Insurance Recommendation Method, System, Device and Storage Medium Based on Temporal Similarity

Abstract

The present invention relates to artificial intelligence technology, and proposes an insurance recommendation method, system, device and storage medium based on temporal similarity. The method includes: obtaining a relationship topology network at the current time period in a database, and obtaining a user adjacency matrix at a historical sampling moment; obtaining a user temporal feature matrix and a user similarity matrix at the historical sampling moment according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment; obtaining an optimal low-dimensional representation feature corresponding to the current sampling moment based on the user adjacency matrix, user attribute feature matrix and user similarity matrix at the current sampling moment in a trained graph convolutional neural network; and obtaining the insurance purchase tendency of users in the database at the current sampling moment based on a trained support vector machine. The embodiments of the present invention can improve the accuracy of user insurance recommendation.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular, to an insurance recommendation method, system, device, and storage medium based on temporal similarity. Background Art

[0002] With the development of science and technology, obtaining data and using data modeling to solve corresponding problems has become a very common technical means. For example, each insurance purchase platform will collect data such as insurance purchases, acceptance, and records of users, and build an insurance recommendation model based on the collected record data to maximize the recommendation of insurance packages that users are interested in and improve the recommendation conversion rate.

[0003] Currently, the commonly used insurance recommendation methods usually rely on the basic customer data of enterprise platforms and use machine learning-based algorithms to predict the demand for insurance product categories. Existing insurance recommendation methods often recommend suitable insurance packages for users based on the users' personal information. However, the needs and hobbies of users are not static, and the existing insurance recommendation methods cannot update the changes in users' needs and hobbies in a timely manner, resulting in low accuracy of the existing insurance recommendation methods. Summary of the Invention

[0004] The present invention provides an insurance recommendation method, system, device, and storage medium based on temporal similarity, and its main purpose is to effectively incorporate the changes in user attribute information over time during the insurance recommendation process and effectively improve the accuracy of insurance recommendations.

[0005] In a first aspect, an embodiment of the present invention provides an insurance recommendation method based on temporal similarity, including:

[0006] Obtain the relationship topology network in the current period in the database. The relationship topology network in the current period includes relationship topology networks at multiple historical sampling times. The relationship topology network at the historical sampling time includes all users in the database, the social relationships between any two users in the database at the historical sampling time, and the user attribute feature matrix at the historical sampling time. The historical sampling time is obtained by sampling the current period, and the historical sampling time closest to the end time in the current period is used as the current sampling time;

[0007] According to the relationship topology network at the historical sampling time, obtain the user adjacency matrix at the historical sampling time;

[0008] According to the user attribute feature matrix at the historical sampling time and the user attribute feature matrix at the previous historical sampling time, obtain the user temporal feature matrix at the historical sampling time, and according to the user temporal feature matrix at the historical sampling time, obtain the user similarity matrix at the historical sampling time;

[0009] Based on the user adjacency matrix, user attribute feature matrix, and user similarity matrix at the current sampling moment, obtain the optimal low-dimensional representation feature corresponding to the current sampling moment from the trained graph convolutional neural network, where the trained graph convolutional neural network is trained by the user adjacency matrix, user attribute feature matrix, and user similarity matrix at the historical sampling moment;

[0010] Based on the optimal low-dimensional representation feature corresponding to the current sampling moment, obtain the insurance purchase tendency of the users in the database at the current sampling moment based on the trained support vector machine.

[0011] Preferably, obtaining the user time-series feature matrix at the historical sampling moment according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment is achieved through the following formula:

[0012]

[0013] where B t represents the user time-series feature matrix at the t-th historical sampling moment, X t represents the user attribute feature matrix at the t-th historical sampling moment, X t-1 represents the user attribute feature matrix at the (t - 1)-th historical sampling moment, m represents the total number of all historical sampling moments and the current sampling moment, and both t and m are positive integers.

[0014] Preferably, obtaining the user similarity matrix at the historical sampling moment according to the user time-series feature matrix at the historical sampling moment is achieved through the following formula:

[0015]

[0016] where represents the element in the i-th row and j-th column at the t-th historical sampling moment, represents the i-th row vector in the user time-series feature matrix at the t-th historical sampling moment, represents the j-th row vector in the user time-series feature matrix at the t-th historical sampling moment, n represents the number of users in the database, and i, j, and n are all positive integers.

