Recommendation method and device, equipment, storage medium and program product

Abstract

The invention provides a recommendation method and device, equipment, a storage medium and a program product, and can be applied to the technical field of information. The method comprises the steps of determining a target behavior matrix and a plurality of auxiliary behavior matrixes corresponding to a plurality of auxiliary behavior types according to interaction behaviors between a user and a plurality of products; processing the target behavior matrix and a reference auxiliary behavior matrix in the plurality of auxiliary behavior matrixes by using a graph convolutional neural network, and determining a conditional probability which represents whether the user is recommended with a product under the condition that the reference auxiliary behavior and the purchase behavior are executed on the product; based on the reference auxiliary behavior matrix, determining an edge probability of a reference auxiliary behavior, the edge probability representing a contribution degree of execution of the reference auxiliary behavior to a recommended product; and according to the conditional probabilities and the marginal probabilities corresponding to the multiple auxiliary behavior types, determining a recommendation degree corresponding to each product, and pushing target product information determined based on the recommendation degrees to the user.

Description

Technical Field

[0001] This application relates to the field of information technology, and more specifically to a recommended method, apparatus, device, storage medium, and program product. Background Technology

[0002] With the rapid development of the internet, more and more users prefer to buy products online. Users not only engage in behaviors such as purchasing and subscribing, but also browsing and saving. Multi-behavior recommendation aims to comprehensively utilize different types of behavioral information to improve the predictive effectiveness of target behaviors. However, mainstream methods are based on relevance modeling, which simply weights, concatenates, or fuses multiple behavioral features using graph modeling, ignoring the deeper causal relationships between behaviors. For example, target behaviors may share common underlying factors with other operational behaviors, such as user activity and product popularity. These factors simultaneously influence both the occurrence of the behavior and the recommendation result, leading to false associations and inaccurate recommendations. Summary of the Invention

[0003] In view of the above problems, this application provides a recommended method, apparatus, device, storage medium and program product.

[0004] According to a first aspect of this application, a recommendation method is provided, comprising: determining a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types based on user interaction behavior with multiple products, wherein matrix elements in the target behavior matrix represent whether the user performs a purchase behavior on the product, and matrix elements in the auxiliary behavior matrices represent whether the user performs auxiliary behaviors related to the purchase behavior on the product, and the multiple auxiliary behavior interaction matrices correspond to multiple auxiliary behavior types respectively; processing the target behavior matrix and the reference auxiliary behavior matrix in the multiple auxiliary behavior matrices using a graph convolutional neural network to determine a conditional probability, wherein the conditional probability represents whether the user is recommended a product when performing reference auxiliary behaviors and purchase behaviors on the product; determining the marginal probability of the reference auxiliary behavior based on the reference auxiliary behavior matrix, wherein the marginal probability represents the contribution of performing reference auxiliary behaviors to the recommended product; determining the recommendation degree corresponding to each product based on the conditional probability and marginal probability corresponding to the multiple auxiliary behavior types, and pushing the target product information determined based on the recommendation degree to the user.

[0005] According to an embodiment of this application, a graph convolutional neural network is used to process the target behavior matrix and the reference auxiliary behavior matrix to determine the conditional probability. This includes: performing a logical operation on the ij-th element in the target behavior matrix and the ij-th element in the reference auxiliary behavior matrix to generate a user-product adjacency matrix, where i represents the ith row and j represents the j-th column; processing the user-product adjacency matrix using a graph convolutional neural network to obtain a user-product joint embedding matrix, which includes M rows of user joint embedding vectors and N rows of product joint embedding vectors, where M is the number of users and N is the number of products; and using the inner product of the m-th user joint embedding vector and the n-th product joint embedding vector as the conditional probability of the m-th user for the n-th product, where M ≥ m > 0 and N ≥ n > 0.

[0006] According to an embodiment of this application, determining the marginal probability of a reference auxiliary behavior based on a reference auxiliary behavior matrix includes: processing the reference auxiliary behavior matrix using a graph convolutional neural network to obtain a user-product auxiliary embedding matrix, the user-product auxiliary embedding matrix including M rows of user auxiliary embedding vectors and N rows of product auxiliary embedding vectors; using the inner product of the m-th user auxiliary embedding vector and the n-th product auxiliary embedding vector as the interaction evaluation value of the m-th user to the n-th product under the reference auxiliary behavior; and determining the marginal probability of the reference auxiliary behavior based on the interaction evaluation value of the m-th user to the n-th product corresponding to each auxiliary behavior type.

[0007] According to an embodiment of this application, determining the marginal probability of a reference auxiliary behavior based on the interaction evaluation value of the m-th user to the n-th product corresponding to each auxiliary behavior type includes: determining a first interaction evaluation value and a second interaction evaluation value among the interaction evaluation values ​​corresponding to each auxiliary behavior type, wherein the first interaction evaluation value is greater than the second interaction evaluation value; normalizing the interaction evaluation value of the reference auxiliary behavior based on the first interaction evaluation value and the second interaction evaluation value to obtain a normalized value corresponding to the reference auxiliary behavior; and determining the marginal probability of the reference auxiliary behavior based on the sum of the normalized value and the normalized values ​​corresponding to multiple auxiliary behavior types.

[0008] According to an embodiment of this application, the recommendation degree of each product is determined based on the conditional probability and marginal probability corresponding to each auxiliary behavior type, including: for each product, fusing the conditional probability and marginal probability corresponding to multiple auxiliary behavior types to obtain the recommendation degree of each product.

