Information recommendation method, device, storage medium and computer equipment

Abstract

The present application provides an information recommendation method, apparatus, storage medium and computer equipment, which are modeled based on the attention mechanism and machine learning algorithm so that the constructed probability prediction model can take into account the long-term and short-term interests of users, and input the acquired user access sequence and the object identification of the candidate object into the probability prediction model so that when predicting the current user's target interest probability for each candidate object, the access order of different historical access objects can be taken into account, so as to accurately locate the user's interest changes and interest accumulation, improve the accuracy of the target interest probability of each candidate object, and then improve the accuracy of the recommended objects obtained based on the target interest probability of each candidate object, that is, improve the accuracy of personalized recommendations for different users.

Description

Technical Field

[0001] The present application mainly relates to the field of data processing technology, and more specifically to an information recommendation method, device, storage medium and computer equipment. Background Art

[0002] The emergence and popularization of the Internet has brought a large amount of information to users, satisfying their needs for information in the information age. However, with the rapid development of the Internet, the amount of online information has increased significantly, making it impossible for users to obtain the information that is truly useful to them when faced with a large amount of information, and the efficiency of using information has been reduced. In order to solve this problem, a personalized information recommendation system is proposed to recommend information and products of interest to users based on their information needs and interests.

[0003] Taking the information recommendation scenario of a community website as an example, currently, it is usually based on the click attributes of each post. Popular posts, such as the top N posts in terms of click volume, are screened out from multiple posts in the community, and the screened posts are pushed to users. This does not take into account the real interests of users, resulting in the recommended posts content that is likely not what the users are interested in. At the same time, a large number of valuable posts will be buried, which is not conducive to the operation of the community website.

[0004] In this regard, the prior art proposes a recommendation method based on collaborative filtering (CF), that is, by calculating the item-item (item can be a post) similarity, the K items that are most similar to the posts visited by the user are obtained as recommended posts. However, for technology community websites, the user access time interval is long, resulting in sparse user access data, which reduces the recommendation efficiency and accuracy. Moreover, if different users visit the same post item within a period of time, but the order of accessing the posts is different, the recommended posts pushed to different users according to this recommendation method based on ItemCF will be the same, which cannot achieve personalized recommendations and also reduces the recommendation accuracy. Summary of the invention

[0005] In view of this, the present invention provides an information recommendation method, the method comprising:

[0006] Acquire a user access sequence and an object identifier of a candidate object, wherein the user access sequence is composed of object identifiers of objects that the user has historically accessed on the application platform;

[0007] Input the user access sequence and the object identifier of the candidate object into a probability prediction model to obtain the probability that the current user is interested in the target of the candidate object, wherein the probability prediction model is obtained by training long-term sample data and short-term sample data based on an attention mechanism and a machine learning algorithm, wherein the long-term sample data refers to the object identifier of the historical access object generated within a specific time period, and the short-term sample data refers to the object identifier of the historical access object generated at the previous moment;

[0008] Based on the target interest probability of the candidate objects, obtaining recommended objects among the candidate objects;

[0009] The recommended object is output.

[0010] In some embodiments, inputting the user access sequence and the object identifier of the candidate object into a probability prediction model to obtain the target interest probability of the current user for the candidate object includes:

[0011] Based on the user access sequence and the object identifier of the candidate object, obtaining a predicted access probability of accessing the candidate object at a current moment;

[0012] Using the predicted access probability of the candidate object, a correlation analysis is performed on the candidate object and the historical access objects generated within a specific time period to obtain a first probability of interest of the current user in the candidate object, and a correlation analysis is performed on the candidate object and the historical access objects generated at the previous moment to obtain a second probability of interest of the current user in the candidate object;

[0013] The first interest probabilities and the second interest probabilities corresponding to different candidate objects are weightedly summed to obtain the target interest probabilities of the current user for the corresponding candidate objects.

[0014] In some embodiments, using the predicted access probability of the candidate object to perform a correlation analysis on the candidate object and historical access objects generated within a specific time period to obtain a first probability of interest of the current user in the candidate object includes:

[0015] The first user access sequence of different users is learned through an attention mechanism to obtain a first attention degree of the current user to the historical access object, wherein the first user access sequence is composed of object identifiers of access objects accessed by the corresponding user on the application platform within a specific time period;

[0016] According to the predicted access probability of the candidate object and the first attention level, the correlation between the corresponding candidate object and the access object accessed by the current user on the application platform within the specific time period is obtained to obtain a first probability of interest of the current user in the corresponding candidate object;

[0017] The step of using the predicted access probability of the candidate object to perform a correlation analysis on the candidate object and the historical access object generated at the last moment to obtain a second probability of interest of the current user in the candidate object includes:

[0018] Analyzing second user access sequences of different users to obtain a second degree of attention of the current user to the historical access object, wherein the second user access sequence is composed of object identifiers of access objects accessed by the corresponding user at the last moment on the application platform;

[0019] According to the predicted access probability of the candidate object and the second attention level, the correlation between the corresponding candidate object and the access object visited by the current user on the application platform at the last moment is obtained to obtain the second interest probability of the current user in the corresponding candidate object.

[0020] In some embodiments, the number of the candidate objects is multiple, and obtaining the recommended objects from the candidate objects based on the target interest probabilities of the candidate objects includes:

[0021] From the plurality of candidate objects, a specific number of candidate objects having a relatively high probability of being of interest to the target are selected as objects to be recommended;

[0022] Based on the attention mechanism and the Bayesian personalized ranking algorithm, the object identifier of the object accessed on the application platform at the last moment is used to obtain the ranking result of the current user's interest in the object to be recommended;

[0023] According to the interest ranking result, the recommended objects for the current user are screened from the objects to be recommended.

[0024] In some embodiments, obtaining the predicted access probability of accessing the candidate object at the current moment based on the user access sequence and the object identifier of the candidate object includes:

[0025] constructing a probability transfer matrix using the user access sequence and the object identifiers of the candidate objects;

[0026] The probability transfer matrix is ​​processed by matrix decomposition to obtain the predicted access probability of the current user accessing the candidate object at the current moment.