[0017] Preferably, obtaining the optimal low-dimensional representation feature corresponding to the current sampling moment based on the user adjacency matrix, user attribute feature matrix, and user similarity matrix at the current sampling moment from the trained graph convolutional neural network is achieved through the following formula:

[0018]

[0019]

[0020]

[0021] i ≠ j, 1 ≤ i ≤ n, 1 ≤ j ≤ n,

[0022] wherein, the current sampling moment is the m-th historical sampling moment, H m represents the optimal low-dimensional representation feature corresponding to the current sampling moment, ε GCN represents the trained graph convolutional neural network, X m represents the user attribute feature matrix at the current sampling moment, A m represents the user adjacency matrix at the current sampling moment, S m represents the user similarity matrix at the current sampling moment, W m represents the preset weight matrix at the current sampling moment, σ represents the rectified linear unit function, represents the degree matrix, I n represents the identity matrix corresponding to the user similarity matrix at the current sampling moment, represents the reference matrix at the current sampling moment, represents the element in the i-th row and j-th column of the degree matrix, represents the element in the i-th row and j-th column of the reference matrix at the current sampling moment, where i, j, m, and n are all positive integers.

[0023] Preferably, the trained graph convolutional neural network is obtained through the following steps:

[0024] Take the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment as the first training sample;

[0025] Use the first training sample to train the initial graph convolutional neural network. If the loss function or the number of training times corresponding to the initial graph convolutional neural network does not meet the target conditions, adjust the parameters of the initial graph convolutional neural network, and re-train the adjusted initial graph convolutional neural network using the training sample until the loss function or the number of training times corresponding to the adjusted initial graph convolutional neural network meets the target conditions, and obtain the trained graph convolutional neural network.

[0026] Preferably, the target conditions include that the number of training times is equal to the preset number, and the difference between the loss functions corresponding to two adjacent trainings is less than the preset loss threshold.

[0027] Preferably, the trained support vector machine is obtained through the following steps:

[0028] Based on the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment, and based on the trained graph convolutional neural network, obtain the optimal low-dimensional representation feature at the historical sampling moment;

[0029] Use the optimal low-dimensional representation feature at the historical sampling moment as the second training sample and the insurance purchase intention at the historical sampling moment as the training label to train the initial support vector machine to obtain the trained support vector machine.

[0030] In a second aspect, an embodiment of the present invention provides an insurance recommendation system based on temporal similarity, including:

[0031] A topology module for obtaining a relationship topology network in the current period in the database. The relationship topology network in the current period includes relationship topology networks at multiple historical sampling moments. The relationship topology network at the historical sampling moment includes all users in the database, the social relationship between any two users in the database at the historical sampling moment, and the user attribute feature matrix at the historical sampling moment. The historical sampling moment is obtained by sampling the current period, and the historical sampling moment closest to the end moment in the current period is used as the current sampling moment;

[0032] An adjacency module for obtaining the user adjacency matrix at the historical sampling moment according to the relationship topology network at the historical sampling moment;

[0033] A similarity module for obtaining the user temporal feature matrix at the historical sampling moment according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment, and obtaining the user similarity matrix at the historical sampling moment according to the user temporal feature matrix at the historical sampling moment;

[0034] A representation module for obtaining the optimal low-dimensional representation feature corresponding to the current sampling moment based on the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment in the trained graph convolutional neural network, where the trained graph convolutional neural network is trained by the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment;

[0035] A prediction module, configured to obtain the insurance purchase tendency of users in the database at the current sampling moment based on the optimal low-dimensional representation features corresponding to the current sampling moment and the trained support vector machine.

[0036] In a third aspect, an embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned insurance recommendation method based on temporal similarity are implemented.

[0037] In a fourth aspect, an embodiment of the present invention provides a computer storage medium storing a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned insurance recommendation method based on temporal similarity are implemented.

[0038] An embodiment of the present invention provides an insurance recommendation method, system, device, and storage medium based on temporal similarity. By using the user adjacency matrix, user attribute feature matrix, and user similarity matrix at multiple historical sampling moments, a graph convolutional neural network is trained to obtain a trained graph convolutional neural network; using the optimal low-dimensional representation matrix at multiple historical sampling moments, a support vector machine is trained to obtain a trained support vector machine; and the trained graph convolutional neural network and the trained support vector machine are used to predict the insurance purchase tendency of users in the database at the current sampling moment. In the embodiment of the present invention, according to the differences between the user attribute feature matrices in the relationship topology network at historical sampling moments, the user temporal feature attributes at historical sampling moments are obtained. Through the user temporal feature attributes, the change of user personal attribute information over time can be considered, thereby improving the accuracy of insurance recommendation; according to the user temporal feature attributes, a user similarity matrix is obtained, and the message aggregation strategy in the graph convolutional neural network is guided by the user similarity matrix, which can fuse the influence of the change of user personal attribute information on the insurance purchase intention, further improving the accuracy of insurance recommendation; through the relationship topology network, the user social relationship can be integrated into the insurance recommendation method, thereby further improving the accuracy of insurance recommendation. Description of the Drawings

