Information pushing method and device, equipment, medium and program product

Abstract

The invention provides an information pushing method and device, relates to the field of artificial intelligence, and can be applied to the technical field of financial science and technology. The method comprises the steps that n pieces of first user information graph data corresponding to n target genera are obtained, users in a first user set serve as nodes of the n pieces of first user information graph data, and connection edges are established between the users with the similarity of corresponding target attributes meeting a preset similarity condition in each piece of first user information graph data; combining the n pieces of first user information graph data, and taking a first historical operation behavior sequence of each user as a node feature of a node corresponding to each user to obtain a first time sequence graph; and inputting the first time sequence diagram to the trained information recommendation model, obtaining a prediction result output by the information recommendation model, and pushing information to the user based on the prediction result. The invention further provides an information recommendation model construction method and device, equipment, a medium and a program product.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence and can be used in the field of financial technology or other fields. More specifically, it relates to an information push method, device, equipment, medium and program product. Background Art

[0002] Effective information push provides great support for meeting user needs, improving product conversion rates, and enhancing user experience. Currently, many information recommendation solutions either find similar users of the target user and then recommend to the target user based on the preferences of similar users, or find similar content to the target user's favorite content for recommendation, or predict the target user's future preferences based on his past behavior and then make recommendations. However, these information recommendation solutions generally consider a relatively single dimension and cannot achieve accurate recommendations based on user characteristics in many aspects and from all angles. Summary of the invention

[0003] In view of the above problems, the present disclosure provides an information push method and apparatus that can improve the accuracy of information push, a method and apparatus for constructing an information recommendation model, and a device, a medium, and a program product.

[0004] According to a first aspect of the present disclosure, an information push method is provided. The method includes: obtaining n first user information graph data, wherein the n first user information graph data all use users in a first user set as nodes, and the n first user information graph data correspond to n target attributes one by one, wherein in each first user information graph data, a connection edge is established between users whose similarity of the corresponding target attribute meets a preset similarity condition, and n is an integer greater than 1; obtaining a first historical operation behavior sequence of each user in the first user set, wherein the first historical operation behavior sequence includes information operation behavior data of the user in m time units, and m is an integer greater than 1; combining the n first user information graph data, and using the first historical operation behavior sequence of each user in the first user set as a node feature of a node corresponding to each user to obtain a first time series graph; and inputting the first time series graph into a trained information recommendation model, and obtaining a prediction result output by the information recommendation model, wherein the prediction result includes information operation behavior data of each user in the first user set in the next s time units; wherein the information recommendation model is constructed using a graph neural network; wherein s is an integer greater than or equal to 1 and less than m; and pushing information to users in the first user set based on the prediction result.

[0005] According to an embodiment of the present disclosure, the combination of n first user information graph data includes: obtaining weights of n target attributes; using the weight of each target attribute as the weight of the first user information graph data corresponding to the target attribute, and performing weighted fusion on the adjacency matrix of the n first user information graph data based on the respective weights of the n first user information graph data.

[0006] According to an embodiment of the present disclosure, obtaining the weights of n target attributes includes: extracting the weights of n target attributes from the training results of the information recommendation model, wherein the weights of the n target attributes are parameters optimized during the training process of the information recommendation model.

[0007] According to an embodiment of the present disclosure, obtaining n first user information graph data includes: obtaining attribute values ​​of users in the first user set on n target attributes; for each target attribute, classifying the first user set based on the attribute value of each user on the target attribute, and establishing connecting edges among users classified into the same category, wherein the similarity condition includes that users belong to the same category.

[0008] According to an embodiment of the present disclosure, the construction process of the information recommendation model includes: obtaining n second user information graph data corresponding one by one to n target attributes, wherein the n second user information graph data all use users in the second user set as nodes, and in each second user information graph data, establish connecting edges between users whose similarity of the corresponding target attributes meets the preset similarity conditions; obtaining the second historical operation behavior sequence of each user in the second user set, wherein the second historical operation behavior sequence includes the information operation behavior data of the user in m time units; obtaining a sequence composed of the information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence, and obtaining the label information of each user; combining the n second user information graph data, and using the second historical operation behavior sequence of each user in the second user set as the node feature of the node corresponding to each user, to obtain a second time series graph; using the second time series graph as the training input and the label information of each user in the second user set as the training output, training the graph neural network, wherein the graph neural network after training is used as the information recommendation model.