[0027] The present application also proposes an information recommendation device, the device comprising:

[0028] A data acquisition module, used to acquire a user access sequence and object identifiers of candidate objects, wherein the user access sequence is composed of object identifiers of objects that the user has historically accessed on the application platform;

[0029] A prediction module is used to input the user access sequence and the object identifier of the candidate object into a probability prediction model to obtain the probability that the current user is interested in the target of the candidate object. The probability prediction model is obtained by training long-term sample data and short-term sample data based on an attention mechanism and a machine learning algorithm. The long-term sample data refers to the object identifier of the historical access object generated within a specific time period, and the short-term sample data refers to the object identifier of the historical access object generated at the previous moment.

[0030] A recommended object obtaining module, used for obtaining a recommended object from the candidate objects based on the target interest probability of the candidate objects;

[0031] The recommended object output module is used to output the recommended object.

[0032] In some embodiments, the prediction module includes:

[0033] A predicted access probability acquisition unit, used to acquire a predicted access probability of accessing the candidate object at a current moment based on the user access sequence and the object identifier of the candidate object;

[0034] A correlation analysis unit is used to use the predicted access probability of the candidate object to perform a correlation analysis on the candidate object and the historical access objects generated within a specific time period to obtain a first probability of interest of the current user in the candidate object, and to perform a correlation analysis on the candidate object and the historical access objects generated at a previous moment to obtain a second probability of interest of the current user in the candidate object;

[0035] The target interest probability obtaining unit is used to perform weighted summation of the first interest probabilities and the second interest probabilities corresponding to different candidate objects respectively, to obtain the target interest probability of the current user for the corresponding candidate object.

[0036] The present application also proposes a storage medium on which a program is stored. The program is called and executed by a processor to implement the above-mentioned information recommendation method.

[0037] The present application also proposes a computer device, the computer device comprising: at least one communication interface, at least one memory and at least one processor, wherein:

[0038] The memory is used to store a program for implementing the above-mentioned information recommendation method;

[0039] The processor is used to load and execute the program stored in the memory to implement the various steps of the above-mentioned information recommendation method.

[0040] It can be seen that compared with the prior art, the present application provides an information recommendation method, apparatus, storage medium and computer device, which are modeled based on the attention mechanism and machine learning algorithm so that the constructed probability prediction model can take into account the user's long-term and short-term interests, and input the acquired user access sequence and the object identification of the candidate object into the probability prediction model, so that when predicting the current user's target interest probability for each candidate object, it can take into account the access order of different historical access objects, so as to accurately locate the user's interest changes and interest accumulation, improve the accuracy of the target interest probability of each candidate object, and then improve the accuracy of the recommended objects obtained based on the target interest probability of each candidate object, that is, improve the accuracy of personalized recommendations for different users. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0042] Figure 1 A schematic diagram showing an optional example of a flow chart of the information recommendation method proposed in the present application;

[0043] Figure 2 A schematic diagram of a scenario showing an optional example of the information recommendation method proposed in the present application;

[0044] Figure 3 A flowchart showing another optional example of the information recommendation method proposed in the present application is shown;

[0045] Figure 4 A schematic diagram showing a process of obtaining a probability transfer matrix for a current user in the information recommendation method proposed in the present application is shown;

[0046] Figure 5 A schematic diagram showing a curve of F1 value changes in recommendation results of the information recommendation method proposed in the present application and the existing information method in an optional scenario;

[0047] Figure 6 A schematic diagram showing a change curve of the F1 value of the recommendation results of the information recommendation method proposed in the present application and the existing information method in another optional scenario;

[0048] Figure 7 A schematic diagram showing the structure of an optional example of the information recommendation device proposed in the present application;

[0049] Figure 8A schematic diagram showing the structure of another optional example of the information recommendation device proposed in the present application;

[0050] Fig. 9 A schematic diagram showing the structure of another optional example of the information recommendation device proposed in the present application;

[0051] Fig.10 A schematic diagram of the hardware structure of a computer device proposed in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] It should be noted that, for the convenience of description, only the parts related to the relevant invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0054] It should be understood that the "system", "device", "unit" and / or "module" used in this application is a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.

[0055] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "a kind" and / or "the" do not refer to the singular, but also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of clearly identified steps and elements, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements. The elements defined by the sentence "includes a..." do not exclude the existence of other identical elements in the process, method, commodity or device that includes the elements.

[0056] Among them, in the description of the embodiments of the present application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, "multiple" refers to two or more than two. The following terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features.

[0057] In addition, flow charts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, the various steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0058] With respect to the existing information recommendation methods described in the background technology section, in order to achieve personalized recommendations for different users and improve the accuracy and flexibility of recommendations, this application proposes to consider the order in which users access objects on the application platform (such as the order in which they click on posts on community websites, etc.) when obtaining recommended information, so as to accurately locate the user's own interest changes and interest accumulation. At the same time, it is also hoped that the user's long-term historical interests and short-term historical interests can be taken into account, so as to improve the accuracy and efficiency of recommendations and meet the personalized recommendation needs of different users.

[0059] Based on this, this application proposes to use the FPMC (Factoring Personalized Markov Chains) model to set a Markov chain transfer matrix (i.e., probability transfer matrix) for each user to obtain the user's long-term preferences and recent concerns to achieve personalized recommendations. However, in the prediction process of the FPMC model, the access object at the current moment is usually predicted based on the access data at the previous moment, and so on, without comprehensively considering the user's objects of interest over a period of time, which affects the accuracy of the recommendation results.

[0060] In order to improve the above problems, this application further proposes to combine the attention mechanism with the FPMC algorithm to accurately and efficiently capture the user's long-term and short-term interests to complete sequence recommendations, thereby improving the accuracy of recommendations.

[0061] Among them, the attention mechanism can be a resource allocation method that allocates computing resources to more important tasks and solves the problem of information overload when the computing power of computer equipment is limited. In neural network learning (i.e., a machine learning algorithm), generally speaking, the more parameters a model has, the stronger the model's expressive power is, and the larger the amount of information stored in the model is. Especially when capturing the long-term interests of users, analyzing a large amount of historical access data can easily cause the problem of information overload. By introducing the attention mechanism, focusing on the information that is more critical to the current task (i.e., the historical access objects that users pay more attention to) among the numerous input information, reducing the attention to other information, and even filtering out irrelevant information, the problem of information overload can be solved and the efficiency of information processing can be improved. This application does not elaborate on the operating principle of the attention mechanism.