[0039] Figure 1 It is a schematic diagram of an application scenario of an insurance recommendation method based on temporal similarity provided by an embodiment of the present invention;

[0040] Figure 2 It is a flowchart of an insurance recommendation method based on temporal similarity provided by an embodiment of the present invention;

[0041] Figure 3 It is a schematic structural diagram of an insurance recommendation system based on temporal similarity provided by an embodiment of the present invention;

[0042] Figure 4 This is a schematic structural diagram of a computer device provided in an embodiment of the present invention.

[0043] The implementation, functional features and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific Embodiments

[0044] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0045] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, "a plurality" means two or more unless otherwise specifically defined.

[0046] Figure 1 This is a schematic application scenario diagram of an insurance recommendation method based on temporal similarity provided in an embodiment of the present invention. As Figure 1 shown, first, the user inputs all users in the database of the current period on the client side. After the client side obtains all users in the database of the current period, it sends all users in the database of the current period to the server side. After the server side receives all users in the database of the current period, it executes the insurance recommendation method based on temporal similarity.

[0047] It should be noted that the server side can be implemented by an independent server or a server cluster composed of multiple servers. The client side can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc., but is not limited thereto. The client side and the server side can be connected through Bluetooth, USB (Universal Serial Bus), or other communication connection methods, and the embodiments of the present invention do not limit this here.

[0048] The embodiments of the present invention can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (abbreviated as AI) is to use a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, sense the environment, acquire knowledge and use the knowledge to obtain the best results of theory, method, technology and application system.

[0049] The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, robotics, biometric technology, speech processing technology, natural language processing technology, as well as machine learning and deep learning.

[0050] With the gradual intelligentization of the economic society, it is very important to accurately and intelligently predict the potential purchase tendencies of users for different insurance products, which is crucial for enhancing the competitiveness and influence of enterprises in the market. Therefore, by using the social relationships between users and the sequential feature data of users, predicting the purchase tendencies of users for different insurance products at different time periods, and recommending specific insurance products to users according to the prediction results, the marketing ability of enterprises can be effectively improved.

[0051] In the prior art, the main recommendation strategy for insurance products is to rely on the user data recorded by enterprises and adopt some machine learning-based algorithms to predict the demand for insurance product categories. Although some existing machine learning-based methods can improve the efficiency of the prediction process, the existing methods ignore the social relationships between users and also ignore the changes in the personal attribute characteristics of users over time, resulting in low accuracy of the existing recommendation methods.

[0052] In some insurance purchase APPs developed by enterprises, if two users follow each other or are platform friends on the APP, and if one user purchases a certain insurance product, then the user's friends also have a relatively high purchase tendency for that insurance product. Therefore, effectively considering the social data between users can improve the prediction accuracy of users' purchase tendencies for insurance products. In addition, the personal attribute information of users will also change over time, that is, the interests, hobbies, occupations, and social information of users may all change over time, and these changing personal attribute information will also affect users' purchase tendencies for insurance. Therefore, effectively characterizing the sequential change characteristics of user attributes and the social relationships between users is very important for improving the prediction accuracy of product purchase tendencies.

[0053] Therefore, the embodiments of the present invention provide an insurance recommendation method based on temporal similarity, which fully considers the social relationships between users and the change relationship of user personal attributes over time, can quickly and accurately predict the purchase tendencies of users for different insurance products, and according to the prediction results, enterprises can recommend different insurance products to different users, thereby effectively enhancing the marketing ability and market competitiveness of enterprises. Figure 2 The flowchart of an insurance recommendation method based on temporal similarity provided by the embodiments of the present invention is as Figure 2 shown, and the method includes:

[0054] S210. Obtain the relationship topology network for the current time period in the database. The relationship topology network for the current time period includes the relationship topology networks for multiple historical sampling moments. The relationship topology network for a historical sampling moment includes all users in the database, the social relationships between any two users in the database at the historical sampling moment, and the user attribute feature matrix at the historical sampling moment. The historical sampling moment is obtained by sampling the current time period, and the historical sampling moment closest to the end moment in the current time period is used as the current sampling moment.

