Product recommendation method and apparatus, storage medium, and electronic device

Abstract

The application discloses a product recommendation method and device, a storage medium and an electronic device. It relates to the field of artificial intelligence, and the method comprises the following steps: extracting a first feature vector from an interaction graph of a video platform, wherein the interaction graph comprises a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction graph; determining a plurality of financial products to be recommended, and extracting a second feature vector for each financial product; for each user, calculating the similarity between the first feature vector and the second feature vector of each financial product; and in the case where the similarity is greater than or equal to a similarity threshold, recommending the financial product corresponding to the similarity greater than or equal to the similarity threshold to the user. Through the application, the problem of low accuracy of the financial products recommended in the video platform in the related art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a product recommendation method and device, a storage medium and an electronic device. BACKGROUND

[0002] With the continuous development of Internet technology, advertising has become one of the important means in network marketing. As a marketing channel with a large user base, video platforms have the advantage of advertising placement, and a precise advertising recommendation system can improve the conversion rate of financial products based on advertising recommendation, thereby improving the advertising value of the platform. In related technologies, an advertising recommendation system trains an advertising recommendation model through user data and pushes advertisements to users based on the advertising recommendation model. However, the advertising recommendation model contains sensitive information, and directly using real user data can easily leak the privacy of users. Moreover, the accuracy of the advertising recommendation method in related technologies is low.

[0003] Traditional privacy protection methods such as anonymization methods are vulnerable to de-anonymization attacks and inference attacks and cannot effectively protect user data. Differential privacy (DP) as a recognized standard for privacy protection is widely used to protect user data. In related technologies, some works that utilize differential privacy to protect user data focus on protecting privacy in a specific data mining task, such as degree distribution and subgraph counting. Compared to protecting privacy in a single task, user data that meets differential privacy can be used for multiple downstream tasks. However, the differential privacy algorithm in related technologies introduces too much noise by directly perturbing the adjacency matrix, or suffers from severe information loss during graph encoding, which reduces the accuracy of the financial product recommendation system.

[0004] In view of the problem of low accuracy of financial products recommended in video platforms in related technologies, no effective solution has been proposed so far. SUMMARY

[0005] The main purpose of the present application is to provide a product recommendation method and device, a storage medium and an electronic device to solve the problem of low accuracy of financial products recommended in video platforms in related technologies.

[0006] To achieve the above object, according to one aspect of the present application, a product recommendation method is provided. The method comprises: extracting a first feature vector from an interaction graph of a video platform, wherein the interaction graph is an interaction graph between all users of the video platform and videos browsed by the users, the interaction graph comprises a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction graph; determining a plurality of financial products to be recommended, and extracting a second feature vector for each financial product; for each user, calculating the similarity between the first feature vector and the second feature vector of each financial product; and in the case where the similarity is greater than or equal to a similarity threshold, recommending the financial product corresponding to the similarity greater than or equal to the similarity threshold to the user.

[0007] Optionally, the user information of each user of the video platform is cached in a database in an encrypted manner, the user information comprises identity information and interaction behavior, and the interaction graph is determined by: decrypting the encrypted user information to obtain the identity information and the interaction behavior, and determining the videos browsed by each user; determining each user as a user node in the interaction graph, and determining each video as a video node in the interaction graph; connecting an edge between two nodes with the first interaction behavior, and connecting an edge between two nodes with the second interaction behavior, to obtain an initial interaction graph, wherein the first interaction behavior is an operation of a user to a video in the process of browsing the video, and the second interaction behavior is an interaction operation between the user and a video creator; and performing differential privacy processing on the initial interaction graph to obtain the interaction graph.

[0008] Optionally, the differential privacy processing on the initial interaction graph to obtain the interaction graph comprises: dividing all nodes in the initial interaction graph into a plurality of communities satisfying differential privacy, wherein each community contains a plurality of nodes; extracting degree information of the nodes in each community, and processing the degree information through a Laplace mechanism to obtain updated degree information, wherein the degree information is an association relationship between each node and other nodes, and the other nodes are nodes in the interaction graph except the node; calculating the edge probability between each node and other nodes through an edge probability calculation model and the updated degree information, and reconstructing the initial interaction graph based on the edge probability to obtain the interaction graph.

[0009] Optionally, the dividing all nodes in the initial interaction graph into a plurality of communities satisfying differential privacy comprises: randomly dividing all nodes in the initial interaction graph into a plurality of initial communities, and merging all nodes in each initial community into a supernode, to obtain a supernode interaction graph containing weights, wherein the weight is the number of nodes contained in the supernode; performing perturbation on the supernode interaction graph through a Laplace mechanism, to obtain a perturbed supernode interaction graph; re-dividing the supernode interaction graph into a plurality of communities through a preset community discovery algorithm; determining the community to which each node belongs through an exponential mechanism, and adjusting the nodes contained in the plurality of communities based on each belonging community, to obtain the plurality of communities satisfying differential privacy.

[0010] Optionally, the reconstructing the initial interaction graph based on the edge probability comprises: judging whether the edge probability between each node and other nodes is greater than or equal to a probability threshold; and in the case that the edge probability is greater than or equal to the probability threshold, connecting a new edge between the node and other nodes, to obtain the reconstructed interaction graph.