[0009] According to an embodiment of this application, a graph convolutional neural network is trained based on the following steps: obtaining a training sample set, which includes a first product with interactive behavior with the evaluated user and a second product without interactive behavior, the interactive behavior including purchase behavior and auxiliary behavior; processing the target behavior matrix corresponding to the first product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types using an initial graph convolutional neural network to obtain a first recommendation degree corresponding to the first product; processing the target behavior matrix corresponding to the second product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types using an initial graph convolutional neural network to obtain a second recommendation degree corresponding to the second product; processing the first recommendation degree and the second recommendation degree based on a Bayesian personalized ranking algorithm to obtain a main loss value; for each auxiliary behavior type, processing the user auxiliary embedding vector corresponding to the evaluated user, the first product auxiliary embedding vector corresponding to the first product, and the second product auxiliary embedding vector corresponding to the second product based on a Bayesian personalized ranking algorithm to obtain an auxiliary behavior loss value; determining a confusion constraint loss value based on the auxiliary behavior loss values ​​corresponding to each of the multiple auxiliary behavior types; and training the initial graph convolutional neural network using the main loss value and the confusion constraint loss value to obtain a trained graph convolutional neural network.

[0010] A second aspect of this application provides a recommendation device, comprising: an auxiliary behavior module, configured to determine a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types based on user interaction behavior with multiple products, wherein matrix elements in the target behavior matrix represent whether the user performs a purchase behavior on the product, and matrix elements in the auxiliary behavior matrices represent whether the user performs auxiliary behaviors related to the purchase behavior on the product, and the multiple auxiliary behavior interaction matrices correspond to multiple auxiliary behavior types; a condition module, configured to process the target behavior matrix and a reference auxiliary behavior matrix in the multiple auxiliary behavior matrices using a graph convolutional neural network to determine conditional probabilities, wherein the conditional probabilities represent whether the user is recommended a product when performing reference auxiliary behaviors and purchase behaviors on the product; a marginal probability determination module, configured to determine the marginal probabilities of reference auxiliary behaviors based on the reference auxiliary behavior matrix, wherein the marginal probabilities represent the contribution of performing reference auxiliary behaviors to the recommended products; and a recommendation degree determination module, configured to determine the recommendation degree corresponding to each product based on the conditional probabilities and marginal probabilities corresponding to the multiple auxiliary behavior types, and push target product information determined based on the recommendation degree to the user.

[0011] A third aspect of this application 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 method described above.

[0012] A fourth aspect of this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0013] The fifth aspect of this application also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.

[0014] According to embodiments of this application, based on the user's interaction behavior with multiple products, a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types can be determined. Matrix elements in the target behavior matrix represent whether the user performs a purchase action on a product, and matrix elements in the auxiliary behavior matrices represent whether the user performs auxiliary actions related to the purchase action. Each of the multiple auxiliary behavior interaction matrices corresponds to a different auxiliary behavior type. By using a graph convolutional neural network to process the target behavior matrix and the reference auxiliary behavior matrix in the multiple auxiliary behavior matrices, conditional probabilities can be determined, revealing the confusing effect of auxiliary behaviors on purchase behavior and system recommendation results. The conditional probability characterizes the impact of auxiliary behaviors on the product... When a user performs both reference-assisted behavior and purchase behavior, the marginal probability of a reference-assisted behavior can be determined based on the reference-assisted behavior matrix. By calculating the conditional probability and marginal probability corresponding to each type of auxiliary behavior separately, and by weighting the conditional probabilities and marginal probabilities corresponding to various auxiliary behaviors, the interference of auxiliary behaviors on the process of generating system recommendation results based on purchase behavior can be further eliminated. This allows the determination of the recommendation degree for each product, and the push of target product information to the user based on the recommendation degree, avoiding recommendation bias caused by false relevance. This maintains the accuracy and robustness of recommendations even in scenarios with too many auxiliary behaviors and highly uneven distribution. Attached Figure Description

[0015] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments of this application with reference to the accompanying drawings.

[0016] Figure 1 The diagram illustrates an application scenario of the recommended method and apparatus according to embodiments of this application.

[0017] Figure 2 A flowchart of a recommended method according to an embodiment of this application is shown.

[0018] Figure 3 A causal construction diagram of multi-behavior recommendations according to an embodiment of this application is shown.

[0019] Figure 4 A schematic diagram of the structure of a recommended method according to an embodiment of this application is shown.

[0020] Figure 5 A structural block diagram of a recommended apparatus according to an embodiment of this application is shown.

[0021] Figure 6 A block diagram of an electronic device suitable for implementing the recommended method according to an embodiment of this application is shown. Detailed Implementation

[0022] The embodiments of this application will now 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 this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0023] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0024] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0025] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0026] In actual Internet application systems, users not only generate target behaviors such as purchasing and subscribing, but also generate various auxiliary behaviors such as browsing, collecting, and adding to the shopping cart. Multi-behavior recommendation aims to comprehensively utilize different types of behavioral information to improve the prediction effect of target behaviors (such as purchasing). However, the mainstream methods are all based on correlation modeling, that is, simply weighting, splicing or graph modeling and fusing multiple behavioral features, ignoring the deeper causal relationship between behaviors, resulting in the following problems: (1) Confusion bias problem: There are common potential factors (such as user activity and product popularity) between auxiliary behaviors (such as browsing) and target behaviors (such as purchasing). These factors affect both the occurrence of behavior and the recommendation result, resulting in false association. (2) Uneven distribution of behavior: The number of auxiliary behaviors is far greater than that of target behaviors, causing the model to over-rely on auxiliary behavior signals and mask the user's true intention. (3) Training target conflict: During the multi-behavior fusion process, the information of auxiliary behaviors may have a negative transfer to the prediction of target behaviors, reducing the recommendation quality.

[0027] The relationship between auxiliary behaviors and target behaviors is not merely correlated, but also includes temporal order and causal structure. For example, browsing behavior not only influences purchase intention but also affects the recommender system's judgment of user interests. Furthermore, the number of these auxiliary behaviors far exceeds the target behavior, leading to a severe imbalance in behavior distribution during model training. This causes the model to become overly reliant on auxiliary behaviors, thus learning spurious preferences.

[0028] While related methods can enhance dependency mining between behaviors through graph neural networks or sequence modeling, they still belong to the correlation paradigm and cannot block the confusion path formed by "auxiliary behaviors simultaneously influencing target behaviors and recommendation results." This causes the model to be misled by the bias of auxiliary behaviors when predicting purchasing behavior, resulting in inaccurate recommendations.