[0055] First, obtain the users in the database for the current time period. In the embodiments of the present invention, the users in the database can be users who have had insurance purchase records in the enterprise, or potential users who intend to purchase insurance in the enterprise. Specifically, it can be determined according to the actual situation, and the embodiments of the present invention do not make specific limitations in this regard. The current time period refers to the current time segment, which is a current time segment, generally the time segment between a past moment and the current moment. The current time period can be from the past year to the current day, or from the past six months to the current day, or from the past three months to the current day. Specifically, it can be determined according to the actual situation, and the embodiments of the present invention do not make specific limitations in this regard.

[0056] For the users in the database for the current time period, obtain the relationship topology network for the current time period. In the embodiments of the present invention, since it is necessary to use the past user personal attribute feature information to predict the user's insurance purchase tendency at the current moment, therefore, the relationship topology network for the current time period in the embodiments of the present invention includes the relationship topology networks for multiple historical sampling moments, and the historical sampling moment can be obtained by dividing the current time period. For the sake of illustration, in the embodiments of the present invention, taking the division of the past year to the current day to obtain multiple historical sampling moments as an example, multiple historical sampling moments are obtained by sampling the past year, which are the past year, the past eleven months,..., the past month, and the current day respectively. Then, the current day is used as the current sampling moment. Therefore, the relationship topology network for the current day includes the relationship topology network for the past year, the relationship topology network for the past eleven months, the relationship topology network for the past ten months,..., the relationship topology network for the past month, and the relationship topology network for the current day, and so on.

[0057] For the relationship topology network at each historical sampling moment, the relationship topology network includes all users in the database at the historical sampling moment, the social relationships between any two users, and the user attribute feature matrix. Generally, each user is regarded as a node of the relationship topology network at the historical sampling moment; if there is a social relationship between any two users, then there is an edge between the corresponding two nodes in the relationship topology network corresponding to the two users. If there is no social relationship between the two users, then there is no edge between the corresponding nodes in the relationship topology network corresponding to the two users; in addition, in the embodiments of the present invention, if any two users have cooperated, or the number of messages sent to each other and the number of chats exceed a preset threshold, it can be considered that the two users have cooperated. If any two users have not cooperated, or the number of messages sent to each other and the number of chats is less than the preset threshold, it is considered that there is no cooperation between the two users, which can be determined according to the actual situation, and the embodiments of the present invention do not make specific limitations in this regard. The relationship topology network at the historical sampling moment in the embodiments of the present invention further includes a user attribute feature matrix, which is obtained by quantifying user attribute information. The user attribute information is personal information related to the user, and specifically may include information such as age, occupation, hobby, income, insurance purchase information, and working years, which can be determined according to the actual situation, and the embodiments of the present invention do not make specific limitations in this regard.

[0058] In the specific implementation process, construct the relationship topology network G of the current time period T ={G 1 , G 2 , G 3 , …, G m}, where T represents the current time period, the current time period is divided into m historical sampling moments, G 1 represents the relationship topology network of the 1st historical sampling moment, G 2 represents the relationship topology network of the 2nd historical sampling moment, G 3 represents the relationship topology network of the 3rd historical sampling moment, G m represents the relationship topology network of the mth historical sampling moment, and the mth historical sampling moment is used as the current sampling moment.

[0059] Among them, in the embodiments of the present invention, G t =(V, E t , X t ) represents the relationship topology network of the tth (1≤t≤m) historical sampling moment, V={V1, V2, …, V n} represents the nodes corresponding to all users, V1 represents the 1st node, that is, the node corresponding to the 1st user, V ndenotes the nth node, that is, the node corresponding to the nth user, where n represents the number of all users in the database. In the embodiments of the present invention, through it is indicated that there is an edge connection between the ith node and the jth node at the tth historical sampling moment, it is indicated that at the tth historical sampling moment, there is an edge connection between node V i and node V j There is an edge connection. In the embodiments of the present invention, through X t it represents the user attribute feature matrix corresponding to the tth historical sampling moment, where X t ∈R n×f , and each row vector thereof represents the personal attribute information of different users, and f represents the number of attributes owned by each user. In the specific implementation process, the personal attribute information of users can be obtained by collecting the text material information of enterprise users and converting the text materials of each user into vector representation information using the bag-of-words model.

[0060] In the embodiments of the present invention, the relationship topology network at the current time period is predicted through the relationship topology network at the historical sampling moment, and the personal attribute information of users at different historical sampling moments is reflected through the user attribute feature matrix in the relationship topology network.

