methods, Article 1 can also be recalled through popular recall methods based on video portraits, external portraits, and WeChat portraits. The recall match degree is 90%. The system will automatically record the recall attribute characteristics corresponding to each recall method.

[0064] In the embodiments of this specification, based on a large amount of data analysis, using features such as multiple portrait sources, multiple recall reasons, and recall scores, and combining incremental extended articles and regional articles, the user's reading needs can be deeply understood and met, thereby enhancing personalization.

[0065] S207. Fusing the multiple recall attribute features corresponding to each recall information to obtain a fused attribute feature corresponding to the recall information.

[0066] In the embodiments of the present specification, after obtaining a plurality of recall attribute features corresponding to each recall information, the obtained plurality of recall attribute features may be fused to obtain a fused attribute feature corresponding to each recall information.

[0067] In a feasible embodiment, considering the pressure of the click prediction model, fewer features can be selected for subsequent click prediction processing. After a large number of experimental verifications, the four features of recall reason, recall portrait, recall reason + recall score, and recall reason + recall portrait can be used as the features predicted by the click prediction model. The reason is that these four features are not only very refined, but also can reflect the importance of different recalls, the importance of different portrait sources, and the matching of users and information.

[0068] In a possible embodiment, if Figure 3As shown, the fusing of the multiple recall attribute features corresponding to each recall information to obtain the fused attribute features corresponding to the recall information may include:

[0069] S301. Perform an OR operation on multiple recall reasons corresponding to each recall information to obtain a first fused attribute feature corresponding to the recall information.

[0070] In a feasible embodiment, for each recall information, an integer value of exponential 2 can be defined for each recall as the recall reason, and the recall reasons are superimposable and decomposable. That is, if an information is recalled multiple times, its recall reasons can be superimposed together through an "or operation" to form a first fused attribute feature corresponding to each recall information. At the same time, the first fused attribute feature can be used to decompress the information by which recalls were made.

[0071] In a feasible embodiment, the information is news information and the scene is a news reading scene for illustration. If N news articles are obtained through S203, article 1 is recalled by three recall methods, namely, primary recall, secondary recall and tag recall. After the above integer value definition, the recall reason of the primary recall is 1, the recall reason of the secondary recall is 2, and the recall reason of the tag recall is 4. Then, after superposition by "or operation", the final recall reason, that is, the first fusion feature is 7. Through the first fusion feature, it is possible to reversely analyze which recalls the article 1 was recalled by.

[0072] In a feasible embodiment, in addition to superimposing the reasons for recall in digital form using exponents of 2, the reasons for recall may also be superimposed in the form of a character string. The embodiments of this specification do not limit the specific method of superimposing the reasons for recall.

[0073] In actual applications, the richer the recall reasons are, the higher the overall click-through rate will be.

[0074] S303. Perform an OR operation on multiple recall portraits corresponding to each recall information to obtain a second fusion attribute feature corresponding to the recall information.

[0075] In the embodiments of the present specification, similar to the recall reasons, for each recall portrait, an integer value of exponential order of 2 can be defined as the recall portrait, and the recall portrait can also be superimposed and reverse analyzed.

[0076] In a feasible embodiment, in addition to superimposing in digital form using exponents of 2, the recall portraits may also be superimposed in the form of strings. The embodiments of this specification do not limit the specific method of superimposing the recall portraits.

[0077] In actual applications, the richer the portrait overlay, the higher the overall click-through rate.

[0078] S305. Calculate the union of multiple recall reasons corresponding to each recall information and multiple recall scores corresponding to the recall information to obtain a third fused attribute feature corresponding to the recall information.

[0079] In the embodiments of the present specification, the third fusion attribute feature, namely the feature of the combination of "recall reason + recall score", calculates the union of multiple recall reasons corresponding to each recall information and multiple recall scores corresponding to the recall information, specifically means that for each recall information, the multiple recall reasons corresponding to the recall information and the multiple recall scores corresponding to the recall information are merged.