[0029] In view of this, this application provides a recommendation method, recommendation apparatus, and device. The method includes: determining a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types based on the interaction behavior between a user and multiple products. The matrix elements in the target behavior matrix represent whether the user performs a purchase behavior on the product, and the matrix elements in the auxiliary behavior matrices represent whether the user performs auxiliary behaviors related to the purchase behavior on the product. Each of the multiple auxiliary behavior interaction matrices corresponds to a multiple auxiliary behavior type. A graph convolutional neural network is used to process the target behavior matrix and the reference auxiliary behavior matrix in the multiple auxiliary behavior matrices to determine conditional probabilities. The conditional probabilities represent whether a product is recommended to the user given that the user performs both the reference auxiliary behavior and the purchase behavior. Based on the reference auxiliary behavior matrix, the marginal probabilities of the reference auxiliary behaviors are determined. The marginal probabilities represent the contribution of performing the reference auxiliary behaviors to the recommended products. Based on the conditional probabilities and marginal probabilities corresponding to the multiple auxiliary behavior types, a recommendation degree corresponding to each product is determined, and the target product information determined based on the recommendation degree is pushed to the user.

[0030] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0031] In scenarios involving automated decision-making using personal information, the methods, devices, and systems provided in this application all offer users corresponding entry points for choosing to agree to or reject the automated decision-making results. If the user chooses to reject, the process proceeds to the expert decision-making stage. Here, "automated decision-making" refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests, or economic, health, and credit status through computer programs, and then making a decision. Here, "expert decision-making" refers to the activity of making decisions by personnel who specialize in a particular field, possess specialized experience, knowledge, and skills, and have reached a certain level of professional expertise.

[0032] Figure 1 The diagram illustrates an application scenario of the recommended method and apparatus according to embodiments of this application.

[0033] like Figure 1 As shown, the application scenario 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 serves as 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 or wireless communication links, or fiber optic cables, etc.

[0034] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via 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 media platform software, etc. (for example only).

[0035] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0036] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process data such as received 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 devices.

[0037] It should be noted that the recommended method provided in this application embodiment can generally be executed by server 105. Correspondingly, the recommended device provided in this application embodiment can generally be located in server 105. The recommended method provided in this application embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the recommended device provided in this application embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105.

[0038] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0039] Figure 2 A flowchart of a recommended method according to an embodiment of this application is shown.

[0040] like Figure 2 As shown, the recommended method of this embodiment includes operations S210 to S240, and this recommended method can be executed by an electronic device.

[0041] In operation S210, based on the user's interaction behavior with multiple products, the target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types are determined.

[0042] In operation S220, a graph convolutional neural network is used to process the target behavior matrix and the reference auxiliary behavior matrix among multiple auxiliary behavior matrices to determine the conditional probability.

[0043] In operation S230, the marginal probabilities of reference auxiliary actions are determined based on the reference auxiliary action matrix. The marginal probabilities represent the contribution of performing reference auxiliary actions to the recommended products.

[0044] In operation S240, the recommendation level of each product is determined based on the conditional probability and marginal probability corresponding to multiple auxiliary behavior types, and the target product information determined based on the recommendation level is pushed to the user.

[0045] Interactive behaviors can include purchasing behaviors and auxiliary behaviors. Purchasing behaviors can include purchasing or subscribing. Auxiliary behaviors can include browsing behaviors, adding to favorites, adding to cart, etc., but are not limited to these, and the embodiments of this application do not limit them. The matrix elements in the target behavior matrix represent whether the user performs a purchasing behavior on the product, and the matrix elements in the auxiliary behavior matrix represent whether the user performs auxiliary behaviors related to the purchasing behavior on the product.

[0046] For each type of auxiliary behavior, an auxiliary behavior matrix can be determined based on whether the user performs an auxiliary behavior related to the purchase of the product. The user ID can represent rows, and the product ID can represent columns. If there is user-product interaction in the auxiliary behavior record, the element at the corresponding position in the auxiliary behavior matrix is ​​1; otherwise, it is 0. The target behavior matrix is ​​constructed in the same way as the auxiliary behavior matrix.

[0047] Multiple auxiliary behavior interaction matrices correspond to multiple auxiliary behavior types. For example, auxiliary behavior types related to browsing behavior correspond to a browsing auxiliary behavior interaction matrix, and auxiliary behavior types related to bookmarking behavior correspond to a bookmarking auxiliary behavior interaction matrix.

[0048] Conditional probability characterizes whether a user is recommended a product after performing a reference-assisted behavior and a purchase behavior. For example, it determines whether the system recommended the product to the user before the purchase, after the user has browsed the product and then purchased it. Graph convolutional neural networks can be used to process the target behavior matrix and the reference-assisted behavior matrix to determine the conditional probability.

[0049] Marginal probabilities represent the contribution of the reference auxiliary behavior to the recommended product. The marginal probabilities of each reference auxiliary behavior can be determined based on the reference auxiliary behavior matrix. Each reference auxiliary behavior corresponds to its own marginal probability.

[0050] Based on the conditional and marginal probabilities corresponding to multiple auxiliary behavior types, the recommendation degree of each product can be determined, as shown in formula (1).

[0051] (1);

[0052] in, The degree of recommendation indicates the causal effect of the target behavior on the recommendation result. Operators represent interventions in the target behavior. Given the target behavior The conditional probability of recommendation result Y when the auxiliary behavior c is met. is the marginal probability of auxiliary behavior c, used to weight the causal contributions of different auxiliary behaviors. C represents the set of auxiliary behaviors.

[0053] Based on the above formula (1), the recommendation level of each product can be calculated, and the information of target products with a recommendation level greater than the preset threshold can be pushed to the user.

[0054] Figure 3 A causal construction diagram of multi-behavior recommendations according to an embodiment of this application is shown.