[0009] According to an embodiment of the present disclosure, the combination of n second user information graph data includes: taking the weight of each target attribute as the weight of the second user information graph data corresponding to the target attribute, and weightedly fusing the adjacency matrices of the n second user information graph data based on the respective weights of the n second user information graph data; taking the second time series graph as training input and the label information of each user in the second user set as training output, training the graph neural network includes synchronously updating the parameters of the graph neural network and the weights of the n target attributes in each round of iterative training; and weightedly fusing the adjacency matrices of the n second user information graph data using the updated weights of the n target attributes to update the second time series graph.

[0010] The second aspect of the disclosed embodiment provides a method for constructing an information recommendation model. The method includes: obtaining n second user information graph data; wherein the n second user information graph data are all nodes of the users in the second user set, and the n second user information graph data correspond to n target attributes one by one, wherein in each second user information graph data, a connection edge is established between users whose similarity of the corresponding target attributes meets the preset similarity condition, and n is an integer greater than 1; obtaining the second historical operation behavior sequence of each user in the second user set, the second historical operation behavior sequence includes the information operation behavior data of the user in m time units; obtaining a sequence composed of the information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence, and obtaining the label information of each user; combining the n second user information graph data, and using the second historical operation behavior sequence of each user in the second user set as the node feature of the node corresponding to each user, to obtain a second time series graph; and using the second time series graph as the training input and the label information of each user in the second user set as the training output to train the graph neural network, wherein the graph neural network after training is used as the information recommendation model.

[0011] A third aspect of the embodiments of the present disclosure provides an information push device, which includes: a first image data acquisition module, a first sequence data acquisition module, a first multi-image combination module, a prediction module, and an information push module.

[0012] The first graph data acquisition module is used to acquire n first user information graph data, wherein the n first user information graph data all have users in the first user set as nodes, and the n first user information graph data correspond one-to-one to n target attributes, wherein in each first user information graph data, a connection edge is established between users whose similarity of the corresponding target attributes meets a preset similarity condition, and n is an integer greater than 1.

[0013] The first sequence data acquisition module is used to acquire a first historical operation behavior sequence of each user in the first user set, where the first historical operation behavior sequence includes information operation behavior data of the user in m time units, where m is an integer greater than 1.

[0014] The first multi-graph combination module is used to combine n first user information graph data, and use the first historical operation behavior sequence of each user in the first user set as the node feature of the node corresponding to each user to obtain a first time series graph.

[0015] The prediction module includes a trained information recommendation model, wherein the prediction module is used to input the first time series graph into the information recommendation model and obtain the prediction result output by the information recommendation model, wherein the prediction result includes the information operation behavior data of each user in the first user set in the next s time units; wherein the information recommendation model is constructed using a graph neural network; wherein s is an integer greater than or equal to 1 and less than m.

[0016] The information push module is used to push information to users in the first user set based on the prediction result.

[0017] According to an embodiment of the present disclosure, the first multi-graph combination module further includes: a weight acquisition submodule and a combination submodule. The weight acquisition submodule is used to obtain the weights of n target attributes. The combination submodule is used to use the weight of each target attribute as the weight of the first user information graph data corresponding to the target attribute, and perform weighted fusion on the adjacency matrix of the n first user information graph data based on the respective weights of the n first user information graph data.

[0018] According to an embodiment of the present disclosure, the weight acquisition submodule is specifically used to extract weights of n target attributes from the training results of the information recommendation model, wherein the weights of the n target attributes are parameters optimized during the training process of the information recommendation model.

[0019] According to an embodiment of the present disclosure, the first graph data acquisition module is specifically used to: obtain attribute values ​​of users in the first user set on n target attributes; for each target attribute, classify the first user set based on the attribute value of each user on the target attribute, and establish connecting edges among users classified into the same category, wherein the similarity condition includes that the users belong to the same category.

[0020] A fourth aspect of the embodiments of the present disclosure provides a device for constructing an information recommendation model, which includes: a second graph data acquisition module, a second sequence data acquisition module, a second multi-graph combination module, a labeling module, and a training module.

[0021] The second graph data acquisition module is used to acquire n second user information graph data; wherein the n second user information graph data all have users in the second user set as nodes, and the n second user information graph data correspond one-to-one to n target attributes, wherein in each second user information graph data, a connection edge is established between users whose similarity of the corresponding target attributes meets a preset similarity condition, and n is an integer greater than 1.

[0022] The second sequence data acquisition module is used to acquire a second historical operation behavior sequence of each user in the second user set, where the second historical operation behavior sequence includes information operation behavior data of the user over m time units.

[0023] The labeling module is used to obtain a sequence consisting of information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence, and obtain label information of each user.