[0080] In the embodiment of this specification, both the primary class and the tag in the user portrait are normalized. Since the primary class recall and the tag recall are inconsistent in the granularity of describing the user's interests, the weight in the portrait cannot be measured by a unified standard. Generally, the weight of the descriptive features with fine granularity is relatively low. In order to solve the above problem, the combined feature of "recall reason and recall score" can be used to characterize the matching situation between the user and the information. It should be noted that the recall reason in the combined feature of "recall reason and recall score" is before the superposition, and this combined feature retains the original information of the recall. In addition, different users have huge differences in the breadth of interests. Especially for users with more interests, their weights are more dispersed, and the recall in the feature will be relatively low, but this will not affect the final prediction of the subsequent click prediction model. For the click prediction model, the higher the recall score, the higher the weight of the feature. The purpose of the embodiment of this specification is to select the candidate information corresponding to the target user, and no comparison between users will be made. Therefore, it is reasonable to use the recall score as a feature.

[0081] S307. Perform an OR operation on multiple recall reasons corresponding to each recall information and multiple recall portraits corresponding to the recall information to obtain a fourth fused attribute feature corresponding to the recall information.

[0082] In the embodiment of this specification, the fourth fusion feature is a feature of the combination of "recall reason + recall portrait". For each recall information, the recall reason corresponding to the recall information and the corresponding recall portrait can be ORed to obtain the fourth fusion attribute feature corresponding to the recall information.

[0083] S209. Input the fused attribute features corresponding to each recalled information into the click prediction model to perform information click prediction processing, and obtain the result of the target user clicking on each recalled information.

[0084] In the embodiments of this specification, in order to consider the synergistic effect of multiple portrait sources and multiple recall methods and facilitate the selection of high-quality recall results, after obtaining the above four fused attribute features, these four fused attribute features can be input into a pre-selected and trained click prediction model for click prediction processing, thereby obtaining the result of the target user clicking on each recall information.

[0085] In the embodiment of this specification, the training process of the click prediction model can be as follows:

[0086] Obtain a sample profile information set of sample users;

[0087] Based on the sample portrait information set, determining a sample recall information set corresponding to the sample user;

[0088] Obtain multiple sample recall attribute features corresponding to each sample recall information;

[0089] Performing fusion processing on the multiple sample recall attribute features corresponding to each sample recall information to obtain the sample fusion attribute features corresponding to the sample recall information;

[0090] The sample fusion attribute features are input into a deep learning model for information click prediction training processing, and the trained model is determined as the click prediction model.

[0091] In the embodiments of this specification, the click prediction model may include any one of the extreme gradient boosting model (eXtremeGradient Boosting, XGBoost), factorization machine (Factorization Machine, FM), field-aware factorization machine (Field-aware Factorization Machine, FFM), and DeepFM model. Among them, the DeepFM model is an integrated FM model and a deep neural network model. The DeepFM model combines the advantages of the breadth and depth models, and jointly trains the FM model and the deep neural network model to simultaneously learn low-order feature combinations and high-order feature combinations.

[0092] Below, the click prediction model is taken as an example of the FM model. Compared with the general linear model, the FM can automatically realize feature crossover, consider the combination effect of features, and has stronger expression ability. The embodiment of this specification adopts the binary crossover FM model, and its objective function is as follows:

[0093]

[0094] Among them, w o and w i are the parameters of the FM model, i and j are variables, and x iis the value of the i-th feature, v i is x i The hidden vector, v j is x j The hidden vector of .

[0095] In a feasible embodiment, inputting the obtained fusion attribute features corresponding to each recall information into a click prediction model to perform information click prediction processing to obtain the result of the target user clicking on each recall information may include:

[0096] The first fusion attribute feature corresponding to each recall information, the second fusion attribute feature corresponding to each recall information, the third fusion attribute feature corresponding to each recall information, and the fourth fusion attribute feature corresponding to each recall information are input into the FM model for information click prediction processing to obtain the probability of the target user clicking on each recall information.

[0097] S2011. Based on the result of the target user clicking on each recalled information, determine a candidate information set corresponding to the target user.

[0098] In the embodiments of this specification, Figure 4 As shown, the determining of the candidate information set corresponding to the target user based on the result of the target user clicking each recalled information may include:

[0099] S401. Based on the probability of the target user clicking on each piece of recalled information, each piece of recalled information is sorted in descending order to obtain a candidate sequence of recalled information.