[0011] Optionally, the extracting the first feature vector from the interaction graph of the video platform comprises: for each target node corresponding to a user in the interaction graph, updating the node feature of each target node through N iterations, to obtain the first feature vector of the target node, wherein N is a positive integer; each iteration update of the node state of the target node comprises: aggregating the features of the nodes adjacent to the target node, to obtain the global feature of the target node; and updating the node feature of the target node through a preset node update function and the global feature.

[0012] Optionally, after the similar financial product is recommended to the user, the method further comprises: obtaining feedback information of the user on the recommended financial product, wherein the feedback information at least comprises one of the following: click rate, conversion rate and satisfaction; and updating the interaction graph based on the feedback information.

[0013] To achieve the above object, according to another aspect of the present application, a product recommendation device is provided. The device comprises: an extraction unit configured to extract a first feature vector from an interaction graph of a video platform, wherein the interaction graph is an interaction graph between all users of the video platform and videos browsed by the users, the interaction graph comprises a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction graph; a determination unit configured to determine a plurality of financial products to be recommended, and extract a second feature vector for each financial product; a calculation unit configured to, for each user, calculate a similarity between the first feature vector and the second feature vector of each financial product; and a recommendation unit configured to, in a case where the similarity is greater than or equal to a similarity threshold, recommend the financial product corresponding to the similarity greater than or equal to the similarity threshold to the user.

[0014] According to the present application, the following steps are adopted: extracting a first feature vector from an interaction graph of a video platform, wherein the interaction graph is an interaction graph between all users of the video platform and videos browsed by the users, the interaction graph comprises a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction graph; determining a plurality of financial products to be recommended, and extracting a second feature vector for each financial product; for each user, calculating a similarity between the first feature vector and the second feature vector of each financial product; and in a case where the similarity is greater than or equal to a similarity threshold, recommending the financial product corresponding to the similarity greater than or equal to the similarity threshold to the user, thereby solving the problem of low accuracy of recommended financial products in the video platform in the related art. By constructing an interaction graph between users and videos of the video platform, extracting a first feature vector of a user based on the interaction graph, and extracting a second feature vector of a financial product to be recommended, the financial product is recommended to the user based on the similarity between the first feature vector and the second feature vector, thereby achieving the effect of improving the accuracy of the recommended financial product in the video platform. BRIEF DESCRIPTION OF DRAWINGS

[0015] The accompanying drawings, which form a part of the present application, are intended to provide further understanding of the present application, and are used to interpret the illustrative embodiments of the present application and their descriptions, and do not constitute improper limitations to the present application. In the drawings:

[0016] Figure 1 is a flowchart of a product recommendation method provided according to an embodiment of the present application;

[0017] Figure 2 is a schematic diagram of a financial product recommendation system provided according to an embodiment of the present application;

[0018] Figure 3 is a flowchart of an optional product recommendation method provided according to an embodiment of the present application;

[0019] Figure 4 is a schematic diagram of a product recommendation device provided according to an embodiment of the present application;

[0020] Figure 5 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] It should be noted that the embodiments and features of the embodiments in the present application can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0022] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0023] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units need not be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, analyzed data, etc.) involved in the present disclosure are all information and data authorized by the user or authorized by all parties.

[0025] For ease of description, some nouns or terms related to the embodiments of the present application are described below:

[0026] PrivGraph algorithm: an algorithm for generating graph data satisfying differential privacy using community information.

[0027] GNN model: Graph Neural Network, a deep learning model based on graph data structure. It is mainly used for modeling, learning and predicting graph data, and has wide applications in social networks, recommendation systems, bioinformatics, etc.

[0028] Adjusted cosine similarity: a measure of the similarity between two vectors.

[0029] Laplacian mechanism perturbation: a differential privacy algorithm applied to privacy protection. It adds some noise to the real data to achieve the purpose of protecting privacy.

[0030] Louvain method: also known as Modularity Optimization Algorithm, a community detection algorithm. The algorithm aims to divide a large network into several communities, so that the connections within the community are tight and the connections between the communities are sparse, thus better revealing the structure and characteristics of the network.

[0031] CL model: Kronecker Product model, a probabilistic model for generating graphs, which can be used to describe the probability of edges between nodes.

[0032] The application will be described below in combination with the preferred implementation steps, Figure 1 The flowchart of the product recommendation method provided by the embodiments of the application is shown in Figure 1 The method comprises the following steps:

[0033] Step S101, extracting a first feature vector from an interaction relationship graph of a video platform, wherein the interaction relationship graph is an interaction relationship graph between all users of the video platform and videos browsed by the users, the interaction relationship graph comprises a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction relationship graph.

[0034] Specifically, the video platform can be a video playing software for a user to browse videos, the video platform can push advertisements of financial products to the user browsing the videos, the database of the video platform stores user information of the user using the video platform, the user information includes identity information and interaction behaviors of the user, the identity information is for example, the region, age and gender of the user, and the interaction behaviors are for example, the operations of the user such as liking, commenting, sharing and collecting the videos, and the user following the creators publishing the videos. The interaction relationship graph can be a correlation relationship graph constructed based on the user information. In the interaction relationship graph, the videos or the users are taken as nodes, and a side is connected between two nodes with interaction behaviors. The first feature vector can be a preference feature of each user when brushing the videos, extracted from the interaction relationship graph, for example, the first feature vector of user a is extracted as game, because user a often brushes game related videos.