[0024] The second multi-graph combination module is used to combine n second user information graph data, and use the second historical operation behavior sequence of each user in the second user set as the node feature of the node corresponding to each user to obtain a second time series graph.

[0025] The training module is used to train the graph neural network using the second time series graph as training input and the label information of each user in the second user set as training output, wherein the graph neural network after training is used as the information recommendation model.

[0026] A third aspect of an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0027] The fourth aspect of the embodiments of the present disclosure further provides a computer-readable storage medium on which a computer program or instructions are stored. When the computer program or instructions are executed by a processor, the steps of the above method are implemented.

[0028] The fifth aspect of the embodiments of the present disclosure further provides a computer program product, including a computer program or instructions, which implement the steps of the above method when the above computer program or instructions are executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0030] Figure 1The application scenario diagram of the information push method, apparatus, device, medium and program product according to the embodiments of the present disclosure is schematically shown;

[0031] Figure 2 The flowchart of the information push method according to an embodiment of the present disclosure is schematically shown;

[0032] Figure 3 The model architecture of the information recommendation model of an embodiment of the present disclosure is schematically shown;

[0033] Figure 4 A flowchart schematically shows a method for constructing an information recommendation model according to an embodiment of the present disclosure;

[0034] Figure 5 A block diagram of an information push device according to an embodiment of the present disclosure is schematically shown;

[0035] Figure 6 A block diagram schematically shows a device for constructing an information recommendation model according to an embodiment of the present disclosure; and

[0036] Figure 7 A block diagram of an electronic device suitable for implementing the information push method or the information recommendation model construction method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0037] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0038] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0039] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having meanings consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner. The terms "first" and "second" in this article are only used to distinguish the terms and do not have a limiting effect.

[0040] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0041] The disclosed embodiments provide an information push method, apparatus, device, medium and program product, which can first construct corresponding user information graph data according to different target attributes, wherein in each user information graph data, users with similarity determined based on the corresponding target attribute are connected through edges, so that the connection relationship in different user information graph data can reflect the similarity of user groups under different target attributes, and at the same time, the user's historical operation behavior sequence is used as the node feature of each user node in the user information graph data to form the input data of the information recommendation model, so that the information recommendation model can comprehensively consider the user similarities on multiple attributes and the historical operation behavior sequence of the user group, and recommend the future information operation behavior of each user, so as to utilize user characteristics in multiple dimensions, improve the accuracy and effectiveness of information push, and enhance the user experience.

[0042] Figure 1 The application scenario diagram of the information push method, apparatus, device, medium and program product according to the embodiments of the present disclosure is schematically shown.

[0043] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0044] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0045] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0046] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device. In one embodiment, an information push model may be deployed in the server 105, and information may be pushed to the first terminal device 101, the second terminal device 102, and the third terminal device 103.

[0047] It should be noted that the information push method or information recommendation model construction method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the information push device or information recommendation model construction device provided in the embodiment of the present disclosure can generally be set in the server 105. The information push method or information recommendation model construction method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the information push device or information recommendation model construction device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0048] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0049] It should be noted that the methods, devices, equipment, media and program products of the embodiments of the present disclosure can be used in the field of financial technology, and can also be used in any field other than the field of financial technology. The present disclosure does not limit the application field.

[0050] In the technical solution of the embodiments of the present disclosure, the user information involved (including but not limited to user personal information, user operation behavior data, various attribute data, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good customs, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0051] The following will be based on Figure 1 The described scene, combined with Figure 2 ~ and 4 describe in detail the information push method of each embodiment of the present disclosure.

[0052] Figure 2 The flowchart of the information push method according to an embodiment of the present disclosure is schematically shown.

[0053] like Figure 2 As shown, the information push method according to this embodiment may include operations S210 to S250.

[0054] In operation S210, n first user information graph data are obtained, wherein the n first user information graph data all have users in the first user set as nodes, and the n first user information graph data correspond one-to-one to n target attributes, wherein in each first user information graph data, a connection edge is established between users whose similarity of the corresponding target attributes meets a preset similarity condition, and n is an integer greater than 1.

[0055] In the implementation of the present disclosure, similarity calculation can be performed on each target attribute separately to set connection edges, and then a connection graph between users (i.e., user information graph data) can be formed for each target attribute.

[0056] The n target attributes may include, but are not limited to, various static attributes such as the user's region, gender, age, education level, work information, and the number of debit and credit cards applied for.

[0057] The similarity of users on the target attribute can be measured in a variety of ways, such as the difference in attribute values ​​(such as numerical values ​​or codes) of different users on the target attribute, distance (such as Euclidean distance) or cosine similarity, etc. Accordingly, the similarity condition can be set accordingly according to the similarity measurement method.