[0100] S403. Take the first N pieces of recall information in the recall information candidate sequence as a candidate information set, where N is a positive integer greater than or equal to 1.

[0101] In a feasible embodiment, the information is news information and the scene is a news reading scene. If 5 news articles are obtained, the FM model predicts that the probability of article 1 being clicked by the user is 80%, the probability of article 2 being clicked by the user is 90%, the probability of article 3 being clicked by the user is 95%, the probability of article 4 being clicked by the user is 85%, and the probability of article 5 being clicked by the user is 75%. Then the candidate sequence of recalled information is: article 3-article 2-article 4-article 1-article 5. The first 3 articles in the sequence, namely article 3, article 2 and article 4, are used as the candidate information set.

[0102] S2013. Push the candidate information to the target user.

[0103] In the embodiment of the present specification, after obtaining the candidate information set, the candidates in the candidate information set may be recommended to the target user, for example, the top 3 articles may be pushed to the target user.

[0104] In the embodiments of this specification, the fused attribute features are input into the click prediction model for click prediction processing, and the number of information sent is strictly controlled. This takes into account both the limitation on the number of recalls and the synergy of multiple portrait sources and multiple recall methods, making it easier to select high-quality recall results. This not only enables the display of high-quality recall articles to users with higher recall quality, but also establishes a multiple selection mechanism through the joint effect of fusion processing and model prediction, thereby reducing the risk of sent articles deviating from user interests.

[0105] In a feasible embodiment, the AUC (Area Under Curve) model evaluation indicator can be selected for the optimization target of the model. Among them, AUC is defined as the area under the ROC (Receiver Operating Characteristic) curve, which can be used to evaluate the quality of binary classifiers. Experiments show that by using the new push method in the embodiment of this specification, the AUC of the FM model in offline testing reaches about 0.89, and the improvement in online effects is also obvious. From the perspective of online effects, the number of daily active users (Daily Active User, DAU) has increased by 220,000.

[0106] In a feasible embodiment, taking the information as news and the scene as news reading scene as an example, if the target user is interested in sports news, three sports-related news articles will be pushed to the user. The specific terminal display schematic diagram is as follows: Figure 5 shown.

[0107] The automated push in the embodiment of the present application realizes the change from "articles looking for people" to "people looking for articles", which lays the foundation for enriching the recall method, expanding the information pool, and deeply optimizing the recall. With many recalls integrated into the automated push framework, the information selection is more refined and the recall quality is higher.

[0108] Figure 6 The figure shows an application scenario diagram of an information push method provided by an embodiment of the present specification. The application scenario is a news reading scenario, and the information is a news article, such as Figure 6As shown in the figure, when it is necessary to push news to the target user, firstly, the push portrait, video portrait, external portrait, seven-day portrait, etc. of the target user are loaded to generate the portrait information set of the target user, and at the same time, the news articles are loaded into the article pool, and different recall methods are adopted to match the portrait information of the target user with the articles in the article pool, so as to obtain multiple recalled articles corresponding to the target user and record the recall reason, recall portrait and recall score corresponding to each recalled article, perform duplicate filtering or regional processing on the obtained multiple news articles, and then fuse the recall reason, recall portrait and recall score corresponding to each recalled article to obtain the fused attribute feature corresponding to each recalled article, and input the fused attribute feature into the FM model for click prediction processing to obtain the top 3 recalled articles, and finally push the top 3 recalled articles to the user. On the one hand, it realizes the user-centric recall of news and can flexibly process the user recall information. On the one hand, it has macro-management and macro-control, which is in line with the product positioning of personalized reading for users. On the other hand, it can follow various recall methods in the feed flow and combine them with specific business scenarios of push for personalized recall. On the other hand, based on a large amount of data analysis, it uses features such as multiple portrait sources, multiple recall reasons, and recall scores, and combines incremental extended articles and regional articles to deeply understand and meet user reading needs, recommend news that users are interested in, and enhance personalization. On the other hand, on the basis of deeply understanding and meeting user reading needs and enhancing personalization, it can give priority to increasing the number of users. On the other hand, based on the fusion of attribute features, a prediction model is used to predict the results of each recall information being clicked, and the prediction results are more accurate and reliable. On the other hand, user-centric recall information is also very friendly to the expansion of the article pool, and will not bring pressure to the system. It can easily cope with the user scale of 550 million and that may continue to grow in the future.