[0035] Step S102, determining a plurality of financial products to be recommended, and extracting a second feature vector for each financial product.

[0036] Specifically, the financial products to be recommended can be financial products provided by the cooperative merchants of the video platform, such as consumer loans and financial products, and the features of the financial products and the contents targeted by the financial products are extracted from the second feature vector. Since the values of the first feature vector and the second feature vector can not be in the same order of magnitude, it is necessary to normalize the first feature vector and the second feature vector, for example, 0-1 normalization or z-score normalization.

[0037] Step S103, for each user, calculating the similarity between the first feature vector and the second feature vector of each financial product.

[0038] Specifically, the similarity between the first feature vector of each user and the second feature vector of each financial product is calculated by adjusting the cosine similarity. The formula is as follows:

[0039]

[0040] Wherein, A i represents the value of the i-th element in the first feature vector, B i represents the value of the i-th element in the second feature vector, represents the average value of all elements in the first feature vector, represents the average value of all elements in the second feature vector, and n is the vector length of the first feature vector or the second feature vector. Sim(A, B) represents the similarity between the first feature vector of each user and the second feature vector of each financial product.

[0041] It should be noted that the value range of the adjusted cosine similarity is -1 to 1, wherein 1 represents complete similarity, -1 represents complete dissimilarity, and 0 represents irrelevance.

[0042] In step S104, in a case where the similarity is greater than or equal to the similarity threshold value, the financial product corresponding to the similarity greater than or equal to the similarity threshold value is recommended to the user.

[0043] Specifically, by setting a recommendation threshold k, that is, the similarity threshold value, when sim(A, B) is greater than or equal to k, the financial product corresponding to the similarity can be recommended to the user.

[0044] The product recommendation method provided in the embodiments of the present application extracts a first feature vector from an interaction relationship graph of a video platform, wherein the interaction relationship graph is an interaction relationship graph between all users of the video platform and videos browsed by the users, the interaction relationship graph includes a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction relationship graph; a plurality of financial products to be recommended are determined, and a second feature vector is extracted for each financial product; for each user, the similarity between the first feature vector and the second feature vector of each financial product is calculated; in a case where the similarity is greater than or equal to a similarity threshold value, the financial product corresponding to the similarity greater than or equal to the similarity threshold value is recommended to the user, thereby solving the problem of low accuracy of recommended financial products in the video platform in the related art. By constructing the interaction relationship graph between the users and the videos of the video platform, extracting the first feature vector of the user based on the interaction relationship graph, and extracting the second feature vector of the financial product to be recommended, the financial product is recommended to the user based on the similarity between the first feature vector and the second feature vector, thereby achieving the effect of improving the accuracy of the recommended financial product in the video platform.

[0045] Before extracting the first feature vector, the interaction relationship graph needs to be constructed first. Optionally, in the product recommendation method provided in the embodiments of the present application, the user information of each user of the video platform is cached in the database after encryption, and the user information includes identity information and interaction behavior. The interaction relationship graph is determined in the following manner: decrypting the encrypted user information to obtain the identity information and the interaction behavior, and determining the videos browsed by each user; determining each user as a user node in the interaction relationship graph, and determining each video as a video node in the interaction relationship graph; connecting an edge between two nodes with a first interaction behavior, and connecting an edge between two nodes with a second interaction behavior, to obtain an initial interaction relationship graph, wherein the first interaction behavior is an operation of a user on a video in the process of browsing the video, and the second interaction behavior is an interaction operation between the user and a video creator; performing differential privacy processing on the initial interaction relationship graph to obtain the interaction relationship graph.

[0046] Specifically, the identity information can be basic information of the user himself, such as region, age, gender, and the like. For example, the user IP address shows that the user is in the southern region, and it is currently summer, so the user may have the demand for cool drinks and summer equipment. The label of the video published by the user. For example, the user often publishes game-related videos, so the user may have the demand for game equipment. The interactive behavior can be the label of the video that the user likes, comments, shares, and collects. For example, the user often watches travel strategy videos and has the above operations, so the user may have the demand for travel products. The above is the most common and most influential important data source of the user portrait of the video platform. In order to protect the user privacy, the data needs to be encrypted and cached in the database. When constructing the interaction relationship graph, the user information cached in the database needs to be decrypted first to obtain the identity information and the interactive behavior used to construct the interaction relationship graph.

[0047] In constructing the interaction relationship graph, for example, a user browses three videos A, B and C, and follows the author b of the B video, collects and likes the C video, and the interaction relationship graph includes five nodes of the a user, the b user, the A video, the B video and the C video. The node corresponding to the a user is connected with the node corresponding to the b user, the node corresponding to the B video and the node corresponding to the C video, respectively, and the b user is connected with the B video. Since the information of the nodes and the edges in the interaction relationship graph also covers the user's privacy data, in order to protect the user's privacy, the initial interaction relationship graph is processed by differential privacy after being constructed, and the interaction relationship graph that can protect the user's privacy is obtained. By constructing the interaction relationship graph, data is provided for extracting the feature vector.