[0061] S220. According to the relationship topology network at the historical sampling moment, obtain the user adjacency matrix at the historical sampling moment;

[0062] Then, according to the relationship topology network at the historical sampling moment, obtain the user adjacency matrix at the historical sampling moment, that is, for the relationship topology network at each historical sampling moment, obtain the user adjacency matrix at each historical sampling moment. In the embodiments of the present invention, the user adjacency matrix is used to quantify the social relationship between users. Specifically, for the user connection matrix at a certain historical sampling moment, the element in the ith row and jth column of the user adjacency matrix represents the edge connection between the ith user and the jth user in the relationship topology network. If there is an edge connection, the value of the element in the ith row and jth column of the user adjacency matrix is 1; if there is no edge connection, the value of the element in the ith row and jth column of the user adjacency matrix is 0. Therefore, in the embodiments of the present invention, the edge connections in the relationship topology network are quantified to obtain the user adjacency matrix.

[0063] In the specific implementation process, for the relationship topology network G t at the tth historical sampling moment, its corresponding user adjacency matrix is A t , if there is an edge connection between node V i and node V j in this relationship topology network, then the values of the ith row and jth column of the user adjacency matrix A t are 1, otherwise 0. It is easy to understand that node V i and node V jThere is an edge connecting nodes V i and node V j indicating that there is a social relationship between the two users represented by node V.

[0064] S230. According to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment, obtain the user time series feature matrix at the historical sampling moment, and according to the user time series feature matrix at the historical sampling moment, obtain the user similarity matrix at the historical sampling moment;

[0065] For the user attribute feature matrix at each historical sampling moment, according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment of this historical moment, obtain the user time series feature matrix at this historical sampling moment. Specifically, in the embodiments of the present invention, the difference between the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment is used to represent the user time series feature matrix at the historical sampling moment, so as to reflect the change of the user's personal attribute information over time. For each historical sampling moment, it is necessary to calculate the user time series feature matrix corresponding to this historical moment. Therefore, the user time series feature matrix at each historical sampling moment is used to reflect the change of the user's personal attribute information over time.

[0066] And according to the user time series feature matrix at the historical sampling moment, by calculating the similarity between the user time series feature matrix at the historical sampling moment and the user time series feature matrices at other historical sampling moments, obtain the user similarity matrix at the historical sampling moment.

[0067] In the specific implementation process, according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment, obtain the user time series feature matrix at the historical sampling moment, which is realized by the following formula:

[0068]

[0069] where B t represents the user time series feature matrix at the t-th historical sampling moment, X t represents the user attribute feature matrix at the t-th historical sampling moment, X t-1 represents the user attribute feature matrix at the (t - 1)-th historical sampling moment, m represents the total number of all historical sampling moments and the current sampling moment, and both t and m are positive integers.

[0070] In the specific implementation process, according to the user time series feature matrix at the historical sampling moment, obtain the user similarity matrix at the historical sampling moment, which is realized by the following formula:

[0071]

[0072] Among them, represents the element in the i-th row and j-th column at the t-th historical sampling moment, represents the i-th row vector in the user time-series feature matrix at the t-th historical sampling moment, represents the j-th row vector in the user time-series feature matrix at the t-th historical sampling moment, n represents the number of users in the database, and i, j, and n are all positive integers.

[0073] In the embodiments of the present invention, by calculating the cosine similarity between the time-series features of two users, the user similarity matrix is obtained. According to the user time-series feature attributes, the user similarity matrix is obtained. By using the user similarity matrix to guide the message aggregation strategy in the graph convolutional neural network, the influence of changes in user personal attribute information on the insurance purchase intention can be integrated, and the accuracy of insurance recommendation can be further improved.

[0074] S240. Based on the trained graph convolutional neural network, according to the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment, obtain the optimal low-dimensional representation feature corresponding to the current sampling moment, where the trained graph convolutional neural network is trained by the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment;

[0075] In the embodiments of the present invention, through the relationship topology network at the current sampling moment, the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment are obtained, and are input into the trained graph convolutional neural network to obtain the optimal low-dimensional representation feature corresponding to the current sampling moment. In the embodiments of the present invention, the trained graph convolutional neural network is obtained by training the graph convolutional neural network with the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment.