[0109] like Figure 7 As shown, the embodiment of the present application also provides an information push device, the device comprising:

[0110] The first acquisition module 701 is used to acquire a portrait information set of a target user.

[0111] The first determination module 703 is used to determine a recall information set corresponding to the target user based on the portrait information set.

[0112] The second acquisition module 705 is used to acquire multiple recall attribute features corresponding to each recall information.

[0113] The fusion module 707 is used to fuse the multiple recall attribute features corresponding to each recall information to obtain a fused attribute feature corresponding to the recall information.

[0114] The prediction module 709 is used to input the fused attribute features corresponding to each recalled information into the click prediction model to perform information click prediction processing to obtain the result of the target user clicking on each recalled information.

[0115] The second determination module 7011 is used to determine a candidate information set corresponding to the target user based on the result of the target user clicking on each recalled information.

[0116] The push module 7013 is used to push the candidate information to the target user.

[0117] In some embodiments, the first determining module 703 may further include:

[0118] The matching recall information acquisition unit is used to perform similarity matching processing on the portrait information in the portrait information set and the information to be pushed in the information pool according to different recall reasons, so as to obtain multiple matching recall information.

[0119] A recall information determination unit is used to determine the multiple matching recall information as a recall information set corresponding to the target user.

[0120] In some embodiments, the second acquisition module 705 may further include:

[0121] Obtain multiple recall reasons corresponding to each recall information, multiple recall portraits corresponding to each recall information, and multiple recall scores corresponding to each recall information.

[0122] Among them, the recall portrait represents at least one portrait information in the portrait information set, the recall reason represents the way of obtaining the recall information, and the recall score represents the matching degree between the portrait information of the target user and the information to be pushed.

[0123] In some embodiments, the fusion module 707 may further include:

[0124] The first fused attribute feature acquisition unit is used to perform an OR operation on multiple recall reasons corresponding to each recall information to obtain a first fused attribute feature corresponding to the recall information.

[0125] The second fused attribute feature acquisition unit is used to perform an OR operation on multiple recall portraits corresponding to each recall information to obtain the second fused attribute feature corresponding to the recall information.

[0126] The third fused attribute feature acquisition unit is used to calculate the union of multiple recall reasons corresponding to each recall information and multiple recall scores corresponding to the recall information to obtain the third fused attribute feature corresponding to the recall information.

[0127] The fourth fused attribute feature acquisition unit is used to perform an OR operation on multiple recall reasons corresponding to each recall information and multiple recall portraits corresponding to the recall information to obtain a fourth fused attribute feature corresponding to the recall information.

[0128] In some embodiments, the prediction module 709 may be further configured to:

[0129] The first fusion attribute feature corresponding to each recall information, the second fusion attribute feature corresponding to each recall information, the third fusion attribute feature corresponding to each recall information, and the fourth fusion attribute feature corresponding to each recall information are input into the click prediction model to perform information click prediction processing to obtain the probability of the target user clicking on each recall information; wherein, the click prediction model is obtained through machine learning training.

[0130] In some embodiments, the second determining module 7011 may further include:

[0131] The recall information candidate sequence acquisition unit is used to arrange each recall information in descending order based on the probability of the target user clicking on each recall information to obtain the recall information candidate sequence.

[0132] The candidate information set determination unit is used to use the first N pieces of recall information in the recall information candidate sequence as the candidate information set, where N is a positive integer greater than or equal to 1.

[0133] In some embodiments, the device may further include a filtering module, and the filtering module is used to perform duplicate filtering or regional filtering on each recall information.

[0134] It should be noted that the information push device provided in the embodiments of this specification and the above-mentioned information push method belong to the same inventive concept.

[0135] An embodiment of the present application also provides an information push device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the information push method provided in the above method embodiment.

[0136] An embodiment of the present application also provides a storage medium, which can be set in a terminal to store at least one instruction, at least one program, a code set or an instruction set related to an information push method in a method embodiment. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the information push method provided by the above-mentioned method embodiment.