[0062] Reference Figure 1 , shows a flow chart of an optional example of the information recommendation method proposed in this application, which can be applied to a computer device, which can be a server (i.e., a recommendation server) or a terminal device with certain data computing capabilities, such as a laptop computer, a desktop computer, etc. This application does not limit the product type of the computer device. Figure 1 As shown, the information recommendation method proposed in this embodiment may include:

[0063] Step S11, obtaining the user access sequence and the object identification of the candidate object;

[0064] In this embodiment, the user access sequence may be composed of object identifiers of historically accessed objects on the application platform, and these object identifiers are arranged in the order of access time of the historically accessed objects. The object identifier may be identification information used to distinguish different accessed objects, and has a unique characteristic, and may specifically be an object ID, etc. The present application does not limit the content of the object identifier.

[0065] In actual applications, users access objects output by the application platform through terminal devices, that is, when users click on objects output by the application platform, corresponding historical access data will be generated. The historical access data can be stored in the terminal device or the application server corresponding to the application platform. Therefore, the computer device can obtain the historical access data of the user from the terminal device or the application server. The historical access data may include: information such as the content and object identification of historical access objects corresponding to various access times. Afterwards, the computer device can use the object identification of the historical access object in the historical range data to generate the user access sequence of the user.

[0066] It should be noted that the method for generating the above-mentioned user access sequence is not limited to the implementation method described above. Moreover, based on the above description of the inventive concept of the present application, the present application needs to capture the user's long-term interests and short-term interests. Therefore, when generating the user access sequence, it is necessary to obtain the historical access data corresponding to a relatively long time period (recorded as a specific time period). According to the above method, a user access sequence that can contain the object identifiers of the historical access objects within the longer time period is generated, so as to ensure that the user's long-term interests can be obtained by subsequent analysis.

[0067] This application does not limit the specific time of the above-mentioned specific time period, such as the past three months or six months starting from the current time. The specific time period can be determined based on the specific recommendation scenario and factors such as the user's requirements for recommendation accuracy. It can also be adjusted according to actual conditions. This application does not elaborate on the method of pre-configuring the specific time period.

[0068] Based on the above analysis, if x i represents the object identifier of the i-th historical access object stored in a specific time period. The user access sequence obtained by the computer device can be [x1, x2, x3, …, x i ,…], the number of elements contained in the user access sequence can be determined according to the number of historical access objects stored in the specific time period. It should be understood that since the user access sequence is arranged in the order of access time, analyzing it can accurately locate the user's interest changes and interest accumulation.

[0069] The candidate objects obtained in step S11 may be objects output by the application platform that have not been visited by the current user and are pre-screened from all objects that can be output by the application platform. Specifically, they may be objects with a high access rate on the application platform, objects related to hot topics in recent community websites, objects with a large number of visits in recent periods of time, or objects randomly collected from objects output by the application platform, etc. This embodiment does not limit the method for selecting candidate objects, and the selection of the candidate objects can be implemented by an application server, a computer device, or a terminal device. This application does not limit the source and number of the candidate objects.

[0070] It should be understood that since user interests or hot topics usually change over time, the candidate objects pre-selected in the above manner may also be continuously updated over time to improve the accuracy of subsequent determination of recommended objects. This application does not limit the method for updating the candidate objects.

[0071] Step S12, inputting the user access sequence and the object identifier of the candidate object into the probability prediction model to obtain the probability of the current user being interested in the target of each candidate object;

[0072] In this embodiment, the probability prediction model can be obtained by training long-term sample data and short-term sample data based on the attention mechanism and machine learning algorithm. The training process of the probability prediction model is not described in detail in this application. Among them, the long-term sample data can refer to the object identification of the historical access object generated in a specific time period, and the short-term sample data refers to the object identification of the historical access object generated in the adjacent previous moment.

[0073] In one possible implementation, combined with the above description of the inventive concept of the present application, the machine learning algorithm used to train the above-mentioned probability prediction model may be the FPMC algorithm. Therefore, the present application may continuously train long-term sample data and short-term sample data based on the operating principles of the FPMC algorithm and the attention mechanism to obtain the above-mentioned probability prediction model.

[0074] Step S13, obtaining recommended objects from the candidate objects based on the target interest probability of each candidate object;

[0075] Step S14: output the recommended object.

[0076] In the actual application of the present application, since the larger the target interest probability of a candidate object is, the more interested the user is in the candidate object in the long and short term, that is, the higher the attention is, the more likely it is to choose it for access at the current moment. Therefore, this embodiment can select a specific number of candidate objects with a large target interest probability from multiple candidate objects as recall objects based on the size of the target interest probability of each candidate object. Afterwards, the selection logic in the recommendation system of the computer device can filter out at least one recommended object that the user may be interested in from the multiple recall objects, and send it to the user client for output.

[0077] The specific implementation process of the selection logic in the above recommendation system is not limited in this application, and can refer to but not limited to the description of the corresponding part of the following embodiment.

[0078] In addition, combined with the above description of the product types of computer devices, the specific implementation methods of step S14 may be different for different types of computer devices, such as sending the obtained recommended objects to other terminal devices for display, saving the recommended objects to a preset storage directory, outputting the recommended objects through a display screen, etc., which can be determined based on actual conditions, and this application will not describe them in detail one by one.

[0079] In summary, in this embodiment, the attention mechanism and the machine learning algorithm are combined for modeling, so that the constructed probability prediction model can take into account the long-term and short-term interests of the user, and at the same time, the model input is obtained in a sequence manner, that is, the user access sequence and the object identifier of the candidate object are obtained as the model input for prediction. Considering the access sequence of different historical access objects, the user's interest changes and interest accumulation can be more accurately located, and the accuracy of the target interest probability of each candidate object obtained can be improved, thereby improving the accuracy of the recommended objects obtained based on the target interest probability of each candidate object, that is, the personalized recommendation accuracy of different users is improved. Improve the accuracy of personalized recommendations.