[0058] In one embodiment, for each target attribute, users are classified based on the attribute value of each user on the target attribute, and connecting edges are established among users classified into the same category to obtain a user information graph. In this embodiment, the similarity condition is that the users belong to the same category. For example, for attributes such as age and gender, the attribute values ​​are directly represented by 0 and 1. For continuous attributes such as age, the attributes can be first converted into discrete attributes by binning, and then the attribute values ​​after conversion into discrete attributes are used to classify the users. Among them, binning is a data preprocessing technology that involves mapping continuous feature values ​​into several bins to discretize continuous data. For regional attributes, they can be classified according to first-, second-, and third-tier cities. For work information attributes, they can be classified according to the attributes of occupations. The method of determining the similarity of users by classification is simple and easy to implement, and the amount of calculation is small.

[0059] In operation S220, a first historical operation behavior sequence of each user in the first user set is obtained, where the first historical operation behavior sequence includes information operation behavior data of the user in m time units, where m is an integer greater than 1. The m time units may be m moments with equal intervals, or may be m time windows with continuous or fixed intervals.

[0060] For example, historical behaviors of users are collected over multiple time units, including but not limited to clicks and browsing behaviors of users on various types or contents of information (such as advertisements and links) pushed in the past, operation records of users purchasing financial products and fund products, operation records of users handling various banking services, etc., to form a historical operation behavior sequence X.

[0061] It should be noted that, in the embodiments of the present disclosure, before obtaining various static attributes and information operation behavior data of a user, authorization to use the corresponding information will be requested from the user, and data collection and acquisition will be performed only after obtaining the user's authorization.

[0062] In operation S230, n first user information graph data are combined, and the first historical operation behavior sequence of each user in the first user set is used as the node feature of the node corresponding to each user to obtain a first time series graph. In order to better represent the time characteristics and user information characteristics of the data, the embodiment of the present disclosure can combine multiple user information graphs and form a time series graph together with the historical operation behavior sequence of each user.

[0063] In one embodiment, combining the n first user information graph data may be concatenating the n first user information graph data in a preset order to form a group of user information graph data.

[0064] In another embodiment, combining the n first user information graph data may be fusing the n first user information graph data into one graph data, such as weighted summing of adjacency matrices of the n first user information graph data to form a weighted undirected graph.

[0065] In one embodiment, when combining n first user information graph data, different weights may be assigned to different first user information graph data. The weight of each target attribute may be used to determine the weight of the first user information graph data corresponding to the target attribute. The weight of each target attribute may reflect the degree of influence of the similarity of each attribute on the similarity of user behavior. By assigning different weights to different first user information graph data, differential combinations of different first user information graph data may be achieved.

[0066] The weight of each target attribute may be obtained based on experience or through statistical analysis, or may be trained. For example, when training an information recommendation model, the weights of the n target attributes are used as parameters to be optimized during the training process.

[0067] It can be seen that the disclosed embodiment can map each user to a point on a graph, and can simply classify the users according to different attributes, such as the user's region, gender, age, education level, work information, and the number of debit and credit cards applied for, etc. Then, the users of the same category under each attribute can be connected to form a weighted undirected graph G=(V,E,A) composed of each user. Where V represents the set of nodes on the graph, each node is a user; E represents the set of edges on the graph, and each edge connects two users. A is the adjacency matrix on the graph, which represents the adjacency relationship of each node on the graph. The adjacency relationship represents the similarity between two users in a certain type of user information. The similarity has a certain weight, which is the weight of the target attribute.

[0068] Next, in operation S240, the first time series graph is input into the trained information recommendation model, and the prediction result output by the information recommendation model is obtained, wherein the prediction result includes the information operation behavior data of each user in the first user set in the next s time units; wherein the information recommendation model is constructed using a graph neural network; wherein s is an integer greater than or equal to 1 and less than m.

[0069] Finally, in operation S250, information is pushed to users in the first user set based on the prediction result.

[0070] It can be seen that the disclosed embodiment abstracts the problem of accurate information push into a prediction problem of a graph neural network. By using the time series graph as the input of the information recommendation model, each user's historical operation behavior sequence can be used as the input of the time series node, and the user's multiple attributes are used as the basis for user similarity division, so that the information recommendation model can predict the user's future preferences based on the user's similarity in different attributes and the user's past behavior operations. Figure 3 The model architecture of the information recommendation model according to an embodiment of the present disclosure is schematically shown.