[0048] In order to protect the user privacy in the interaction relationship graph, the initial interaction relationship graph needs to be processed by differential privacy. Optionally, in the product recommendation method provided in the embodiments of the present application, the initial interaction relationship graph is processed by differential privacy to obtain the interaction relationship graph, including: dividing all nodes in the initial interaction relationship graph into a plurality of communities that satisfy differential privacy, wherein each community contains a plurality of nodes; extracting the degree information of the nodes in each community, and processing the degree information through the Laplace mechanism to obtain updated degree information, wherein the degree information is the association relationship between each node and other nodes, and the other nodes are nodes other than the node in the interaction relationship graph; calculating the edge probability between each node and other nodes through an edge probability calculation model and the updated degree information, and reconstructing the initial interaction relationship graph based on the edge probability to obtain the interaction relationship graph.

[0049] Specifically, the initial interaction relationship graph can be processed by the PrivGraph algorithm for differential privacy. First, all nodes in the initial interaction relationship graph need to be divided into a plurality of communities satisfying differential privacy. In the process of dividing the communities, the differential privacy processing of the node information in the interaction relationship graph is realized by perturbation through the Laplace mechanism. The degree information, i.e., the edge information between nodes in the interaction relationship graph. After extracting the degree information of the nodes from the communities, the degree information is processed again through the Laplace mechanism. Through the CL model (Kronecker Product, Kronecker Product model), i.e., the edge probability calculation model and the degree information, the edge probability between each node and other nodes is calculated. A new edge is connected between two nodes whose edge probability is greater than or equal to a probability threshold, thereby reconstructing the initial interaction relationship graph to obtain an interaction relationship graph satisfying differential privacy processing. The embodiment processes the initial interaction relationship graph for differential privacy, thereby protecting the privacy of users in the interaction relationship graph.

[0050] Optionally, in the product recommendation method provided in the embodiment of the application, dividing all nodes in the initial interaction relationship graph into a plurality of communities satisfying differential privacy comprises: randomly dividing all nodes in the initial interaction relationship graph into a plurality of initial communities, and merging all nodes in each initial community into a super node to obtain a super node interaction relationship graph containing weights, wherein the weight is the number of nodes contained by the super node; perturbing the super node interaction relationship graph through the Laplace mechanism to obtain a perturbed super node interaction relationship graph; re-dividing the super node interaction relationship graph into a plurality of communities through a preset community discovery algorithm; determining the community to which each node belongs through the exponential mechanism, and adjusting the nodes contained in the plurality of communities based on each belonging community to obtain a plurality of communities satisfying differential privacy.

[0051] Specifically, all nodes can be divided into communities satisfying differential privacy by PrivGraph. Community division includes community initialization and community adjustment. In community initialization, the original nodes in the initial interaction relationship graph are randomly divided into several communities, the nodes in the same community are merged into a super node, and then a super node graph with weights is formed; in order to guarantee the differential privacy of user data, the super node interaction relationship graph is perturbed by the Laplace mechanism; then, the Louvain algorithm, i.e., the community discovery algorithm, is used to detect the communities of the perturbed super node interaction relationship graph, thereby obtaining the initial community division. In community adjustment, the utility function is set as the number of edges from the node to each community, and the exponential mechanism is used to select the final community of all nodes to reduce the influence caused by the random division of the nodes at the beginning.

[0052] Next, the degree information within and between communities is disturbed and extracted respectively. For the edges within the community, the degree information of the node within the community is extracted, and Laplace mechanism and post-processing are used to ensure differential privacy. For the edges between different communities, the number of edges between the two communities is extracted, and Laplace mechanism and post-processing are used for disturbance. The embodiment divides all nodes in the initial interaction relationship graph into multiple communities that meet differential privacy, and then facilitates the extraction of degree information that can protect user privacy.

[0053] Optionally, in the product recommendation method provided by the embodiment of the application, the initial interaction relationship graph is reconstructed based on the edge probability to obtain the interaction relationship graph, including: judging whether the edge probability between each node and other nodes is greater than or equal to a probability threshold; in the case that the edge probability is greater than or equal to the probability threshold, connecting a new edge between the node and other nodes to obtain the reconstructed interaction relationship graph.

[0054] Specifically, the degree information extracted in the previous step is used to reconstruct within and between communities. For the community, the CL model is used to calculate the edge probability between nodes and reconstruct the nodes within the community. For the community, the edge probability between nodes is calculated based on the extracted degree information and the number of nodes in the two communities and reconstructed. By merging the reconstructed edges within and between communities, the final reconstructed interaction relationship graph can be obtained. The embodiment protects the user privacy in the interaction relationship graph by reconstructing the interaction relationship graph.