[0055] like Figure 3 As shown, a causal graph for multi-behavior recommendations can be constructed. This graph includes multiple variables: an auxiliary behavior set C, a target behavior T, and a recommendation result Y. The auxiliary behavior set C can include auxiliary behavior 1, auxiliary behavior 2, ..., auxiliary behavior k. This causal graph can contain three core paths: the causal path of auxiliary behavior set C to target behavior T (C→T), such as browsing increasing the probability of purchase; the direct causal path of auxiliary behavior set C to recommendation result Y (C→Y), where auxiliary behaviors themselves are also considered interest signals by the system; and the core causal path of target behavior T to recommendation result Y (T→Y), where the core goal of the recommendation system is to predict the target behavior. Since C simultaneously affects T and Y, a backdoor path T→Y is formed (T←C→Y), which is the root cause of confounding bias. Therefore, each auxiliary behavior can be considered an independent confounding variable. This application intervenes to disconnect the backdoor path to accurately measure the pure causal effect of T on Y.

[0056] According to embodiments of this application, based on the user's interaction behavior with multiple products, a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types can be determined. Matrix elements in the target behavior matrix represent whether the user performs a purchase action on a product, and matrix elements in the auxiliary behavior matrices represent whether the user performs auxiliary actions related to the purchase action. Each of the multiple auxiliary behavior interaction matrices corresponds to a different auxiliary behavior type. By using a graph convolutional neural network to process the target behavior matrix and the reference auxiliary behavior matrix in the multiple auxiliary behavior matrices, conditional probabilities can be determined, revealing the confusing effect of auxiliary behaviors on purchase behavior and system recommendation results. The conditional probability characterizes the impact of auxiliary behaviors on the product... When a user performs both reference-assisted behavior and purchase behavior, the marginal probability of a reference-assisted behavior can be determined based on the reference-assisted behavior matrix. By calculating the conditional probability and marginal probability corresponding to each type of auxiliary behavior separately, and by weighting the conditional probabilities and marginal probabilities corresponding to various auxiliary behaviors, the interference of auxiliary behaviors on the process of generating system recommendation results based on purchase behavior can be further eliminated. This allows the determination of the recommendation degree for each product, and the push of target product information to the user based on the recommendation degree, avoiding recommendation bias caused by false relevance. This maintains the accuracy and robustness of recommendations even in scenarios with too many auxiliary behaviors and highly uneven distribution.

[0057] According to an embodiment of this application, a graph convolutional neural network is used to process the target behavior matrix and the reference auxiliary behavior matrix to determine the conditional probability. This includes: performing a logical operation on the ij-th element in the target behavior matrix and the ij-th element in the reference auxiliary behavior matrix to generate a user-product adjacency matrix, where i represents the ith row and j represents the j-th column; processing the user-product adjacency matrix using a graph convolutional neural network to obtain a user-product joint embedding matrix, which includes M rows of user joint embedding vectors and N rows of product joint embedding vectors, where M is the number of users and N is the number of products; and using the inner product of the m-th user joint embedding vector and the n-th product joint embedding vector as the conditional probability of the m-th user for the n-th product, where M ≥ m > 0 and N ≥ n > 0.

[0058] In order to obtain "simultaneous target behavior" User interests under the constraint of auxiliary behavior c can be used to perform element-level logical OR operations on the positions corresponding to the target behavior matrix and the reference auxiliary behavior matrix to generate a user product adjacency matrix, as shown in formula (2).

[0059] (2);

[0060] Here, "|" represents the element-wise logical OR operation. This represents the user product adjacency matrix, used to represent the reference auxiliary behavior matrix. and target behavior matrix As long as it exists ,but element at the corresponding position Otherwise, it is 0 (i and j represent the row number and column number in the matrix, respectively).

[0061] Transform the user-product adjacency matrix into the user-product bipartite graph adjacency matrix required by the graph convolutional neural network. As shown in formula (3). Then, perform multi-layer propagation of the graph convolutional neural network to obtain the joint embedding matrix of user products. As shown in formulas (4) and (5)

[0062] (3);

[0063] (4);

[0064] (5);

[0065] Where D is the degree matrix, and the diagonal elements of this matrix are matrix 2. The sum of elements in each row of the graph is 0, and all other elements are 0. l represents the l-th layer of the graph convolutional neural network, and L represents the L-th layer of the graph convolutional neural network. When hour, It is a randomly initialized matrix with the following shape: M represents the number of users in the system, N represents the number of products in the system, and w represents the number of columns in the matrix. This represents the user product joint embedding matrix at layer l. Let represent the user-product joint embedding matrix at layer l+1. This means that the first M rows of the user-product joint embedding matrix represent users, and the last M rows represent users. The row represents the product. This is the user-product joint embedding matrix. Finally, the m-th user joint embedding vector, indexed from the m-th row of the user-product joint embedding matrix, represents the user with ID m, and is denoted as... Index to the A row vector represents the embedding of the product with number n, denoted as .

[0066] The inner product of the joint embedding vector of the m-th user and the joint embedding vector of the n-th product is taken as the conditional probability of the m-th user for the n-th product, where M≥m>0 and N≥n>0.

[0067] Taking user u and product i as an example, the conditional probability of user u for product i is determined according to the following formula. .

[0068] (6);

[0069] Where ReLU is the activation function. Let u represent the joint embedding vector of user u. Let represent the joint embedding vector of the i-th product, used to show that the probability value is non-negative.

[0070] According to embodiments of this application, unlike simply fusing multiple behaviors or relying on graph neural networks to mine behavior co-occurrence, this application first clearly characterizes the causal structure between auxiliary behaviors, target behaviors, and recommendation results. By calculating the conditional probability under the constraints of both target behaviors and auxiliary behaviors, key confusion paths affecting recommendation accuracy can be identified, and the true causal contribution of target behaviors to user interests can be directly estimated, thereby avoiding recommendation bias caused by false relevance.