[0071] Combination Figure 3 In constructing the information recommendation model, the disclosed embodiment not only considers the time dependency, but also the dependency between users on multiple attributes. Thus, an improved model of Long Short-Term Memory (LSTM) network based on graph neural network is used to construct LSTM for each node and edge on the user information graph data. At the same time, the encoder-decoder framework is used to decode the historical operation behavior sequence ( , and The decoder predicts multiple time units in the future and outputs , It is the user’s information operation behavior data in the future time unit L-1 and time unit L. At the same time, in order to improve the accuracy of prediction, an attention mechanism is added to the graph neural network, so that the encoder can extract effective information from the combination of multiple user information graphs, thereby optimizing the model.

[0072] for Figure 3 The encoder-decoder model in the example is mainly to combine the information of the historical sequence, the combination of multiple user information graphs and the attention mechanism to encode the historical sequence information, so as to make subsequent decoding predictions. The decoder part predicts the preferences of future users based on the learning content of the encoder. In one embodiment, the construction process of the information recommendation model can refer to Figure 4 shown.

[0073] Figure 4 The flowchart of the method for constructing an information recommendation model according to an embodiment of the present disclosure is schematically shown. Figure 4 As shown, according to this embodiment, the method for constructing an information recommendation model may include operations S410 to S480.

[0074] First, in operation S410, n second user information graph data corresponding one-to-one to the n target attributes are obtained, wherein the n second user information graph data all use users in the second user set as nodes, and in each second user information graph data, connection edges are established between users whose similarities of the corresponding target attributes meet preset similarity conditions.

[0075] And in operation S420, a second historical operation behavior sequence of each user in the second user set is obtained, where the second historical operation behavior sequence includes information operation behavior data of the user over m time units.

[0076] Then, in operation S430, a sequence consisting of information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence is obtained to obtain label information of each user.

[0077] Next, the graph neural network is trained through a loop iteration of operation S440 to operation S470.

[0078] Specifically, in operation S440, the weight of each target attribute is used as the weight of the second user information graph data corresponding to the target attribute, the adjacency matrix of the n second user information graph data is weightedly fused based on the respective weights of the n second user information graph data, and the second historical operation behavior sequence of each user in the second user set is used as the node feature of the node corresponding to each user to obtain a second time series graph.

[0079] In operation S450, the graph neural network is trained using the second time series graph as a training input and the label information of each user in the second user set as a training output.

[0080] Then, in operation S460, after each round of training is completed, it is determined whether the iterative training termination condition is met. If not, operation S470 is performed; if yes, operation S480 is performed. The iterative training termination condition may be, for example, that the cross entropy loss function that measures the performance of the model during the training process is minimized, or that the number of iterations reaches a preset threshold.

[0081] In operation S470, the parameters of the graph neural network and the weights of the n target attributes are updated synchronously. Then, the process returns to operation S440 to perform weighted fusion on the adjacency matrix of the n second user information graph data using the updated weights of the n target attributes to update the second time series graph, and perform a new round of iterative training on the graph neural network with updated parameters.

[0082] In operation S480, when the iterative training termination condition is reached, the trained graph neural network is used as the information recommendation model.

[0083] In this embodiment, during the training of the graph neural network, the weights of the n target attributes are also updated as training parameters, so that the second time series graph can be continuously iterated and updated during the training of the graph neural network. In this way, the influence of attributes that are more closely related to user behavior can be continuously amplified, while the influence of attributes that are not closely related to user behavior can be reduced, thereby reducing the interference of attributes that are not closely related to user behavior prediction. In this way, when the model predicts user behavior, the importance of different user information graphs in the model input data is adjusted by the weight of the target attribute, and the model performance is further optimized, making the model prediction more accurate. In the related art, multiple attributes of the user are often collected, and then the similarities between users are distinguished based on the merge processing of the multiple attributes. In this way, the degree of association of multiple attributes with user behavior cannot be effectively distinguished, and thus more targeted information cannot be provided when predicting user behavior as in the embodiment of the present disclosure.

[0084] The disclosed embodiment can more accurately capture the relationship between different attributes of users and user behaviors by combining different weights through a multi-user information graph, rather than analyzing user similarities by combining all user attributes together and then connecting similar users as in the related art, which makes it impossible to find out which attributes of users with similar behaviors have a greater impact and which attributes may actually have little impact.

[0085] It can be seen that the information recommendation model constructed by the embodiment of the present disclosure can comprehensively consider the user similarities on multiple attributes and the historical operation behavior sequences of the user group to predict the future information operation behavior of each user, thereby making use of user characteristics in multiple dimensions, improving the accuracy and effectiveness of information push, and enhancing the user experience.