[0055] The first feature vector is extracted by the graph neural network model. Optionally, in the product recommendation method provided by the embodiment of the application, the first feature vector is extracted from the interaction relationship graph of the video platform, including: for each target node corresponding to a user in the interaction relationship graph, updating the node feature of each target node through N iterations to obtain the first feature vector of the target node, wherein N is a positive integer; each iteration update of the node state of the target node includes: aggregating the features of the adjacent nodes of the target node to obtain the global feature of the target node; updating the node feature of the target node by using a preset node update function and the global feature.

[0056] Specifically, the graph data in the reconstructed interaction relationship graph is learned and feature extracted using a graph neural network model. The structure of the graph neural network model can be a classical graph neural network model such as GCN (Graph Convolutional Network) or GraphSAGE (Graph Sample and aggreGate). The steps of using the graph neural network model to extract features of the graph data include: first, determining the representation form of the graph, and converting the graph data of the interaction relationship graph into the form of nodes and edges to meet the input requirements of the graph neural network model. Then, a node update function is defined. The graph neural network model updates the state of the node by the neighbor nodes of the node at any depth, that is, the features of the nodes adjacent to the target node are aggregated to obtain the global features of the target node. Generally, some node update functions such as graph convolutional layers and pooling layers are used to extract features of the nodes. Next, all nodes in the interaction relationship graph are iteratively updated. The node update function is applied to each node, and all nodes are iteratively updated. After the iterative update, each node obtains a vector representing its features, that is, a first feature vector. The first feature vectors of all nodes are summarized to serve as the final representation of the interaction relationship graph. The first feature vector is extracted to provide data for calculating the similarity between the features of the user and the features of the financial product.

[0057] Optionally, in the product recommendation method provided in the embodiments of the present application, after the financial product corresponding to the similarity is recommended to the user, the method further includes: obtaining feedback information of the user on the recommended financial product, wherein the feedback information at least includes one of the following: click rate, conversion rate and satisfaction; and updating the interaction relationship graph based on the feedback information.

[0058] Specifically, after the financial product is recommended to the user, the interaction relationship graph is updated in time by collecting the feedback information of the user on the financial product, so as to improve the accuracy of the first feature vector extracted from the interaction relationship graph.

[0059] According to another embodiment of the present application, a financial product recommendation system is also provided, Figure 2 is a schematic diagram of the financial product recommendation system provided by the embodiments of the present application. As Figure 2 shown, the system includes:

[0060] The data acquisition module 201 is configured to acquire the basic information of the user of the video platform and the interactive behaviors of the user on the video such as likes, comments, collections and shares as important data sources of the system.

[0061] The data preprocessing module 202 has two parts: one part is to process the user data of the video platform by PrivGraph, and the other part is to vectorize the financial product.

[0062] The graph neural network model training module 203: using a graph neural network model to learn and extract features from the reconstructed graph data, wherein the structure of the graph neural network model can be a classic GNN model such as GCN, GraphSAGE, etc.

[0063] The financial product recommendation module 204: selecting a financial product vector that meets the conditions in the data storage module 206, and calculating the similarity between the financial product features and the user features using an adjusted cosine similarity.

[0064] The feedback evaluation module 205: collects feedback information of users on recommended financial products, including click rate, conversion rate, satisfaction, etc., and then updates the financial product library and user behavior data graph according to the user feedback information.

[0065] The data storage module 206: stores the input data and output data of the remaining modules.

[0066] The financial product library 207: stores the original data of the financial products.

[0067] The financial product recommendation system provided by the embodiments of the present application optimizes the performance of the graph data processed by differential privacy in downstream tasks, i.e., financial product recommendation, while improving user privacy security. The combination of GNN model for feature extraction effectively improves the financial product matching and recommendation effect.

[0068] According to another embodiment of the present application, an optional product recommendation method applied to the financial product recommendation system is also provided, Figure 3 is a flowchart of the optional product recommendation method provided by the embodiments of the present application. As Figure 3 shown, the method comprises:

[0069] Step 301: encrypt and cache the user information and interactive behavior in the viewing process of the video platform.

[0070] Step 302: decrypt and convert the cached data into graph data, and then use PrivGraph to process differential privacy and reconstruct a new graph that satisfies differential privacy.

[0071] Step 303: vectorize the financial product feature vectors in the financial product library to obtain the feature vectors of the financial products.

[0072] Step 304: use a graph neural network model to extract features from the graph data to obtain the feature vectors of the graph nodes.

[0073] Step 305: perform similarity matching between the feature vectors of the nodes and the feature vectors of the financial products, set a recommendation threshold of 0.8, and when the similarity is greater than k, recommend financial products to the node users or the community where the nodes are located.

[0074] Step 306: Collect user feedback information on recommended financial products, including click rate, conversion rate, satisfaction, etc., and then update the financial product library and user behavior data graph according to the user feedback information.

[0075] The optional product recommendation method provided by the embodiment of the application optimizes the performance of the graph data processed by differential privacy in downstream tasks, i.e., financial product recommendation, while improving user privacy security. The GNN model is combined for feature extraction, which effectively improves the financial product matching and recommendation effect.

[0076] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0077] The embodiment of the application also provides a product recommendation device. It should be noted that the product recommendation device of the embodiment of the application can be used to execute the product recommendation method provided by the embodiment of the application. The product recommendation device provided by the embodiment of the application is introduced as follows.