[0071] According to an embodiment of this application, determining the marginal probability of a reference auxiliary behavior based on a reference auxiliary behavior matrix includes: processing the reference auxiliary behavior matrix using a graph convolutional neural network to obtain a user-product auxiliary embedding matrix, the user-product auxiliary embedding matrix including M rows of user auxiliary embedding vectors and N rows of product auxiliary embedding vectors; using the inner product of the m-th user auxiliary embedding vector and the n-th product auxiliary embedding vector as the interaction evaluation value of the m-th user to the n-th product under the reference auxiliary behavior; and determining the marginal probability of the reference auxiliary behavior based on the interaction evaluation value of the m-th user to the n-th product corresponding to each auxiliary behavior type.

[0072] The reference auxiliary behavior matrix is ​​used as the input to the graph convolutional neural network for calculation. The calculation process is similar to the steps described above for obtaining the joint embedding matrix of the user product, i.e., using the reference auxiliary behavior matrix... Replace the user product adjacency matrix in this step This yields the embeddings of user m and product n specific to the reference auxiliary behavior c, denoted as the m-th user auxiliary embedding vector. and the nth product auxiliary embedding vector .

[0073] The inner product of the auxiliary embedding vector of the m-th user and the auxiliary embedding vector of the n-th product is used as the interaction evaluation value of the m-th user to the n-th product under the reference auxiliary behavior c. As shown in formula (7).

[0074] (7);

[0075] The marginal probability of a reference assistive behavior can be determined based on the interaction evaluation value of the m-th user with the n-th product corresponding to each assistive behavior type.

[0076] According to embodiments of this application, observable auxiliary behavior interaction data is used to quantify its impact as a confounding factor, i.e., marginal probability. The method is intuitive and computationally reliable. When a new user behavior type is added to the business scenario, it is only necessary to include it in the auxiliary behavior set and calculate its corresponding probability term to achieve model expansion, without redesigning or training a complex latent variable inference network. Furthermore, by adopting a backdoor adjustment method, the interference of auxiliary behaviors on the prediction of target behavior is effectively eliminated, revealing the confounding effect of auxiliary behaviors on target behavior and recommendation results. In scenarios with too many auxiliary behaviors and highly uneven distribution, recommendation accuracy and robustness can be maintained.

[0077] According to an embodiment of this application, determining the marginal probability of a reference auxiliary behavior based on the interaction evaluation value of the m-th user to the n-th product corresponding to each auxiliary behavior type includes: determining a first interaction evaluation value and a second interaction evaluation value among the interaction evaluation values ​​corresponding to each auxiliary behavior type, wherein the first interaction evaluation value is greater than the second interaction evaluation value; normalizing the interaction evaluation value of the reference auxiliary behavior based on the first interaction evaluation value and the second interaction evaluation value to obtain a normalized value corresponding to the reference auxiliary behavior; and determining the marginal probability of the reference auxiliary behavior based on the sum of the normalized value and the normalized values ​​corresponding to multiple auxiliary behavior types.

[0078] To ensure that the sum of the marginal probabilities of all assistive behaviors is 1, normalization can be used. Among the interaction evaluation values ​​corresponding to each assistive behavior type, a first interaction evaluation value is determined. Second interaction evaluation value The first interaction evaluation value is greater than the second interaction evaluation value. Based on the first and second interaction evaluation values, the interaction evaluation value of the assistive behavior will be referenced. Normalization is performed to obtain the normalized value corresponding to the reference auxiliary behavior. As shown in formula (8).

[0079] (8);

[0080] Where j represents the reference auxiliary behavior, and C represents the set of auxiliary behaviors.

[0081] The marginal probability of the reference auxiliary behavior is determined by the sum of the normalized value and the normalized values ​​corresponding to each of the multiple auxiliary behavior types, as shown in formula (9).

[0082] (9);

[0083] in, This represents the sum of normalized values. This represents the marginal probability of the reference auxiliary behavior.

[0084] According to the embodiments of this application, by using the first interaction evaluation value and the second interaction evaluation value as a dynamic scale for normalization, the marginal probability of the reference auxiliary behavior is guaranteed to be a comparable probability, while allowing the probability value to better reflect the proportion of the auxiliary behavior among all auxiliary behaviors.

[0085] According to an embodiment of this application, the recommendation degree of each product is determined based on the conditional probability and marginal probability corresponding to each auxiliary behavior type, including: for each product, fusing the conditional probability and marginal probability corresponding to multiple auxiliary behavior types to obtain the recommendation degree of each product.

[0086] For each product, the conditional and marginal probabilities corresponding to multiple auxiliary behavior types of user u for product i can be combined to obtain the recommendation degree of product i for user u. As shown in formula (10).

[0087] (10);

[0088] in, Let C represent the auxiliary behavior, and C represent the set of auxiliary behaviors. Represents conditional probability. This represents the marginal probability.

[0089] According to an embodiment of this application, a graph convolutional neural network is trained based on the following steps: obtaining a training sample set, which includes a first product with interactive behavior with the evaluated user and a second product without interactive behavior, the interactive behavior including purchase behavior and auxiliary behavior; processing the target behavior matrix corresponding to the first product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types using an initial graph convolutional neural network to obtain a first recommendation degree corresponding to the first product; processing the target behavior matrix corresponding to the second product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types using an initial graph convolutional neural network to obtain a second recommendation degree corresponding to the second product; processing the first recommendation degree and the second recommendation degree based on a Bayesian personalized ranking algorithm to obtain a main loss value; for each auxiliary behavior type, processing the user auxiliary embedding vector corresponding to the evaluated user, the first product auxiliary embedding vector corresponding to the first product, and the second product auxiliary embedding vector corresponding to the second product based on a Bayesian personalized ranking algorithm to obtain an auxiliary behavior loss value; determining a confusion constraint loss value based on the auxiliary behavior loss values ​​corresponding to each of the multiple auxiliary behavior types; and training the initial graph convolutional neural network using the main loss value and the confusion constraint loss value to obtain a trained graph convolutional neural network.

[0090] A training sample set S can be obtained, which may include a first product p that has interactive behavior with the user being evaluated and a second product r that does not have interactive behavior. Interactive behavior may include purchasing behavior and auxiliary behavior.