[0086] The disclosed embodiments can extract multi-angle and comprehensive user information, including multiple target attributes of users (such as basic information or static attribute information) and historical operation behavior information of users (such as past click records, browsing records, purchase records of financial products and other operation records of users on advertisements), and then use the learning method of graph neural networks to predict user behavior, so as to push more accurate and pain-point information to users (such as advertisements, activity links, financial services, etc.).

[0087] Figure 5 A block diagram of an information push device according to an embodiment of the present disclosure is schematically shown.

[0088] like Figure 5 As shown, according to this embodiment, the information pushing device 500 may include a first image data acquisition module 510 , a first sequence data acquisition module 520 , a first multi-image combination module 530 , a prediction module 540 and an information pushing module 550 .

[0089] The first graph data acquisition module 510 is used to acquire n first user information graph data, wherein the n first user information graph data all have users in the first user set as nodes, and the n first user information graph data correspond one-to-one to n target attributes, wherein in each first user information graph data, a connection edge is established between users whose similarity of the corresponding target attribute meets a preset similarity condition, and n is an integer greater than 1. In one embodiment, the first graph data acquisition module 510 can perform the operation S210 described above.

[0090] The first sequence data acquisition module 520 is used to acquire the first historical operation behavior sequence of each user in the first user set, where the first historical operation behavior sequence includes the information operation behavior data of the user in m time units, where m is an integer greater than 1. In one embodiment, the first sequence data acquisition module 520 may perform the operation S220 described above.

[0091] The first multi-graph combination module 530 is used to combine n first user information graph data, and use the first historical operation behavior sequence of each user in the first user set as the node feature of the node corresponding to each user to obtain a first time series graph. In one embodiment, the first multi-graph combination module 530 can perform the operation S230 described above.

[0092] The prediction module 540 includes a trained information recommendation model, wherein the prediction module 540 is used to input the first time series graph into the information recommendation model and obtain the prediction result output by the information recommendation model, wherein the prediction result includes the information operation behavior data of each user in the first user set in the next s time units; wherein the information recommendation model is constructed using a graph neural network; wherein s is an integer greater than or equal to 1 and less than m. In one embodiment, the prediction module 540 can perform the operation S240 described above.

[0093] The information push module 550 is used to push information to users in the first user set based on the prediction result. In one embodiment, the information push module 550 may perform the operation S250 described above.

[0094] The information push device 500 can execute reference Figure 2~Figure 4 For the information push method introduced, please refer to the previous introduction, which will not be repeated here.

[0095] Figure 6 A block diagram schematically shows a device for constructing an information recommendation model according to an embodiment of the present disclosure.

[0096] like Figure 6As shown, the information recommendation model construction device 600 may include a second graph data acquisition module 610, a second sequence data acquisition module 620, a second multi-graph combination module 630, a labeling module 640 and a training module 650. In one embodiment, the information recommendation model construction device 600 may be integrated into the information push device 500.

[0097] The second graph data acquisition module 610 is used to acquire n second user information graph data; wherein the n second user information graph data all use users in the second user set as nodes, and the n second user information graph data correspond to n target attributes one by one, wherein in each second user information graph data, a connection edge is established between users whose similarity of the corresponding target attribute meets a preset similarity condition, and n is an integer greater than 1. In one embodiment, the second graph data acquisition module 610 can perform the operation S410 described above.

[0098] The second sequence data acquisition module 620 is used to acquire the second historical operation behavior sequence of each user in the second user set, and the second historical operation behavior sequence includes the information operation behavior data of the user in m time units. In one embodiment, the second sequence data acquisition module 620 can perform the operation S420 described above.

[0099] The second multi-graph combination module 630 is used to combine n second user information graph data, and use the second historical operation behavior sequence of each user in the second user set as the node feature of the node corresponding to each user to obtain a second time series graph. In one embodiment, the second multi-graph combination module 630 can perform the operation S440 described above.

[0100] The marking module 640 is used to obtain a sequence of information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence, and obtain label information of each user. In one embodiment, the marking module 640 can perform the operation S430 described above.

[0101] The training module 650 is used to train the graph neural network using the second time series graph as training input and the label information of each user in the second user set as training output, wherein the trained graph neural network is used as the information recommendation model. In one embodiment, the training module 650 can perform the operations S450, S460, S470, and S480 described above.