[0078] Figure 4 is a schematic diagram of the product recommendation device provided by the embodiment of the application. As shown in Figure 4 , the device includes:

[0079] The extraction unit 401 is configured to extract a first feature vector from an interaction relationship graph of a video platform, wherein the interaction relationship graph is an interaction relationship graph between all users of the video platform and videos browsed by the users, the interaction relationship graph includes a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction relationship graph.

[0080] The determination unit 402 is configured to determine a plurality of financial products to be recommended, and extract a second feature vector for each financial product.

[0081] The calculation unit 403 is configured to calculate, for each user, a similarity between the first feature vector and the second feature vector of each financial product.

[0082] The recommendation unit 404 is configured to recommend, in a case where the similarity is greater than or equal to a similarity threshold, a financial product corresponding to the similarity greater than or equal to the similarity threshold to the user.

[0083] The product recommendation device provided in the embodiments of the present application extracts a first feature vector from an interaction relationship graph of a video platform through an extraction unit 401, wherein the interaction relationship graph is an interaction relationship graph between all users of the video platform and videos browsed by the users, the interaction relationship graph includes a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction relationship graph; a determination unit 402 determines a plurality of financial products to be recommended and extracts a second feature vector for each financial product; a calculation unit 403 calculates, for each user, a similarity between the first feature vector and the second feature vector of each financial product; and a recommendation unit 404 recommends, in a case where the similarity is greater than or equal to a similarity threshold, a financial product corresponding to the similarity greater than or equal to the similarity threshold to the user, thereby solving the problem of low accuracy of a recommended financial product in the related art, constructing an interaction relationship graph between users and videos of the video platform, extracting a first feature vector of a user based on the interaction relationship graph, extracting a second feature vector of a financial product to be recommended, and recommending a financial product to a user based on the similarity between the first feature vector and the second feature vector, thereby achieving the effect of improving the accuracy of a recommended financial product in the video platform.

[0084] Optionally, in the product recommendation device provided in the embodiments of the present application, user information of each user of the video platform is cached in a database in an encrypted manner, the user information includes identity information and interaction behaviors, the extraction unit 401 includes: a decryption module configured to decrypt the encrypted user information to obtain the identity information and the interaction behaviors and determine videos browsed by each user; a determination module configured to determine each user as a user node in the interaction relationship graph and determine each video as a video node in the interaction relationship graph; a connection module configured to connect an edge between two nodes having a first interaction behavior and connect an edge between two nodes having a second interaction behavior to obtain an initial interaction relationship graph, wherein the first interaction behavior is an operation of a user to a video in a process of browsing the video, and the second interaction behavior is an interaction operation between the user and a video creator; and a privacy processing module configured to perform differential privacy processing on the initial interaction relationship graph to obtain the interaction relationship graph.

[0085] Optionally, in the product recommendation apparatus provided by the embodiment of the present application, the privacy processing module comprises: a division sub-module, configured to divide all nodes in the initial interaction relationship graph into a plurality of communities satisfying differential privacy, wherein each community contains a plurality of nodes; an extraction sub-module, configured to extract degree information of nodes in each community, and process the degree information through a Laplace mechanism to obtain updated degree information, wherein the degree information is an association relationship between each node and other nodes, and the other nodes are nodes other than the node in the interaction relationship graph; and a calculation sub-module, configured to calculate an edge probability between each node and other nodes through an edge probability calculation model and the updated degree information, and reconstruct the initial interaction relationship graph based on the edge probability to obtain the interaction relationship graph.

[0086] Optionally, in the product recommendation apparatus provided by the embodiment of the present application, the division sub-module comprises: a first community division module, configured to randomly divide all nodes in the initial interaction relationship graph into a plurality of initial communities, and combine all nodes in each initial community into a supernode to obtain a supernode interaction relationship graph containing weights, wherein the weight is the number of nodes contained by the supernode; a perturbation module, configured to perturb the supernode interaction relationship graph through a Laplace mechanism to obtain a perturbed supernode interaction relationship graph; a second community division module, configured to redivide the supernode interaction relationship graph into a plurality of communities through a preset community discovery algorithm; and a node adjustment module, configured to determine the community to which each node belongs through an exponential mechanism, and adjust the nodes contained in the plurality of communities based on each belonging community to obtain the plurality of communities satisfying differential privacy.

[0087] Optionally, in the product recommendation apparatus provided by the embodiment of the present application, the calculation sub-module comprises: a judgment module, configured to judge whether the edge probability between each node and other nodes is greater than or equal to a probability threshold; and a reconstruction module, configured to connect a new edge between the node and other nodes to obtain a reconstructed interaction relationship graph in the case that the edge probability is greater than or equal to the probability threshold.

[0088] Optionally, in the product recommendation apparatus provided by the embodiment of the present application, the extraction unit 401 comprises: an updating module, configured to update the node feature of each target node corresponding to each user in the interaction relationship graph through N times of iteration to obtain a first feature vector of the target node, wherein N is a positive integer; and each iteration update of the node state of the target node comprises: aggregating the features of nodes adjacent to the target node to obtain a global feature of the target node; and updating the node feature of the target node through a preset node updating function and the global feature.