[0091] By using an initial graph convolutional neural network to process the target behavior matrix corresponding to the first product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types, the first recommendation score corresponding to the first product can be obtained. By using an initial graph convolutional neural network to process the target behavior matrix corresponding to the second product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types, the second recommendation score corresponding to the second product can be obtained. .

[0092] Using the Bayesian personalized ranking algorithm, we can ensure that positive samples score higher than negative samples. Positive samples represent the first product that has interacted with the evaluated user, and negative samples represent the second product that has not interacted with the evaluated user.

[0093] Specifically, the first and second recommendation scores can be processed based on the Bayesian personalized ranking algorithm to obtain the main loss value. As shown in formula (11).

[0094] (11);

[0095] Where p represents products in the training set that have interacted with user u, r represents products that have not interacted with user u, and S represents the training sample set. This indicates the first recommendation level between user u and product p. This represents the second recommendation level between user u and product r. This represents the sigmoid function.

[0096] For each type of auxiliary behavior, the user auxiliary embedding vector corresponding to the evaluated user is processed based on a Bayesian personalized ranking algorithm. The first product's auxiliary embedding vector Second product auxiliary embedding vector corresponding to the second product Obtain the auxiliary behavior loss value As shown in formula (12).

[0097] (12);

[0098] in, This represents the sigmoid function.

[0099] Determine the confusion constraint loss value based on the corresponding auxiliary behavior loss values ​​for each of the multiple auxiliary behavior types. As shown in formula (13).

[0100] (13);

[0101] Where C is the set of auxiliary behaviors.

[0102] Using the main loss value and confusion constraint loss value Determined total loss value Train the initial graph convolutional neural network to obtain the trained graph convolutional neural network, as shown in formula (14).

[0103] (14);

[0104] in This is a hyperparameter used to balance recommendation tasks and confusion constraints.

[0105] According to embodiments of this application, by constructing a graph convolutional neural network that integrates causal interventions, the causal effects of target behaviors can be better estimated, thereby improving recommendation quality. By combining causal modeling with a lightweight graph convolutional structure, stable performance is maintained not only in environments with an extremely large number of auxiliary behaviors and severely uneven behavior distribution, but also the interpretability of recommendation results is significantly improved, enabling the model to quantitatively explain the impact of different auxiliary behaviors on prediction results.

[0106] Figure 4 A schematic diagram of the structure of a recommended method according to an embodiment of this application is shown.

[0107] like Figure 4 As shown, based on the reference auxiliary behavior matrix and target behavior matrix Determine the adjacency matrix of user products The user product adjacency matrix is ​​processed using a Light Graph Convolutional Network (LightGCN). Initial user joint embedding vector and initial product joint embedding vector To obtain the joint embedding matrix of user products User product joint embedding matrix Including the joint embedding vector of user u Joint embedding vector of product i The joint embedding vector of user u. Joint embedding vectors with i products The inner product of and is used as the conditional probability of user u for product i. .

[0108] Processing reference auxiliary behavior matrices using graph convolutional neural networks Initial user joint embedding vector and initial product joint embedding vector The auxiliary embedding vector of the u-th user is obtained. and the auxiliary embedding vector of the i-th product Based on the u-th user-assisted embedding vector and the auxiliary embedding vector of the i-th product The inner product determines the marginal probability of the reference auxiliary behavior. Based on the conditional probabilities corresponding to multiple auxiliary behavior types. and marginal probability Determine the recommendation level for each product.

[0109] Figure 5 A structural block diagram of a recommended apparatus according to an embodiment of this application is shown.

[0110] like Figure 5 As shown, the recommendation device in this embodiment includes an auxiliary behavior module 510, a condition module 520, an edge probability determination module 530, and a recommendation degree determination module 540.

[0111] The auxiliary behavior module 510 is used to determine the target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types based on the user's interaction behavior with multiple products. The matrix elements in the target behavior matrix represent whether the user performs a purchase behavior on the product, and the matrix elements in the auxiliary behavior matrix represent whether the user performs auxiliary behaviors related to the purchase behavior on the product. The multiple auxiliary behavior interaction matrices correspond to the multiple auxiliary behavior types.

[0112] Condition module 520 is used to process the target behavior matrix and the reference auxiliary behavior matrix among multiple auxiliary behavior matrices using a graph convolutional neural network to determine the conditional probability. The conditional probability represents whether the user is recommended a product given that the reference auxiliary behavior and the purchase behavior are performed on the product.

[0113] The edge probability determination module 530 is used to determine the edge probability of the reference auxiliary behavior based on the reference auxiliary behavior matrix. The edge probability represents the contribution of performing the reference auxiliary behavior to the recommended product.

[0114] The recommendation degree determination module 540 is used to determine the recommendation degree of each product based on the conditional probability and marginal probability corresponding to multiple auxiliary behavior types, and push the target product information based on the recommendation degree determination to the user.

[0115] According to embodiments of this application, based on the user's interaction behavior with multiple products, a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types can be determined. Matrix elements in the target behavior matrix represent whether the user performs a purchase action on a product, and matrix elements in the auxiliary behavior matrices represent whether the user performs auxiliary actions related to the purchase action. Each of the multiple auxiliary behavior interaction matrices corresponds to a different auxiliary behavior type. By using a graph convolutional neural network to process the target behavior matrix and the reference auxiliary behavior matrix in the multiple auxiliary behavior matrices, conditional probabilities can be determined, revealing the confusing effect of auxiliary behaviors on purchase behavior and system recommendation results. The conditional probability characterizes the impact of auxiliary behaviors on the product... When a user performs both reference-assisted behavior and purchase behavior, the marginal probability of a reference-assisted behavior can be determined based on the reference-assisted behavior matrix. By calculating the conditional probability and marginal probability corresponding to each type of auxiliary behavior separately, and by weighting the conditional probabilities and marginal probabilities corresponding to various auxiliary behaviors, the interference of auxiliary behaviors on the process of generating system recommendation results based on purchase behavior can be further eliminated. This allows the determination of the recommendation degree for each product, and the push of target product information to the user based on the recommendation degree, avoiding recommendation bias caused by false relevance. This maintains the accuracy and robustness of recommendations even in scenarios with too many auxiliary behaviors and highly uneven distribution.