[0102] In one embodiment, the second multi-graph combination module 630 is also used to: use the weight of each target attribute as the weight of the second user information graph data corresponding to the target attribute, and perform weighted fusion on the adjacency matrix of the n second user information graph data based on the respective weights of the n second user information graph data. Accordingly, the training module 650 is also used to synchronously update the parameters of the graph neural network and the weights of the n target attributes in each round of iterative training of the graph neural network, and call the second multi-graph combination module 630 to perform weighted fusion on the adjacency matrix of the n second user information graph data using the updated weights of the n target attributes to update the second time series graph.

[0103] The information recommendation model construction device 600 can perform reference Figure 4 The construction method of the information recommendation model introduced in this article is specifically described in the previous article and will not be repeated here.

[0104] According to an embodiment of the present disclosure, any multiple modules among the first graph data acquisition module 510, the first sequence data acquisition module 520, the first multi-graph combination module 530, the prediction module 540, the information push module 550, the second graph data acquisition module 610, the second sequence data acquisition module 620, the second multi-graph combination module 630, the annotation module 640 and the training module 650 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the first graph data acquisition module 510, the first sequence data acquisition module 520, the first multi-graph combination module 530, the prediction module 540, the information push module 550, the second graph data acquisition module 610, the second sequence data acquisition module 620, the second multi-graph combination module 630, the labeling module 640 and the training module 650 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or may be implemented by hardware or firmware such as any other reasonable way of integrating or packaging circuits, or may be implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any of them. Alternatively, at least one of the first image data acquisition module 510, the first sequence data acquisition module 520, the first multi-image combination module 530, the prediction module 540, the information push module 550, the second image data acquisition module 610, the second sequence data acquisition module 620, the second multi-image combination module 630, the labeling module 640 and the training module 650 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is executed.

[0105] Figure 7 A block diagram of an electronic device suitable for implementing the information push method or the information recommendation model construction method according to an embodiment of the present disclosure is schematically shown.

[0106] like Figure 7 As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 to a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0107] In RAM 903, various programs and data required for the operation of electronic device 900 are stored. Processor 901, ROM 902 and RAM 903 are connected to each other via bus 904. Processor 901 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 902 and / or RAM 903. It should be noted that the program can also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.

[0108] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the input / output (I / O) interface 905: an input portion 906 including a keyboard, a mouse, etc.; an output portion 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 908 including a hard disk, etc.; and a communication portion 909 including a network interface card such as a LAN card, a modem, etc. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage portion 908 as needed.

[0109] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0110] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0111] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiment of the present disclosure.

[0112] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 901. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0113] In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of a signal on a network medium, and downloaded and installed through the communication part 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0114] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0115] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0116] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0117] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0118] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for pushing information, wherein: The method comprises: Obtain n first user information graph data, wherein the n first user information graph data all have users in the first user set as nodes, and the n first user information graph data correspond one-to-one to n target attributes, wherein in each first user information graph data, a connection edge is established between users whose similarity of the corresponding target attribute meets a preset similarity condition, and n is an integer greater than 1; Acquire a first historical operation behavior sequence of each user in the first user set, where the first historical operation behavior sequence includes information operation behavior data of the user in m time units, where m is an integer greater than 1; Combining n first user information graph data, and using the first historical operation behavior sequence of each user in the first user set as the node feature of the node corresponding to each user, to obtain a first time series graph; and Inputting the first time series graph into the trained information recommendation model, and obtaining the prediction result output by the information recommendation model, wherein the prediction result includes the information operation behavior data of each user in the first user set in the next s time units; wherein the information recommendation model is constructed by using a graph neural network; wherein s is an integer greater than or equal to 1 and less than m; and Information is pushed to users in the first user set based on the prediction result.

2. The method according to claim 1, wherein: The combination of n first user information graph data includes: Get the weights of n target attributes; The weight of each target attribute is used as the weight of the first user information graph data corresponding to the target attribute, and the adjacency matrices of the n first user information graph data are weightedly fused based on the respective weights of the n first user information graph data.

3. The method according to claim 2, wherein: The step of obtaining the weights of n target attributes includes: The weights of the n target attributes are extracted from the training results of the information recommendation model, wherein the weights of the n target attributes are parameters optimized during the training process of the information recommendation model.

4. The method according to claim 1, wherein: The obtaining of n first user information graph data includes: Obtaining attribute values ​​of users in the first user set on n target attributes; For each target attribute, the first user set is classified based on the attribute value of each user on the target attribute, and a connection edge is established among users classified into the same category, wherein the similarity condition includes that the users belong to the same category.