[0076] Specifically, in the embodiments of the present invention, based on the trained graph convolutional neural network, according to the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment, obtain the optimal low-dimensional representation feature corresponding to the current sampling moment. The specific calculation formula is as follows:

[0077]

[0078]

[0079]

[0080] i ≠ j, 1 ≤ i ≤ n, 1 ≤ j ≤ n,

[0081] where the current sampling moment is the m-th historical sampling moment, H m represents the optimal low-dimensional representation feature corresponding to the current sampling moment, ε GCN represents the trained graph convolutional neural network, X m represents the user attribute feature matrix at the current sampling moment, A m represents the user adjacency matrix at the current sampling moment, S m represents the user similarity matrix at the current sampling moment, W m represents the preset weight matrix at the current sampling moment, σ represents the rectified linear unit function, represents the degree matrix, I n represents the identity matrix corresponding to the user similarity matrix at the current sampling moment, represents the reference matrix at the current sampling moment, represents the element in the i-th row and j-th column of the degree matrix, represents the element in the i-th row and j-th column of the reference matrix at the current sampling moment, where i, j, m, and n are all positive integers.

[0082] In the specific implementation process, the trained graph convolutional neural network is obtained by training the graph convolutional neural network using the user adjacency matrix of the previous m - 1 historical sampling moments, the user attribute feature matrix of the previous m - 1 historical sampling moments, and the user similarity matrix of the previous m - 1 historical sampling moments. In the embodiment of the present invention, the m-th historical sampling moment is the time point closest to the end moment of the current period, so the m-th historical sampling moment is used as the current sampling moment.

[0083] For the user adjacency matrix of the previous m - 1 historical sampling moments, the user attribute feature matrix of the previous m - 1 historical sampling moments, and the user similarity matrix of the previous m - 1 historical sampling moments, use these training data to perform unsupervised training on the initial graph convolutional neural network. If the loss function of the training meets the target conditions or the number of training times reaches the preset number of times, the trained graph convolutional neural network can be obtained.

[0084] S250. According to the optimal low-dimensional representation feature corresponding to the current sampling moment, based on the trained support vector machine, obtain the insurance purchase tendency of the users in the database at the current sampling moment.

[0085] After obtaining the trained graph convolutional neural network, input the user adjacency matrix at the first m - 1 historical sampling moments, the user attribute feature matrix at the first m - 1 historical sampling moments, and the user similarity matrix at the first m - 1 historical sampling moments into the trained graph convolutional neural network, and the optimal low-dimensional representation features at the first m - 1 historical sampling moments can be obtained.

[0086] For the optimal low-dimensional representation features at the first m - 1 historical sampling moments, take 70% of them as training data, 30% as test data, and use the insurance purchase situations of users at the first m - 1 historical sampling moments as labels. Train the support vector machine using the training data and labels to obtain the trained support vector machine. Then, test the trained support vector machine using the test data and labels. If the test accuracy is greater than the preset accuracy, the trained support vector machine can be directly used to predict the insurance purchase tendency of users in the current period; if the test accuracy is not greater than the preset threshold, the trained support vector machine needs to be retrained using the training data and labels until the test accuracy meets the requirements.

[0087] After obtaining the trained support vector machine, input the optimal low-dimensional representation features at the m-th historical sampling moment into the trained support vector machine to obtain the current insurance purchase tendency of this user.

[0088] The embodiment of the present invention provides an insurance recommendation method based on temporal similarity. Train a graph convolutional neural network through the user adjacency matrix, user attribute feature matrix, and user similarity matrix at multiple historical sampling moments to obtain the trained graph convolutional neural network; train a support vector machine using the optimal low-dimensional representation matrices at multiple historical sampling moments to obtain the trained support vector machine; and use the trained graph convolutional neural network and the trained support vector machine to predict the insurance purchase tendency of users in the database at the current sampling moment. In the embodiment of the present invention, according to the differences between the user attribute feature matrices in the relationship topology network at historical sampling moments, the user temporal feature attributes at historical sampling moments are obtained. Through these user temporal feature attributes, the change situation of user personal attribute information over time can be considered, thereby improving the accuracy of insurance recommendation; according to the user temporal feature attributes, a user similarity matrix is obtained. By using the user similarity matrix to guide the message aggregation strategy in the graph convolutional neural network, the influence of changes in user personal attribute information on the insurance purchase intention can be integrated, further improving the accuracy of insurance recommendation; through the relationship topology network, the user social relationship can be integrated into this insurance recommendation method, thereby further improving the accuracy of insurance recommendation.

[0089] On the basis of the above embodiment, preferably, the trained graph convolutional neural network is obtained through the following steps:

[0090] Take the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment as the first training sample;

[0091] Use the first training sample to train the initial graph convolutional neural network. If the loss function or the number of training times corresponding to the initial graph convolutional neural network does not meet the target conditions, adjust the parameters of the initial graph convolutional neural network, and re-train the adjusted graph convolutional neural network using the training sample until the loss function or the number of training times corresponding to the adjusted graph convolutional neural network meets the target conditions, and obtain the trained graph convolutional neural network.