[0080] Based on the above analysis, we take the application scenario where a computer device is used as a recommendation server and a user uses a terminal device to visit community website A as an example to illustrate how the recommendation server pushes recommended posts that the user may be interested in. Figure 2 In the scenario flow diagram shown, a user can access community website A on a terminal device, that is, access the application server corresponding to the community website. The historical access data generated during the access process can be stored in the application server. For example, a user can click on one or more posts from the posts output by community website A to generate corresponding click records, which include information such as click time and click post ID. This information can be stored in the application server or terminal device as historical access data. This application is only explained by taking storage in the application server as an example.

[0081] When a user views a post output by community network A, the recommendation server can obtain the historical access data corresponding to the user ID in a specific time period from the application server in the manner described in the above embodiment, and use the post IDs of the historically clicked posts contained therein to form a user access sequence. In addition, the recommendation server can also preliminarily screen out multiple candidate posts from multiple posts that the user has not clicked on contained in the application server, and the candidate posts can be continuously updated over time. The specific implementation process will not be described in detail.

[0082] Afterwards, the recommendation server can input the user access sequence composed of the post IDs of multiple historical clicked posts in sequence and the post IDs of multiple candidate posts into the pre-trained probability prediction model, and can obtain the target interest probability of the user in each candidate post. Afterwards, the recommended post for the user can be determined from the multiple candidate posts based on the target interest probability of each candidate post, and the recommended post can be directly sent to the user's terminal device for output using the user ID. Of course, the terminal device can also output corresponding prompt information, and after detecting the user's confirmation instruction to view the recommended post, it will output the corresponding recommended post. This application does not limit the specific output method of the recommended post.

[0083] Of course, if Figure 2 As shown, the recommendation server obtains the recommended posts for the user, which can be fed back to the user's terminal device for output through the application server. This application does not limit the specific implementation method of how to feed back at least one recommended post of the user to the user community website A for output.

[0084] It should be noted that in some embodiments, the above-mentioned application server and the recommendation server can be the same computer device, that is, the recommendation system can be deployed in the application server of the community website A and used as the recommendation server; of course, in some other embodiments, the above-mentioned recommendation server and the terminal device can be the same computer device, and this application does not limit this and can be determined according to actual needs.

[0085] In addition, if Figure 2 In the implementation process of the recommendation server pushing personalized recommended posts to different users is similar, but because the user access sequences corresponding to different users are different, even if multiple users click to view the same type of posts within a specific time period, the click order of viewing posts is often different. According to the above method, the recommended posts pushed to each of these multiple users will also be different.

[0086] Reference Figure 3 , shows a flow chart of another optional example of the information recommendation method proposed in the present application. This embodiment may be an optional refinement of the information recommendation method described in the above embodiment, such as Figure 3 As shown, the information recommendation method proposed in this embodiment may include:

[0087] Step S21, obtaining the user access sequence and the object identifier of the candidate object;

[0088] Regarding the content of the user access sequence and the method for obtaining it, as well as the method for obtaining the object identification of the candidate object, reference may be made to the description of the corresponding parts of the above embodiment.

[0089] Step S22, based on the user access sequence and the object identifier of the candidate object, obtaining the predicted access probability of accessing the candidate object at the current moment;

[0090] In combination with the above description of the inventive concept of the present application, in order to achieve personalized recommendations, the present application will construct a probability transfer matrix for each user. The constituent elements of the probability transfer matrix are the predicted access probability of the user to the candidate object at the current moment. It can be seen that step S22 is actually the process of constructing a probability transfer matrix for the current user. The present application does not elaborate on the specific construction process of the probability transfer matrix.

[0091] Assume that user 1 clicks on posts a, b, and c output by the community website at time t-3, clicks on posts b and c output by the community website at time t-2, and clicks on posts a and b output by the community website at time t-1. This application needs to predict which one or more posts output by the community website the user will click on at time t. Specifically, this application can use these historical data to predict the probability of the user clicking on each candidate post output by the community website that has not been clicked at time t.

[0092] Specifically, in order to accurately locate the user's interest migration and interest accumulation, such as Figure 4 As shown in the schematic diagram of the probability transfer matrix acquisition process, the present application can use the user's clicks on the posts at time t-3 to obtain the probability of the user clicking on each candidate post at time t-2; use the user's clicks on the posts at time t-2 to obtain the probability of the user clicking on each candidate post at time t-1, and so on, to finally obtain the probability of the user clicking on each candidate post at time t.

[0093] In practical applications, such as Figure 4 As shown, the data of the obtained probability transfer matrix is ​​relatively sparse. To this end, this application will fit the sparse matrix into a dense matrix by matrix decomposition to determine Figure 4 The predicted access probability of the position corresponding to the question mark in the figure will not be described in detail in this application.

[0094] Based on the above analysis, the above step S22 may include: constructing a probability transfer matrix using the user access sequence and the object identifier of the candidate object, processing the probability transfer matrix by matrix decomposition, and obtaining the predicted access probability of the current user accessing the candidate object at the current moment.

[0095] In some embodiments, according to the requirements of specific application scenarios, the present application can also combine the attention mechanism with the traditional matrix decomposition method to improve the accuracy of the predicted access probability of the user visiting the candidate object at the current moment obtained after matrix decomposition.

[0096] Step S23, using the predicted access probability of the candidate object, performing a correlation analysis between the candidate object and the historical access objects generated in a specific time period, and obtaining a first probability of interest of the current user in the candidate object;

[0097] Combined with the above analysis of the inventive concept of this application, this application combines the attention mechanism with the FPMC algorithm to analyze the probability transfer matrix constructed for the user, and obtain what objects the user has been interested in output from the community website for a long time, and what objects the user has been interested in output from the community website recently. Combined with the above description of the attention mechanism and the FPMC algorithm, this application will combine the attention mechanism to implement the analysis of the user's long-term interests.

[0098] In a possible implementation, historical access data of different users who visit the community website can be obtained. According to the above method, user access sequences corresponding to different users are obtained, which are recorded as first user access sequences. It can be seen that the first user access sequence is composed of object identifiers of access objects on the application platform (i.e., the application platform where the community website is located) visited by the corresponding users within a specific time period. The specific generation process is not described in detail.