[0137] Optionally, in the embodiment of this specification, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the above storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0138] The memory described in the embodiments of this specification can be used to store software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, application programs required for functions, etc.; the data storage area may store data created according to the use of the device, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide the processor with access to the memory.

[0139] The information push method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal, a server or a similar computing device. Taking running on a server as an example, Figure 8 1 is a hardware structure block diagram of a server of an information push method provided in an embodiment of the present application. Figure 8As shown, the server 800 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 810 (the processor 810 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 830 for storing data, and one or more storage media 820 (such as one or more mass storage devices) for storing application programs 823 or data 822. Among them, the memory 830 and the storage medium 820 can be short-term storage or permanent storage. The program stored in the storage medium 820 may include one or more modules, each of which may include a series of instruction operations on the server. Furthermore, the central processing unit 810 can be configured to communicate with the storage medium 820 and execute a series of instruction operations in the storage medium 820 on the server 800. The server 800 may also include one or more power supplies 860, one or more wired or wireless network interfaces 850, one or more input and output interfaces 840, and / or one or more operating systems 821, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0140] The input / output interface 840 may be used to receive or send data via a network. The specific example of the network may include a wireless network provided by a communication provider of the server 800. In one example, the input / output interface 840 includes a network adapter (Network Interface Controller, NIC), which may be connected to other network devices via a base station so as to communicate with the Internet. In one example, the input / output interface 840 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0141] It can be understood by those skilled in the art that Figure 8 The structure shown is for illustration only and does not limit the structure of the above electronic device. Figure 8 More or fewer components as shown, or with Figure 8 Different configurations shown.

[0142] It should be noted that the above-mentioned sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above-mentioned specific embodiments of this specification are described. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0143] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and server embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0144] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0145] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. An information push method, characterized in that: The method comprises: Obtain the target user's portrait information set; Based on the portrait information set, determining a recall information set corresponding to the target user; Acquire multiple recall attribute features corresponding to each recall information; the multiple recall attribute features corresponding to each recall information include multiple recall reasons corresponding to each recall information, multiple recall portraits corresponding to each recall information, and multiple recall scores corresponding to each recall information; the recall portrait represents at least one portrait information in the portrait information set, the recall reason represents the way of obtaining the recall information, and the recall score represents the matching degree between the portrait information of the target user and the information to be pushed; The multiple recall attribute features corresponding to each recall information are fused to obtain the fused attribute features corresponding to the recall information, including: performing an OR operation on the multiple recall reasons corresponding to each recall information to obtain the first fused attribute features corresponding to the recall information; performing an OR operation on the multiple recall portraits corresponding to each recall information to obtain the second fused attribute features corresponding to the recall information; calculating the union of the multiple recall reasons corresponding to each recall information and the multiple recall scores corresponding to the recall information to obtain the third fused attribute features corresponding to the recall information; performing an OR operation on the multiple recall reasons corresponding to each recall information and the multiple recall portraits corresponding to the recall information to obtain the fourth fused attribute features corresponding to the recall information; Inputting the fused attribute features corresponding to each recalled information into the click prediction model to perform information click prediction processing, and obtaining the result of the target user clicking on each recalled information; Determine a candidate information set corresponding to the target user based on the result of the target user clicking each recalled information; The candidate information is pushed to the target user.

2. The method according to claim 1, characterized in that The determining, based on the portrait information set, a recall information set corresponding to the target user includes: According to different recall reasons, the portrait information in the portrait information set is matched with the information to be pushed in the information pool to obtain a plurality of matching recall information; The plurality of matching recall information are determined as a recall information set corresponding to the target user.

3. The method according to claim 2, characterized in that The obtaining of multiple recall attribute features corresponding to each recall information includes: Obtain multiple recall reasons corresponding to each recall information, multiple recall portraits corresponding to each recall information, and multiple recall scores corresponding to each recall information.

4. The method according to claim 1, characterized in that: The fusion attribute features corresponding to each recalled information are input into the click prediction model to perform information click prediction processing, and the result of the target user clicking on each recalled information is obtained, including: The first fusion attribute feature corresponding to each recall information, the second fusion attribute feature corresponding to each recall information, the third fusion attribute feature corresponding to each recall information, and the fourth fusion attribute feature corresponding to each recall information are input into the click prediction model to perform information click prediction processing to obtain the probability of the target user clicking on each recall information; wherein, the click prediction model is obtained through machine learning training.