[0116] The condition module 520 includes: adjacency unit, embedding unit and conditional probability unit.

[0117] The adjacency unit is used to perform logical operations between the ij-th element in the target behavior matrix and the ij-th element in the reference auxiliary behavior matrix to generate a user product adjacency matrix, where i represents the ij-th row and j represents the j-th column.

[0118] The embedding unit is used to process the user-product adjacency matrix using a graph convolutional neural network to obtain a user-product joint embedding matrix. The user-product joint embedding matrix includes M rows of user joint embedding vectors and N rows of product joint embedding vectors, where M is the number of users and N is the number of products.

[0119] The conditional probability unit is used to take the inner product of the joint embedding vector of the m-th user and the joint embedding vector of the n-th product as the conditional probability of the m-th user for the n-th product, where M≥m>0 and N≥n>0.

[0120] The edge probability determination module 530 includes: an auxiliary embedding unit, an interactive evaluation value unit, and an edge probability unit.

[0121] The auxiliary embedding unit is used to process the reference auxiliary behavior matrix using a graph convolutional neural network to obtain the user product auxiliary embedding matrix, which includes M rows of user auxiliary embedding vectors and N rows of product auxiliary embedding vectors.

[0122] The interaction evaluation value unit is used to take the inner product of the m-th user's auxiliary embedding vector and the n-th product's auxiliary embedding vector as the interaction evaluation value of the m-th user to the n-th product under the reference auxiliary behavior.

[0123] The marginal probability unit is used to determine the marginal probability of a reference auxiliary behavior based on the interaction evaluation value of the m-th user to the n-th product corresponding to each auxiliary behavior type.

[0124] The marginal probability unit includes: the interactive evaluation value subunit and the normalization subunit.

[0125] The interaction evaluation value subunit is used to determine the first interaction evaluation value and the second interaction evaluation value among the interaction evaluation values ​​corresponding to each auxiliary behavior type, wherein the first interaction evaluation value is greater than the second interaction evaluation value.

[0126] The normalization subunit is used to normalize the interaction evaluation value of the reference auxiliary behavior based on the first interaction evaluation value and the second interaction evaluation value, so as to obtain the normalized value corresponding to the reference auxiliary behavior.

[0127] The marginal probability subunit is used to determine the marginal probability of a reference auxiliary behavior based on the sum of the normalized value and the normalized values ​​corresponding to each of the multiple auxiliary behavior types.

[0128] The recommendation determination module 540 includes: recommendation unit.

[0129] The recommendation unit is used to combine the conditional probabilities and marginal probabilities of multiple auxiliary behavior types for each product to obtain the recommendation score for each product.

[0130] The embedding unit includes: an acquisition subunit, a first recommendation subunit, a second recommendation subunit, a main loss subunit, an auxiliary loss subunit, a confusion constraint subunit, and a training subunit.

[0131] The acquisition sub-unit is used to acquire a training sample set, which includes a first product that has interactive behavior with the user being evaluated and a second product that does not have interactive behavior. Interactive behavior includes purchasing behavior and auxiliary behavior.

[0132] The first recommendation subunit is used to process the target behavior matrix corresponding to the first product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types using the initial graph convolutional neural network to obtain the first recommendation degree corresponding to the first product.

[0133] The second recommendation subunit is used to process the target behavior matrix corresponding to the second product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types using the initial graph convolutional neural network to obtain the second recommendation degree corresponding to the second product.

[0134] The main loss subunit is used to process the first and second recommendation scores based on the Bayesian personalized ranking algorithm to obtain the main loss value.

[0135] The auxiliary loss subunit is used to process the user auxiliary embedding vector corresponding to the evaluated user, the first product auxiliary embedding vector corresponding to the first product, and the second product auxiliary embedding vector corresponding to the second product based on the Bayesian personalized ranking algorithm for each auxiliary behavior type, so as to obtain the auxiliary behavior loss value.

[0136] The confuse constraint subunit is used to determine the confuse constraint loss value based on the auxiliary behavior loss value corresponding to each of the multiple auxiliary behavior types.

[0137] The training subunit is used to train the initial graph convolutional neural network using the main loss value and the confusion constraint loss value, resulting in the trained graph convolutional neural network.

[0138] According to embodiments of this application, any multiple modules among the auxiliary behavior module 510, condition module 520, edge probability determination module 530, and recommendation degree determination module 540 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the auxiliary behavior module 510, condition module 520, edge probability determination module 530, and recommendation degree determination module 540 can be at least partially implemented as hardware circuitry, 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-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the auxiliary behavior module 510, condition module 520, edge probability determination module 530, and recommendation degree determination module 540 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0139] Figure 6 A block diagram of an electronic device suitable for implementing a partition table partitioning method and a data processing method according to an embodiment of this application is shown.

[0140] like Figure 6As shown, an electronic device 600 according to an embodiment of this application includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a ROM 602 (Read-Only Memory) or a program loaded from a storage portion 608 into a RAM 603 (Random Access Memory). The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0141] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0142] According to embodiments of this application, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0143] This application also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0144] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include ROM 602 and / or RAM 603 and / or one or more memories other than ROM 602 and RAM 603 described above.

[0145] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the recommended methods provided in the embodiments of this application.

[0146] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this application embodiment. According to the embodiments of this application, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0147] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0148] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, it performs the functions defined in the system of this application embodiment. According to the embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0149] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0150] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0151] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this application can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments and / or claims of this application can be combined or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

[0152] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Without departing from the scope of this application, those skilled in the art can make various substitutions and modifications, all of which should fall within the scope of this application.