[0099] The computer device can learn the first user access sequence of different users through the attention mechanism to obtain the first attention degree of the current user to its historical access object. This application does not elaborate on the implementation process of how to use the attention mechanism to learn and obtain the first attention degree, which can be determined in combination with the operation principle of the attention mechanism. Among them, the current user usually has multiple historical access objects in a specific time period. According to this processing method, the first attention degree of each of the multiple historical accesses of the current user in the specific time period can be obtained, thereby generating a corresponding attention vector.

[0100] Afterwards, the computer device can use the predicted access probabilities of the multiple candidate objects obtained above and the first attention levels of the multiple historical access objects to obtain the correlation between the corresponding candidate objects and the access objects visited by the current user on the application platform within a specific time period, that is, to obtain the degree of association between each candidate object and the historical access objects within a specific time period, so as to obtain the current user's stable interest, that is, long-term interest, within the specific time period. To facilitate subsequent processing, the first probability of interest of the current user in the corresponding candidate object can be obtained based on the correlation analysis results.

[0101] Optionally, after obtaining the correlation analysis results between each of the multiple candidate objects and the objects visited by the current user on the application platform within a specific time period, since the specific time period is usually a continuous time determined from the previous moment adjacent to the current moment, which includes the objects recently visited by the current user, this embodiment directly averages the multiple correlation analysis results between each candidate object and the multiple historical visited objects. The average value can better indicate the current user's stable interest in the candidate object, so that the first interest probability of the candidate object obtained based on the average value is more accurate.

[0102] It should be understood that the greater the first probability of interest of a candidate object, the higher the degree of association between the candidate object and the historical objects visited by the user in the past specific time period, that is, the more similar the candidate object is to the object that the user pays attention to in the specific time period, the more likely the candidate object is the object that the user is interested in at the current moment. This application does not limit the specific correspondence between the first probability of interest and the above-mentioned correlation analysis result. If the correlation analysis result is a score, the first probability of interest can also be a corresponding score, or a probability value generated based on the score according to a certain proportional relationship, etc.

[0103] Step S24, using the predicted access probability of the candidate object, performing a correlation analysis between the candidate object and the historical access object generated at the last moment, to obtain a second probability of interest of the current user in the candidate object;

[0104] Referring to the description of the corresponding part of the above step S23, the content executed in step S24 is mainly used to obtain the current user's recent interests, that is, the degree of association between each candidate object at the current moment and each historical access object at the previous moment.

[0105] In one possible implementation, a first-order Markov chain may be used to implement the content of step S24. Specifically, the second user access sequences of different users are analyzed to obtain the second degree of attention of the current user to the historical access objects at the previous moment. The second user access sequence is composed of the object identifiers of the access objects that the corresponding users accessed on the application platform at the previous moment. The generation process of the second user access sequence is similar to that of the above-mentioned first user access sequence and is not described in detail.

[0106] It can be seen that the second attention level indicates the current user's interest in each historically visited object at the last moment. Afterwards, based on the predicted access probability of each candidate object and the second attention level corresponding to each historically visited object at the last moment, the correlation between the corresponding candidate object and the visit object of the current user on the application platform at the last moment is obtained to obtain the second interest probability of the current user in the corresponding candidate object. Since the correlation analysis is similar to the long-term interest analysis process of the corresponding part of step S23 above, the analysis process of the current user's short-term interest is not repeated in this application.

[0107] Step S25, performing weighted summation of the first interest probabilities and the second interest probabilities corresponding to different candidate objects respectively, to obtain the target interest probability of the current user for the corresponding candidate object;

[0108] Following the above analysis, the first interest probability is obtained by analyzing the current user's long-term interests (such as object access interests within a specific time period), and the second interest probability is obtained by analyzing the current user's short-term interests (such as object access interests at the previous moment). Therefore, this embodiment combines the first interest probability and the second interest probability to determine the target interest probability of the current user in the corresponding candidate object, so that the target interest probability has both the long-term interests and short-term interests of the current user, thereby making the recommendations determined thereby more accurate and reliable.

[0109] Among them, when obtaining the target interest probability of a candidate object, the weights of the first interest probability and the second interest probability of the corresponding candidate object can be determined according to actual needs, and this application does not limit the specific values ​​of the weights.

[0110] Step S26, selecting a specific number of candidate objects with a relatively high probability of being of interest to the target from among the multiple candidate objects as objects to be recommended;

[0111] Generally, the number of objects that are output by community websites and available for users to access is very large, and will continue to increase over time. The number of candidate objects initially selected in this embodiment is often also large, and they need to be further screened out in order to quickly and accurately obtain recommended objects for the current user. Therefore, this embodiment obtains the target interest probability of each of the multiple candidate objects in the above manner, that is, after combining the long-term interests and short-term interests of the current user, predicting the probability that the current user is interested in each candidate object at the current moment, the candidate objects of long-term interest and short-term interest can be screened out from the multiple candidate objects as the objects to be recommended. This embodiment does not limit the specific implementation process of step S26.

[0112] In some embodiments, the candidate objects may be sorted according to their target interest probabilities, and then a specific number of candidate objects with larger target interest probabilities may be selected as the objects to be recommended, i.e., the recalled objects in the recommendation system. Alternatively, the candidate objects with target interest probabilities greater than a specific probability threshold may be directly selected as the objects to be recommended, and so on.

[0113] Step S27, based on the attention mechanism and the Bayesian personalized ranking algorithm, using the object identifier of the access object on the application platform at the last moment, obtain the ranking result of the current user's interest in the recommended object;

[0114] In some embodiments, after recalling the recommended objects that the current user may be interested in from a large number of candidate objects in the above manner, the Bayesian Personalized Ranking (BPR) method can be used to rank the recommended objects. Specifically, the FPMC algorithm used for screening the recommended objects from multiple candidate objects can be simplified to achieve efficient ranking of the recommended objects. The specific implementation process can be determined based on the operational principle of the Bayesian personalized ranking.

[0115] Among them, the implementation of this Bayesian personalized sorting method is based on the assumption that the preference behaviors between each user are independent of each other, that is, the user's preference between visiting object i and visiting object j is unrelated to other users; and the same user's preferences for different items are independent, that is, the user's preference between visiting object i and visiting object j is unrelated to other visiting objects. Based on these two assumptions, this embodiment can use the Bayesian personalized sorting method to implement personalized sorting of the obtained multiple objects to be recommended, obtain the current user's ranking score for the multiple objects to be recommended, and sort the multiple objects to be recommended according to the ranking score. The size of the ranking score can indicate the probability that the current user is interested in the corresponding object to be recommended. The larger the ranking score, the greater the probability that the current user is interested in the corresponding object to be recommended. This application does not elaborate on the specific process of how to use the Bayesian personalized sorting method to implement the sorting of multiple objects to be recommended.