[0089] Optionally, in the product recommendation device provided in the embodiments of this application, the device further includes: an acquisition unit, used to acquire user feedback information on recommended financial products, wherein the feedback information includes at least one of the following: click-through rate, conversion rate, and satisfaction; and an update unit, used to update the interaction relationship graph based on the feedback information.

[0090] The product recommendation device includes a processor and a memory. The extraction unit 401, the determination unit 402, the calculation unit 403, and the recommendation unit 404 are all stored in the memory as program units. The processor executes the program units stored in the memory to achieve the corresponding functions.

[0091] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can improve the accuracy of financial product recommendations on the video platform.

[0092] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0093] This invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements a product recommendation method.

[0094] This invention provides a processor for running a program, wherein the program executes a product recommendation method during runtime.

[0095] Figure 5 This is a schematic diagram of an electronic device provided according to an embodiment of this application. For example... Figure 5 As shown, the electronic device 501 includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: extracting a first feature vector from the interaction relationship graph of the video platform, wherein the interaction relationship graph is a graph of the interaction relationships between all users of the video platform and the videos viewed by the users, the interaction relationship graph includes multiple nodes and edges connecting the nodes, each node corresponds to a user or a video, and each edge represents a user's interaction behavior with a video or other users, and the first feature vector is the feature vector of the node corresponding to the user in the interaction relationship graph; determining multiple financial products to be recommended, and extracting a second feature vector for each financial product; for each user, calculating the similarity between the first feature vector and the second feature vector of each financial product; and recommending the financial products corresponding to similarities greater than or equal to the similarity threshold to the user if the similarity is greater than or equal to the similarity threshold. The device in this article can be a server, PC, PAD, mobile phone, etc.

[0096] The application further provides a computer program product suitable for executing the program of the following method steps when executed on a data processing device: extracting a first feature vector from an interaction graph of a video platform, wherein the interaction graph is an interaction graph between all users of the video platform and videos browsed by the users, the interaction graph comprises a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction graph; determining a plurality of financial products to be recommended, and extracting a second feature vector for each financial product; for each user, calculating a similarity between the first feature vector and the second feature vector of each financial product; and in a case where the similarity is greater than or equal to a similarity threshold, recommending the financial product corresponding to the similarity greater than or equal to the similarity threshold to the user.

[0097] Those skilled in the art will understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) containing computer-usable program code.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that carries out the functions specified in one or more flows and / or blocks.

[0099] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 an apparatus that carries out the functions specified in one or more flows and / or blocks.

[0100] These computer program instructions can also be loaded into a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 Figure 1

[0101] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0102] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) about which the processor can execute instructions. The memory can also include non-volatile memory, such as read only memory (ROM), electrically programmable read only memory (EPROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), erasable programmable read only memory (EPROM), flash memory, or a combination of non-volatile memories in different types. The memory can also include a compact disk read only memory (CD-ROM), digital versatile disk (DVD), Blu-ray, or another non-transitory computer readable medium, which is non-volatile and non-transitory in nature, but volatile in that it can lose its content if the power to the computer is turned off or if the computer crashes. The memory is an example of a computer readable medium.

[0103] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically programmable read only memory (EEPROM), flash memory or other memory technology, compact disk read only memory (CD-ROM), digital versatile disk (DVD), or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media such as modulated data signals and carriers.

[0104] It should also be noted that the terms "comprising," "including," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements recited, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0105] ​​Those skilled in the art will appreciate that embodiments of the present application can be devised for a method, a system, or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) embodying computer-readable program code thereon for use by or in connection with an instruction execution system. For the purposes of this description, a computer-usable or computer readable storage medium can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.

[0106] The foregoing is merely illustrative of the embodiments of this application, and is not intended to limit the application. Numerous variations and modifications can be possible to the embodiments without departing from the spirit and scope of the application. Any equivalent modifications or variations, made within the spirit and scope of the application, should be considered within the scope of the application.

Claims

1. A product recommendation method characterized by, The method comprises the following steps: extracting a first feature vector from an interaction graph of a video platform, wherein the interaction graph is an interaction graph between all users of the video platform and videos browsed by the users, the interaction graph comprises a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user to a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction graph; determining a plurality of financial products to be recommended, and extracting a second feature vector for each financial product; for each user, calculating the similarity between the first feature vector and the second feature vector of each financial product; in the case where the similarity is greater than or equal to a similarity threshold, recommending the financial product corresponding to the similarity greater than or equal to the similarity threshold to the user; the user information of each user of the video platform is cached in a database in an encrypted manner, the user information comprises identity information and interaction behavior, and the interaction graph is determined by the following method: decrypting the encrypted user information to obtain the identity information and the interaction behavior, and determining the videos browsed by each user; each user is determined as a user node in the interaction graph, and each video is determined as a video node in the interaction graph; connecting an edge between two nodes with a first interaction behavior, and connecting an edge between two nodes with a second interaction behavior, to obtain an initial interaction graph, wherein the first interaction behavior is an operation of a user on a video during browsing the video, and the second interaction behavior is an interaction operation between a user and a video creator; performing differential privacy processing on the initial interaction graph to obtain the interaction graph; wherein the differential privacy processing on the initial interaction graph to obtain the interaction graph comprises: dividing all nodes in the initial interaction graph into a plurality of communities satisfying differential privacy, wherein each community contains a plurality of nodes; extracting degree information of nodes in each community, and processing the degree information through a Laplace mechanism to obtain updated degree information, wherein the degree information is the association relationship between each node and other nodes in the interaction graph, and the other nodes are nodes other than the node in the interaction graph; calculating the edge probability between each node and the other nodes through an edge probability calculation model and the updated degree information, and reconstructing the initial interaction graph based on the edge probability to obtain the interaction graph.