5. The method according to claim 4, characterized in that The determining of a candidate information set corresponding to the target user based on the result of the target user clicking each recalled information includes: Based on the probability of the target user clicking on each recall information, each recall information is sorted in descending order to obtain a recall information candidate sequence; The first N pieces of recall information in the recall information candidate sequence are taken as a candidate information set, where N is a positive integer greater than or equal to 1.

6. The method according to claim 1, characterized in that After determining the recall information set corresponding to the target user based on the portrait information set, the method further includes: Each recalled information is filtered for duplicate removal or region filtering.

7. An information push device, characterized in that: The device comprises: The first acquisition module is used to acquire a portrait information set of a target user; A first determination module, configured to determine a recall information set corresponding to the target user based on the portrait information set; The second acquisition module is used to acquire multiple recall attribute features corresponding to each recall information; the multiple recall attribute features corresponding to each recall information include multiple recall reasons corresponding to each recall information, multiple recall portraits corresponding to each recall information, and multiple recall scores corresponding to each recall information; the recall portrait represents at least one portrait information in the portrait information set, the recall reason represents the way of acquiring the recall information, and the recall score represents the matching degree between the portrait information of the target user and the information to be pushed; A fusion module is used to fuse the multiple recall attribute features corresponding to each recall information to obtain the fused attribute features corresponding to the recall information; the fusion module includes: a first fused attribute feature acquisition unit, used to perform an OR operation on the multiple recall reasons corresponding to each recall information to obtain the first fused attribute feature corresponding to the recall information; a second fused attribute feature acquisition unit, used to perform an OR operation on the multiple recall portraits corresponding to each recall information to obtain the second fused attribute feature corresponding to the recall information; a third fused attribute feature acquisition unit, used to calculate the union of the multiple recall reasons corresponding to each recall information and the multiple recall scores corresponding to the recall information to obtain the third fused attribute feature corresponding to the recall information; a fourth fused attribute feature acquisition unit, used to perform an OR operation on the multiple recall reasons corresponding to each recall information and the multiple recall portraits corresponding to the recall information to obtain the fourth fused attribute feature corresponding to the recall information; A prediction module is used to input the fusion attribute features corresponding to each recalled information into a click prediction model to perform information click prediction processing, and obtain the result of the target user clicking on each recalled information; A second determination module, configured to determine a candidate information set corresponding to the target user based on the result of the target user clicking each recalled information; The push module is used to push the candidate information to the target user.

8. The device according to claim 7, characterized in that The first determining module includes: A matching recall information acquisition unit, used to perform similarity matching processing on the portrait information in the portrait information set and the information to be pushed in the information pool according to different recall reasons, so as to obtain a plurality of matching recall information; A recall information determination unit is used to determine the multiple matching recall information as a recall information set corresponding to the target user.

9. The device according to claim 8, characterized in that The second acquisition module is used to: Obtain multiple recall reasons corresponding to each recall information, multiple recall portraits corresponding to each recall information, and multiple recall scores corresponding to each recall information.

10. The device according to claim 7, characterized in that The prediction module is used to: The first fusion attribute feature corresponding to each recall information, the second fusion attribute feature corresponding to each recall information, the third fusion attribute feature corresponding to each recall information, and the fourth fusion attribute feature corresponding to each recall information are input into the click prediction model to perform information click prediction processing to obtain the probability of the target user clicking on each recall information; wherein, the click prediction model is obtained through machine learning training.

11. The device according to claim 10, characterized in that The second determining module includes: A recall information candidate sequence acquisition unit, configured to arrange each recall information in descending order based on the probability of the target user clicking on each recall information, to obtain a recall information candidate sequence; The candidate information set determination unit is used to use the first N pieces of recall information in the recall information candidate sequence as the candidate information set, where N is a positive integer greater than or equal to 1.

12. The device according to claim 7, characterized in that The device also includes a filtering module, which is used to perform duplicate removal filtering or regional filtering on each recall information.

13. A device, characterized in that The device includes: a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the information push method as described in any one of claims 1-6.

14. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the information push method as described in any one of claims 1-6.

Images