[0116] In actual applications, the user's interest preference for the access object at the current moment will not change suddenly, and is usually the same as the interest preference at the previous moment. Therefore, in order to improve the sorting efficiency, for the recalled objects to be recommended, based on the attention mechanism, only the object identifier of the current user's access object on the application platform at the previous moment is used to obtain the probability of the current user's interest in multiple objects to be recommended, and according to the interest probability, the multiple objects to be recommended are sorted to obtain the current user's interest in the recommended objects. The specific implementation process is not described in detail. In combination with how to use the attention mechanism to analyze the user's long-term interests in the process of obtaining the objects to be recommended above, the process of using the attention mechanism to analyze the user's short-term interests in this embodiment can be determined.

[0117] Based on the information recommendation processing method described in the above steps S22 to S27, that is, inputting the user access sequence and the object identification of the candidate object into the probability prediction model, how to obtain the specific implementation method of the probability that the current user is interested in the target of the candidate object, the present application can train the long-term sample data and the short-term sample data according to this method to obtain the probability prediction model, and the specific training process will not be described in detail.

[0118] Step S28, filtering the recommended objects for the current user from the objects to be recommended according to the interest ranking results;

[0119] Step S29: output the recommended object.

[0120] Regarding step S28, specifically, according to the result of the interest ranking, a certain number of recommended objects with a high interest level of the current user are screened out from the recommended objects, and the objects are fed back to the user client for display, or recommended objects with interest levels reaching a specific threshold are screened out for output. The specific implementation process is not described in detail, and can be referred to but not limited to Figure 2 The illustrated scenario describes the implementation method of the embodiment.

[0121] In summary, in this embodiment, in the recall stage of the recommendation system, this embodiment uses the attention mechanism to obtain the current user's long-term interest in the historical access objects output by the application platform, that is, to analyze the correlation between each candidate object and the historical access objects in a specific time period to obtain the long-term interest; and uses a first-order Markov chain to obtain the current user's short-term interest in the historical access objects output by the application platform, that is, to analyze the correlation between each candidate object and the historical access objects at the previous moment to obtain the short-term interest, and then, the first interest probability representing the long-term interest and the second interest probability representing the short-term interest are weighted and summed to obtain the current user's target interest probability for each candidate object, and accordingly, a specific number of objects to be recommended with a large target interest probability are screened out by the recommendation system. It can be seen that the objects to be recommended take into account the long-term and short-term interests of the user, and improve the comprehensiveness and accuracy of the objects to be recommended recalled by the recommendation system.

[0122] In the sorting stage of the recommendation system, this embodiment combines the attention mechanism to screen the recommended objects that the current user is more interested in in the short term from the recalled objects to be recommended, and feeds them back to the current user. Since the objects to be recommended take into account both the long-term and short-term interests of the user, on this basis, from the perspective of the user's short-term interests, the recommended objects that the user may be interested in at the current moment are selected, which greatly improves the accuracy and reliability of the recommendation.

[0123] Based on the above embodiments, the information recommendation method proposed in the present application, that is, the information recommendation method of the present application, which takes into account both the long-term and short-term interests of users and is implemented using the attention mechanism and the FPMC algorithm, is recorded as the first information recommendation method, and the information recommendation method based on the traditional FPMC algorithm is recorded as the second information recommendation method, and the information recommendation method based on the user's long-term interests and implemented using the FPMC algorithm is recorded as the third information recommendation method. The data is measured on a public movielens dataset, that is, a dataset about movie ratings, and the click data of logged-in users on a community website. The test results are as follows:

[0124] Test results on the movielens dataset:

[0125] Information Recommendation Method topN Accuracy Recall F1 value Second information recommendation method 1 25.25 1.63 3.06 First information recommendation method 1 28.33 1.92 3.60 Third information recommendation method 1 28.01 1.84 3.45 Second information recommendation method 3 22.14 4.27 7.16 First information recommendation method 3 24.55 4.66 7.83 Third information recommendation method 3 24.56 4.63 7.79 Second information recommendation method 5 20.51 6.38 9.73 First information recommendation method 5 22.93 7.03 10.76 Third information recommendation method 5 22.79 6.93 10.63

[0126] Table 1

[0127] As can be seen from Table 1 above, the comparison of the above-listed information recommendation methods can be analyzed from the indicators such as accuracy (i.e. the above-mentioned correct rate), precision, recall, F1 value, etc. The accuracy rate can be the percentage of correctly predicted results in the total samples; the precision rate can be the probability of actually being a positive sample among all the samples predicted to be positive, that is, the probability that the recommended object is the object actually visited by the user, which represents the overall prediction accuracy; the recall rate can be the probability of being predicted as a positive sample among the actually positive samples, that is, the probability that the actually visited object is the object to be recommended; the F1 value, i.e., the F1 score, can be the score of the balance point when the precision rate and the recall rate are considered at the same time and both are maximized at the same time and a balance is achieved. This application does not elaborate on the methods for obtaining these indicators.

[0128] Based on this, no matter how many recommended objects (topN) are finally obtained, the first information recommendation method proposed in this application has the highest accuracy. Even when the top3 is selected, the accuracy of the third information recommendation method is slightly higher than that of the first information recommendation method. However, the recall rate and F1 value of the first information recommendation method are higher than those of the third information recommendation method. Moreover, from the perspective of the F1 value, Figure 5 As shown in the F1 value variation curve, according to the information recommendation method proposed in this application, the recommended objects pushed to users can better meet user needs.