2. The method of claim 1, wherein, dividing all nodes in the initial interaction graph into a plurality of communities satisfying differential privacy comprises: randomly dividing all nodes in the initial interaction graph into a plurality of initial communities, and merging all nodes in each initial community into a super node to obtain a super node interaction graph containing weights, wherein the weight is the number of nodes contained in the super node; perturbing the super node interaction graph through the Laplace mechanism to obtain a perturbed super node interaction graph; re-dividing the super node interaction graph into a plurality of communities through a preset community discovery algorithm; Each node belongs to a community is determined by an exponential mechanism, and the nodes contained in the plurality of communities are adjusted based on each belonging community, to obtain a plurality of communities satisfying differential privacy.

3. The method of claim 1, wherein, The initial interaction relationship graph is reconstructed based on the edge probability, to obtain the interaction relationship graph including: It is judged whether the edge probability between each node and other nodes is greater than or equal to a probability threshold; In the case that the edge probability is greater than or equal to the probability threshold, a new edge is connected between the node and the other nodes, to obtain a reconstructed interaction relationship graph.

4. The method of claim 1, wherein, The first feature vector is extracted from the interaction relationship graph of the video platform, including: For each target node corresponding to each user in the interaction relationship graph, the node feature of each target node is updated through N iterations, to obtain the first feature vector of the target node, wherein N is a positive integer; Each iteration update of the node state of the target node includes: aggregating the features of the nodes adjacent to the target node to obtain the global feature of the target node; The node feature of the target node is updated by a preset node update function and the global feature.

5. The method of claim 1, wherein, After the financial product corresponding to the similarity is recommended to the user, the method further includes: Obtaining feedback information of the user on the recommended financial product, wherein the feedback information at least includes one of the following: click rate, conversion rate and satisfaction; Updating the interaction relationship graph based on the feedback information.

6. A product recommendation device characterized by comprising: Including: An extraction unit is configured to extract a first feature vector from an interaction relationship graph of a video platform, wherein the interaction relationship graph is an interaction relationship graph between all users of the video platform and videos browsed by the users, the interaction relationship graph includes a plurality of nodes and edges connecting the nodes, each node corresponds to a user or a video, each edge represents an interaction behavior of a user on a video or another user, and the first feature vector is a feature vector of a node corresponding to a user in the interaction relationship graph; A determination unit is configured to determine a plurality of financial products to be recommended, and extract a second feature vector for each financial product; A calculation unit is configured to calculate, for each user, a similarity between the first feature vector and the second feature vector of each financial product; A recommendation unit is configured to recommend, in the case that the similarity is greater than or equal to a similarity threshold, a financial product corresponding to the similarity greater than or equal to the similarity threshold to the user. The user information of each user of the video platform is cached in a database, the user information including identity information and interaction behaviors, the extraction unit including: a decryption module configured to decrypt the encrypted user information to obtain the identity information and the interaction behaviors, and determine videos browsed by each user; a determination module configured to determine each user as a user node in the interaction relationship graph, and determine each video as a video node in the interaction relationship graph; a connection module configured to connect an edge between two nodes having a first interaction behavior, and connect an edge between two nodes having a second interaction behavior, to obtain an initial interaction relationship graph, wherein the first interaction behavior is an operation of a user on a video in the process of browsing the video, and the second interaction behavior is an interactive operation between the user and a video creator; and a privacy processing module configured to perform differential privacy processing on the initial interaction relationship graph to obtain the interaction relationship graph. The privacy processing module includes: a division sub-module configured to divide all nodes in the initial interaction relationship graph into a plurality of communities satisfying differential privacy, wherein each community contains a plurality of nodes; an extraction sub-module configured to extract degree information of nodes in each community, and process the degree information through a Laplace mechanism to obtain updated degree information, wherein the degree information is an association relationship between each node and other nodes, and the other nodes are nodes other than the node in the interaction relationship graph; and a calculation sub-module configured to calculate an edge probability between each node and the other nodes through an edge probability calculation model and the updated degree information, and reconstruct the initial interaction relationship graph based on the edge probability to obtain the interaction relationship graph.

7. A non-volatile storage medium, characterized by The non-volatile storage medium includes a stored program, wherein the program, when executed, controls a device in which the non-volatile storage medium is located to perform the product recommendation method of any one of claims 1 to 5.

8. An electronic device, comprising: The device includes one or more processors and a memory configured to store one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the product recommendation method of any one of claims 1 to 5.

Images