Claims

1. A recommendation method, characterized in that, include: Based on the user's interaction behavior with multiple products, a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types are determined. The matrix elements in the target behavior matrix represent whether the user performs a purchase behavior on the product, and the matrix elements in the auxiliary behavior matrix represent whether the user performs auxiliary behaviors related to the purchase behavior on the product. Each of the multiple auxiliary behavior interaction matrices corresponds to a multiple auxiliary behavior type. A graph convolutional neural network is used to process the target behavior matrix and a reference auxiliary behavior matrix among a plurality of auxiliary behavior matrices to determine a conditional probability. The conditional probability represents whether the user is recommended the product when performing a reference auxiliary behavior and a purchase behavior on the product. The reference auxiliary behavior matrix is ​​any one of the plurality of auxiliary behavior matrices. Based on the reference auxiliary behavior matrix, the marginal probability of the reference auxiliary behavior is determined, and the marginal probability represents the contribution of performing the reference auxiliary behavior to recommending the product; Based on the conditional and marginal probabilities corresponding to multiple auxiliary behavior types, the recommendation level for each product is determined, and the target product information determined based on the recommendation level is pushed to the user.

2. The method according to claim 1, characterized in that, The process of using a graph convolutional neural network to process the target behavior matrix and the reference auxiliary behavior matrix to determine the conditional probability includes: Perform logical operations between the ij-th element in the target behavior matrix and the ij-th element in the reference auxiliary behavior matrix to generate a user product adjacency matrix, where i represents the ij-th row and j represents the j-th column; The user-product adjacency matrix is ​​processed using a graph convolutional neural network to obtain a user-product joint embedding matrix. The user-product joint embedding matrix includes M rows of user joint embedding vectors and N rows of product joint embedding vectors, where M is the number of users and N is the number of products. The inner product of the joint embedding vector of the m-th user and the joint embedding vector of the n-th product is taken as the conditional probability of the m-th user for the n-th product, where M≥m>0 and N≥n>0.

3. The method according to claim 1, characterized in that, Determining the marginal probability of the reference auxiliary behavior based on the reference auxiliary behavior matrix includes: The user product auxiliary embedding matrix is ​​obtained by processing the reference auxiliary behavior matrix using a graph convolutional neural network. The user product auxiliary embedding matrix includes M rows of user auxiliary embedding vectors and N rows of product embedding vectors. The inner product of the m-th user's auxiliary embedding vector and the n-th product's auxiliary embedding vector is used as the interaction evaluation value of the m-th user to the n-th product under the reference auxiliary behavior. The marginal probability of the reference auxiliary behavior is determined based on the interaction evaluation value of the m-th user to the n-th product corresponding to each type of auxiliary behavior.

4. The method according to claim 3, characterized in that, The step of determining the marginal probability of the reference assistive behavior based on the interaction evaluation value of the m-th user to the n-th product corresponding to each assistive behavior type includes: In the interaction evaluation values ​​corresponding to each type of assistive behavior, a first interaction evaluation value and a second interaction evaluation value are determined, wherein the first interaction evaluation value is greater than the second interaction evaluation value; Based on the first interaction evaluation value and the second interaction evaluation value, the interaction evaluation value of the reference auxiliary behavior is normalized to obtain the normalized value corresponding to the reference auxiliary behavior. The marginal probability of the reference auxiliary behavior is determined by the sum of the normalized value corresponding to the reference auxiliary behavior and the normalized values ​​corresponding to each of the multiple auxiliary behavior types.

5. The method according to claim 1, characterized in that, The process of determining the recommendation level for each product based on the conditional probability and marginal probability corresponding to each type of auxiliary behavior includes: For each product, the conditional probability and marginal probability corresponding to each of the multiple auxiliary behavior types are combined to obtain the recommendation degree corresponding to each product.

6. The method according to claim 1, characterized in that, The graph convolutional neural network is trained based on the following steps: Obtain a training sample set, which includes a first product that has interactive behavior with the user being evaluated and a second product that does not have interactive behavior, wherein the interactive behavior includes the purchase behavior and the auxiliary behavior; The initial graph convolutional neural network is used to process the target behavior matrix corresponding to the first product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types to obtain the first recommendation degree corresponding to the first product. The initial graph convolutional neural network is used to process the target behavior matrix corresponding to the second product and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types to obtain the second recommendation degree corresponding to the second product. The first recommendation score and the second recommendation score are processed based on the Bayesian personalized ranking algorithm to obtain the main loss value; For each type of auxiliary behavior, the user auxiliary embedding vector corresponding to the evaluated user, the first product auxiliary embedding vector corresponding to the first product, and the second product auxiliary embedding vector corresponding to the second product are processed based on the Bayesian personalized ranking algorithm to obtain the auxiliary behavior loss value. Determine the confusion constraint loss value based on the auxiliary behavior loss value corresponding to each of the multiple auxiliary behavior types; The initial graph convolutional neural network is trained using the main loss value and the confusion constraint loss value to obtain the trained graph convolutional neural network.

7. A recommended device, characterized in that, include: The auxiliary behavior module is used to determine a target behavior matrix and multiple auxiliary behavior matrices corresponding to multiple auxiliary behavior types based on the user's interaction behavior with multiple products. The matrix elements in the target behavior matrix represent whether the user performs a purchase behavior on the product, and the matrix elements in the auxiliary behavior matrix represent whether the user performs auxiliary behaviors related to the purchase behavior on the product. The multiple auxiliary behavior interaction matrices correspond to multiple auxiliary behavior types. The condition module is used to process the target behavior matrix and a reference auxiliary behavior matrix among a plurality of auxiliary behavior matrices using a graph convolutional neural network to determine a conditional probability. The conditional probability represents whether the user is recommended the product when performing a reference auxiliary behavior and a purchase behavior on the product. The reference auxiliary behavior matrix is ​​any one of the plurality of auxiliary behavior matrices. The edge probability determination module is used to determine the edge probability of the reference auxiliary behavior based on the reference auxiliary behavior matrix, wherein the edge probability represents the contribution of performing the reference auxiliary behavior to recommending the product; The recommendation degree determination module is used to determine the recommendation degree corresponding to each product based on the conditional probability and marginal probability corresponding to multiple auxiliary behavior types, and push the target product information based on the recommendation degree determination to the user.

8. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that 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 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images