5. The method according to claim 1, wherein: The construction process of the information recommendation model includes: Obtain n second user information graph data corresponding to the n target attributes one by one, wherein the n second user information graph data all use users in the second user set as nodes, and establish connection edges between users whose similarity of the corresponding target attributes meets a preset similarity condition in each second user information graph data; Acquire a second historical operation behavior sequence of each user in the second user set, where the second historical operation behavior sequence includes information operation behavior data of the user over m time units; Obtain a sequence of information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence, and obtain label information of each user; Combining n second user information graph data, and using the second historical operation behavior sequence of each user in the second user set as the node feature of the node corresponding to each user, to obtain a second time series graph; The graph neural network is trained using the second time series graph as training input and the label information of each user in the second user set as training output, wherein the trained graph neural network is used as the information recommendation model.

6. The method according to claim 5, wherein: The combining of the n second user information graph data includes: taking the weight of each target attribute as the weight of the second user information graph data corresponding to the target attribute, and performing weighted fusion on the adjacency matrix of the n second user information graph data based on the respective weights of the n second user information graph data; The training of the graph neural network using the second time series graph as training input and the label information of each user in the second user set as training output includes: in each round of iterative training, synchronously updating the parameters of the graph neural network and the weights of n target attributes; and using the updated weights of the n target attributes to weightedly fuse the adjacency matrices of the n second user information graph data to update the second time series graph.

7. A method for constructing an information recommendation model, wherein: The method comprises: Obtain n second user information graph data; wherein the n second user information graph data all have users in the second user set as nodes, and the n second user information graph data correspond one-to-one to the n target attributes, wherein in each second user information graph data, a connection edge is established between users whose similarity of the corresponding target attribute meets a preset similarity condition, and n is an integer greater than 1; Acquire a second historical operation behavior sequence of each user in the second user set, where the second historical operation behavior sequence includes information operation behavior data of the user over m time units; Obtain a sequence of information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence, and obtain label information of each user; Combining n second user information graph data, and using the second historical operation behavior sequence of each user in the second user set as the node feature of the node corresponding to each user, to obtain a second time series graph; The graph neural network is trained using the second time series graph as training input and the label information of each user in the second user set as training output, wherein the trained graph neural network is used as the information recommendation model.

8. An information push device, wherein: The device comprises: A first graph data acquisition module, used to acquire n first user information graph data, wherein the n first user information graph data all have users in the first user set as nodes, and the n first user information graph data correspond one-to-one to n target attributes, wherein in each first user information graph data, a connection edge is established between users whose similarity of the corresponding target attribute meets a preset similarity condition, and n is an integer greater than 1; A first sequence data acquisition module, used to acquire a first historical operation behavior sequence of each user in the first user set, wherein the first historical operation behavior sequence includes information operation behavior data of the user in m time units, where m is an integer greater than 1; A first multi-graph combination module, used to combine n first user information graph data, and use the first historical operation behavior sequence of each user in the first user set as the node feature of the node corresponding to each user to obtain a first time series graph; and A prediction module, comprising a trained information recommendation model, wherein the prediction module is used to input the first time series graph into the information recommendation model, and obtain a prediction result output by the information recommendation model, wherein the prediction result includes information operation behavior data of each user in the first user set in the next s time units; wherein the information recommendation model is constructed using a graph neural network; wherein s is an integer greater than or equal to 1 and less than m; and An information push module is used to push information to users in the first user set based on the prediction result.

9. A device for constructing an information recommendation model, wherein: The device comprises: A second graph data acquisition module, used to acquire n second user information graph data; wherein the n second user information graph data all have users in the second user set as nodes, and the n second user information graph data correspond one-to-one to n target attributes, wherein in each second user information graph data, a connection edge is established between users whose similarity of the corresponding target attribute meets a preset similarity condition, and n is an integer greater than 1; A second sequence data acquisition module, used to acquire a second historical operation behavior sequence of each user in the second user set, wherein the second historical operation behavior sequence includes information operation behavior data of the user over m time units; a labeling module, used to obtain a sequence of information operation behavior data of each user in the second user set in s time units after the second historical operation behavior sequence, and obtain label information of each user; A second multi-graph combination module is used to combine n second user information graph data, and use the second historical operation behavior sequence of each user in the second user set as the node feature of the node corresponding to each user to obtain a second time series graph; A training module is used to train a graph neural network using the second time series graph as training input and the label information of each user in the second user set as training output, wherein the graph neural network after training is used as the information recommendation model.

10. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, The one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

11. A computer-readable storage medium having a computer program or instruction stored thereon, wherein: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

12. A computer program product comprising a computer program or instructions, wherein: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images