[0129] Similarly, the test results on the click data (such as clicks on posts) of logged-in users on a community website are as follows:

[0130] Information Recommendation Method topN Accuracy Recall F1 value Second information recommendation method 1 18.49 4.55 7.30 First information recommendation method 1 18.76 4.77 7.61 Third information recommendation method 1 18.54 4.68 7.47 Second information recommendation method 3 14.09 10.35 11.93 First information recommendation method 3 14.56 10.36 12.11 Third information recommendation method 3 14.15 10.38 11.98 Second information recommendation method 5 11.93 14.43 13.06 First information recommendation method 5 12.33 14.57 13.36 Third information recommendation method 5 12.11 14.46 13.18

[0131] Table 1

[0132] Combining the data in Table 1, and Figure 6 From the F1 value variation curves of the community secretaries corresponding to the different information recommendation systems shown, it can be seen that according to the information recommendation method proposed in this application, the recommended community secretaries pushed to users can better meet the needs of users.

[0133] Reference Figure 7 , shows a schematic diagram of the structure of an optional example of the information recommendation device proposed in the present application, the device can be applied to a computer device, and the device may include:

[0134] A data acquisition module 11, used to acquire a user access sequence and object identifiers of candidate objects, wherein the user access sequence is composed of object identifiers of objects that the user has historically accessed on the application platform;

[0135] Prediction module 12, used for inputting the user access sequence and the object identification of the candidate object into a probability prediction model to obtain the probability that the current user is interested in the target of the candidate object;

[0136] The probability prediction model is obtained by training long-term sample data and short-term sample data based on the attention mechanism and machine learning algorithm. The long-term sample data refers to the object identifiers of historical access objects generated within a specific time period, and the short-term sample data refers to the object identifiers of historical access objects generated at the previous moment.

[0137] A recommended object obtaining module 13 is used to obtain a recommended object from the candidate objects based on the target interest probability of the candidate objects;

[0138] The recommended object output module 14 is used to output the recommended object.

[0139] In some embodiments, Figure 8 As shown, the prediction module 12 may include:

[0140] The predicted access probability acquisition unit 121 is used to acquire the predicted access probability of accessing the candidate object at the current moment based on the user access sequence and the object identifier of the candidate object;

[0141] In a possible implementation, the predicted access probability acquisition unit 121 may include:

[0142] A probability transfer matrix construction unit, used to construct a probability transfer matrix using the user access sequence and the object identifier of the candidate object;

[0143] The predicted access probability obtaining unit is used to process the probability transfer matrix by matrix decomposition to obtain the predicted access probability of the current user visiting the candidate object at the current moment.

[0144] The correlation analysis unit 122 is used to perform a correlation analysis on the candidate object and the historical access objects generated in a specific time period using the predicted access probability of the candidate object to obtain a first probability of interest of the current user in the candidate object, and to perform a correlation analysis on the candidate object and the historical access objects generated at the previous moment to obtain a second probability of interest of the current user in the candidate object;

[0145] In a possible implementation, the correlation analysis unit 122 may include:

[0146] A first attention degree obtaining unit is used to learn first user access sequences of different users through an attention mechanism to obtain a first attention degree of the current user to the historical access object, wherein the first user access sequence is composed of object identifiers of access objects on the application platform accessed by the corresponding user within a specific time period;

[0147] A first correlation acquisition unit is used to acquire the correlation between the corresponding candidate object and the access object accessed by the current user on the application platform within the specific time period according to the predicted access probability of the candidate object and the first attention level, so as to obtain a first probability of interest of the current user in the corresponding candidate object;

[0148] A second attention degree obtaining unit is used to analyze second user access sequences of different users to obtain a second attention degree of the current user to the historical access object, wherein the second user access sequence is composed of object identifiers of access objects accessed by the corresponding user at the previous moment to the application platform;

[0149] The second correlation acquisition unit is used to obtain the correlation between the corresponding candidate object and the access object visited by the current user on the application platform at the last moment based on the predicted access probability of the candidate object and the second attention level, so as to obtain the second probability of interest of the current user in the corresponding candidate object.

[0150] The target interest probability obtaining unit 123 is used to perform weighted summation of the first interest probabilities and the second interest probabilities corresponding to different candidate objects, to obtain the target interest probability of the current user for the corresponding candidate object.

[0151] Usually, the number of candidate objects obtained by pre-screening is multiple. In this case, if Fig. 9 As shown, the above-mentioned recommended object obtaining module 13 may include:

[0152] A to-be-recommended object selection unit 131 is used to select a specific number of candidate objects with a relatively high probability of being of interest to the target from the plurality of candidate objects as to-be-recommended objects;

[0153] A sorting unit 132 is used to obtain a sorting result of the current user's interest in the object to be recommended based on an attention mechanism and a Bayesian personalized sorting algorithm, using the object identifier of the object accessed on the application platform at the last moment;

[0154] The recommended object screening unit 133 is used to screen the recommended objects for the current user from the objects to be recommended according to the interest level sorting result.

[0155] It should be noted that the various modules, units, etc. in the above-mentioned device embodiments can be stored in the memory as program modules, and the processor executes the above-mentioned program modules stored in the memory to implement the corresponding functions. For the functions implemented by each program module and its combination, as well as the technical effects achieved, please refer to the description of the corresponding parts of the above-mentioned method embodiments, which will not be repeated in this embodiment.

[0156] The present application also provides a storage medium on which a program can be stored. The program can be called and loaded by a processor to implement the various steps of the information recommendation method described in the above embodiment.

[0157] Reference Fig.10 , shows a schematic diagram of the hardware structure of a computer device proposed in an embodiment of the present application, the computer device may include: at least one communication interface 21, at least one memory 22 and at least one processor 23, wherein:

[0158] The communication interface 21 can be an interface of a communication module, such as an interface of a GSM module, a WIFI module, a GPRS module, etc., which can realize data interaction with other devices. It can also include interfaces such as USB interface, serial / parallel port, etc., which are used to realize data interaction between internal components of the computer device. It can be determined according to the product type of the computer device, and this application will not describe them in detail one by one.

[0159] The memory 22 is used to store a program for implementing the information recommendation method described in any of the above method embodiments; the processor 23 is used to load and execute the program stored in the memory 22 to implement the various steps of the information recommendation method described in the above corresponding method embodiment. The specific implementation process can refer to the description of the corresponding part of the above embodiment.

[0160] In the embodiment of the present application, the memory 22 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device. The processor 23 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices.

[0161] In one possible implementation, the memory 22 may include a program storage area and a data storage area. The program storage area may store an operating system, an application required for at least one function, a program for implementing the information recommendation method proposed in the present application, etc.; the data storage area may store data generated during the use of the computer device, such as historical access data of users visiting the community website, candidate objects, recommended objects, etc.