[0092] In the embodiment of the present invention, the user adjacency matrix at the first m - 1 historical sampling moments, the user attribute feature matrix at the first m - 1 historical sampling moments, and the user similarity matrix at the first m - 1 historical sampling moments are used as training samples, and the initial graph convolutional neural network is trained unsupervised using these training data. If the loss function of the training meets the target conditions or the number of training times reaches the preset number of times, the trained graph convolutional neural network can be obtained.

[0093] It should be noted that in the embodiment of the present invention, the target conditions are that the number of training times reaches the preset number of times, and the difference between the loss functions corresponding to two adjacent trainings is less than the preset loss threshold. The target conditions in the embodiment of the present invention include two aspects: the number of training times reaches the preset number of times, and the difference between the loss functions corresponding to two adjacent trainings is less than the preset loss threshold. As long as one of the target conditions is met, the training can be ended. It is easy to understand that the number of training times reaching the preset number of times is to avoid the problem of falling into an infinite loop when the loss value cannot meet the conditions all the time; the difference between the loss functions corresponding to two adjacent trainings is less than the preset loss threshold. When the difference between the loss functions between two adjacent trainings is relatively small, it means that the parameters of the graph convolutional neural network have reached a relatively good state, and even if the parameters of the graph convolutional neural network are adjusted again, the loss value cannot be made smaller. Therefore, when the difference between the loss functions between two adjacent trainings is less than the preset loss threshold, the training can be ended.

[0094] On the basis of the above - mentioned embodiment, preferably, the trained support vector machine is obtained through the following steps:

[0095] According to the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment, based on the trained graph convolutional neural network, obtain the optimal low - dimensional representation features at the historical sampling moment;

[0096] Using the optimal low-dimensional representation feature at the historical sampling moment as the second training sample and the insurance purchase intention at the historical sampling moment as the training label, train the initial support vector machine to obtain the trained support vector machine.

[0097] Specifically, in the embodiment of the present invention, after obtaining the trained graph convolutional neural network, input the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment into the trained graph convolutional neural network to obtain the optimal low-dimensional representation feature at the historical sampling moment. Then, use the optimal low-dimensional representation feature at the historical sampling moment as the training sample and the actual insurance purchase intention of the user at the historical sampling moment as the label to train the support vector machine to obtain the trained support vector machine.

[0098] Figure 3 The structure diagram of an insurance recommendation system based on temporal similarity provided by an embodiment of the present invention is shown in Figure 3 As shown, the system includes a topology module 310, an adjacency module 320, a similarity module 330, a representation module 340, and a prediction module 350, where:

[0099] The topology module 310 is used to obtain the relationship topology network in the current period in the database. The relationship topology network in the current period includes relationship topology networks at multiple historical sampling moments. The relationship topology network at the historical sampling moment includes all users in the database, the social relationship between any two users in the database at the historical sampling moment, and the user attribute feature matrix at the historical sampling moment. The historical sampling moment is obtained by sampling the current period, and the historical sampling moment closest to the end moment in the current period is used as the current sampling moment;

[0100] The adjacency module 320 is used to obtain the user adjacency matrix at the historical sampling moment according to the relationship topology network at the historical sampling moment;

[0101] The similarity module 330 is used to obtain the user temporal feature matrix at the historical sampling moment according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment, and obtain the user similarity matrix at the historical sampling moment according to the user temporal feature matrix at the historical sampling moment;

[0102] The representation module 340 is used to obtain the optimal low-dimensional representation feature corresponding to the current sampling moment based on the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment in the trained graph convolutional neural network, where the trained graph convolutional neural network is trained by the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment;

[0103] The prediction module 350 is used to obtain the insurance purchase tendency of the users in the database at the current sampling moment based on the optimal low-dimensional representation feature corresponding to the current sampling moment and the trained support vector machine.

[0104] This embodiment is a system embodiment corresponding to the above method, and its specific implementation process is the same as that of the above method embodiment. For details, please refer to the above method embodiment, and this system embodiment will not be specifically limited here.

[0105] On the basis of the above embodiment, preferably, in the similarity module, the user temporal feature matrix at the historical sampling moment is obtained according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment, and is implemented by the following formula:

[0106]

[0107] where B t represents the user temporal feature matrix at the t-th historical sampling moment, X t represents the user attribute feature matrix at the t-th historical sampling moment, X t-1 represents the user attribute feature matrix at the (t - 1)-th historical sampling moment, m represents the total number of all historical sampling moments and the current sampling moment, and t and m are positive integers.