[0162] It should be understood that Fig.10 The structure of the computer device shown does not constitute a limitation on the computer device in the embodiment of the present application. In practical applications, the computer device may include Fig.10 More or fewer components as shown, or combinations of certain components, are not listed here one by one in this application.

[0163] Finally, it should be noted that the various embodiments in this specification are described in a progressive or parallel manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0164] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An information recommendation method, the method comprising: Acquire a user access sequence and an object identifier of a candidate object, wherein the user access sequence is composed of object identifiers of objects that the user has historically accessed on the application platform; The user access sequence and the object identifier of the candidate object are input into a probability prediction model to obtain the probability that the current user is interested in the target of the candidate object, wherein the probability prediction model is obtained by training long-term sample data and short-term sample data based on an attention mechanism and a machine learning algorithm, wherein the long-term sample data refers to the object identifier of the historical access object generated within a time period, and the short-term sample data refers to the object identifier of the historical access object generated at the previous moment; wherein the machine learning algorithm includes an FPMC algorithm; Based on the target interest probability of the candidate objects, obtaining recommended objects among the candidate objects; The recommended object is output.

2. The method according to claim 1, wherein the step of inputting the user access sequence and the object identifier of the candidate object into a probability prediction model to obtain the probability that the current user is interested in the target of the candidate object comprises: Based on the user access sequence and the object identifier of the candidate object, obtaining a predicted access probability of accessing the candidate object at a current moment; Using the predicted access probability of the candidate object, a correlation analysis is performed on the candidate object and the historical access objects generated within a time period to obtain a first probability of interest of the current user in the candidate object, and a correlation analysis is performed on the candidate object and the historical access objects generated at the previous moment to obtain a second probability of interest of the current user in the candidate object; The first interest probabilities and the second interest probabilities corresponding to different candidate objects are weightedly summed to obtain the target interest probabilities of the current user for the corresponding candidate objects.

3. The method according to claim 2, wherein the predicted access probability of the candidate object is used to perform a correlation analysis between the candidate object and the historical access objects generated within a time period to obtain the first probability of the current user being interested in the candidate object, comprising: The first user access sequence of different users is learned through an attention mechanism to obtain a first attention degree of the current user to the historical access object, wherein the first user access sequence is composed of object identifiers of access objects accessed by the corresponding user on the application platform within a time period; According to the predicted access probability of the candidate object and the first attention level, the correlation between the corresponding candidate object and the access object accessed by the current user on the application platform within the time period is obtained to obtain a first probability of interest of the current user in the corresponding candidate object; The step of using the predicted access probability of the candidate object to perform a correlation analysis on the candidate object and the historical access object generated at the last moment to obtain a second probability of interest of the current user in the candidate object includes: Analyzing second user access sequences of different users to obtain a second degree of attention of the current user to the historical access object, wherein the second user access sequence is composed of object identifiers of access objects accessed by the corresponding user at the last moment on the application platform; According to the predicted access probability of the candidate object and the second attention level, the correlation between the corresponding candidate object and the access object visited by the current user on the application platform at the last moment is obtained to obtain the second interest probability of the current user in the corresponding candidate object.

4. The method according to any one of claims 1 to 3, wherein the number of candidate objects is multiple, and obtaining the recommended objects from the candidate objects based on the target interest probabilities of the candidate objects comprises: From the plurality of candidate objects, selecting a first number of candidate objects whose target interest probability is greater than a probability threshold as objects to be recommended; Based on the attention mechanism and the Bayesian personalized ranking algorithm, the object identifier of the object accessed on the application platform at the last moment is used to obtain the ranking result of the current user's interest in the object to be recommended; According to the interest ranking result, the recommended objects for the current user are screened from the objects to be recommended.

5. The method according to claim 2, wherein obtaining the predicted access probability of accessing the candidate object at the current moment based on the user access sequence and the object identifier of the candidate object comprises: constructing a probability transfer matrix using the user access sequence and the object identifiers of the candidate objects; The probability transfer matrix is ​​processed by matrix decomposition to obtain the predicted access probability of the current user accessing the candidate object at the current moment.

6. An information recommendation device, comprising: A data acquisition module, used to acquire a user access sequence and object identifiers of candidate objects, wherein the user access sequence is composed of object identifiers of objects that the user has historically accessed on the application platform; A prediction module, used for inputting the user access sequence and the object identifier of the candidate object into a probability prediction model to obtain the probability that the current user is interested in the target of the candidate object, wherein the probability prediction model is obtained by training long-term sample data and short-term sample data based on an attention mechanism and a machine learning algorithm, wherein the long-term sample data refers to the object identifier of the historical access object generated within a time period, and the short-term sample data refers to the object identifier of the historical access object generated at the previous moment; wherein the machine learning algorithm includes an FPMC algorithm; A recommended object obtaining module, used for obtaining a recommended object from the candidate objects based on the target interest probability of the candidate objects; The recommended object output module is used to output the recommended object.

7. The device according to claim 6, wherein the prediction module comprises: A predicted access probability acquisition unit, used to acquire a predicted access probability of accessing the candidate object at a current moment based on the user access sequence and the object identifier of the candidate object; A correlation analysis unit is used to use the predicted access probability of the candidate object to perform a correlation analysis on the candidate object and the historical access objects generated within a time period to obtain a first probability of interest of the current user in the candidate object, and to perform a correlation analysis on the candidate object and the historical access objects generated at a previous moment to obtain a second probability of interest of the current user in the candidate object; The target interest probability obtaining unit is used to perform weighted summation of the first interest probabilities and the second interest probabilities corresponding to different candidate objects respectively, to obtain the target interest probability of the current user for the corresponding candidate object.

8. A storage medium having a program stored thereon, wherein the program is called and executed by a processor to implement the information recommendation method according to any one of claims 1 to 5.

9. A computer device, comprising: At least one communication interface, at least one memory, and at least one processor, wherein: The memory is used to store a program for implementing the information recommendation method according to any one of claims 1 to 5; The processor is used to load and execute the program stored in the memory to implement each step of the information recommendation method according to any one of claims 1 to 5.

Images