[0108] On the basis of the above embodiment, preferably, in the similarity module, the user similarity matrix at the historical sampling moment is obtained according to the user temporal feature matrix at the historical sampling moment, and is implemented by the following formula:

[0109]

[0110] where represents the element in the i-th row and j-th column at the t-th historical sampling moment, represents the i-th row vector in the user temporal feature matrix at the t-th historical sampling moment, represents the j-th row vector in the user temporal feature matrix at the t-th historical sampling moment, n represents the number of users in the database, and i, j, and n are all positive integers.

[0111] Based on the above embodiments, preferably, in the representation module, according to the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment, in the trained graph convolutional neural network, the optimal low-dimensional representation feature corresponding to the current sampling moment is obtained, which is realized by the following formula:

[0112]

[0113]

[0114]

[0115] i≠j, 1≤i≤n, 1≤j≤n,

[0116] wherein, the current sampling moment is the m-th historical sampling moment, H m represents the optimal low-dimensional representation feature corresponding to the current sampling moment, ε GCN represents the trained graph convolutional neural network, X m represents the user attribute feature matrix at the current sampling moment, A m represents the user adjacency matrix at the current sampling moment, S m represents the user similarity matrix at the current sampling moment, W m represents the preset weight matrix at the current sampling moment, σ represents the rectified linear unit function, represents the degree matrix, I n represents the identity matrix corresponding to the user similarity matrix at the current sampling moment, represents the reference matrix at the current sampling moment, represents the element in the i-th row and j-th column of the degree matrix, represents the element in the i-th row and j-th column of the reference matrix at the current sampling moment, and i, j, m, n are all positive integers.

[0117] Based on the above embodiments, preferably, the adjacency module includes an adjacency construction unit, wherein:

[0118] The adjacency construction unit is used for the i-th user and the j-th user in the relationship topology network at the historical sampling moment. If there is an edge between the node corresponding to the i-th user and the node corresponding to the j-th user in the relationship topology network, the element in the i-th row and j-th column of the user adjacency matrix at the historical sampling moment is assigned a value of 1; otherwise, the element in the i-th row and j-th column of the user adjacency matrix at the historical sampling moment is assigned a value of 0, where i and j are both positive integers.

[0119] Based on the above embodiments, preferably, the representation module includes a sample unit and a training unit, where:

[0120] The sample unit is used to use the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment as training samples;

[0121] The training unit is used to train the initial graph convolutional neural network with the training samples. If the loss function corresponding to the initial graph convolutional neural network does not meet the target condition, adjust the parameters of the initial graph convolutional neural network, and re-train the initial graph convolutional neural network with the training samples until the loss function corresponding to the adjusted initial graph convolutional neural network meets the target condition, and use the adjusted initial graph convolutional neural network as the trained graph convolutional neural network.

[0122] Based on the above embodiments, preferably, the target condition includes that the number of training times reaches a preset number, or the difference between the loss functions corresponding to two adjacent trainings is less than a preset loss threshold.

[0123] Based on the above embodiments, preferably, the prediction module includes a characterization unit and a vector machine unit, where:

[0124] The standard unit is used to obtain the optimal low-dimensional representation feature at the historical sampling moment based on the trained graph convolutional neural network according to the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment;

[0125] The vector machine unit is used to train the initial support vector machine with the optimal low-dimensional representation feature at the historical sampling moment as the second training sample and the insurance purchase intention at the historical sampling moment as the training label to obtain the trained support vector machine.

[0126] Each module in the above insurance recommendation system based on temporal similarity can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0127] Figure 4 It is a schematic structural diagram of a computer device provided in an embodiment of the present invention. The computer device can be a server, and its internal structure diagram can be as Figure 4As shown in the figure. The computer device includes a processor, a memory, a network interface, and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a computer storage medium and an internal memory. The computer storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the computer storage medium. The database of the computer device is used to store the data generated or obtained during the execution of the insurance recommendation method based on temporal similarity, such as the relationship topology network at the current time period and the trained graph convolutional neural network. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements an insurance recommendation method based on temporal similarity.

[0128] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the insurance recommendation method based on temporal similarity in the above embodiment. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the embodiment of the insurance recommendation system based on temporal similarity.

[0129] In one embodiment, a computer storage medium is provided. A computer program is stored on the computer storage medium. When the computer program is executed by the processor, it implements the steps of the insurance recommendation method based on temporal similarity in the above embodiment. Alternatively, when the computer program is executed by the processor, it implements the functions of each module / unit in the above embodiment of the insurance recommendation system based on temporal similarity.

[0130] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in this application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0131] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional unit and module is used as an example for illustration. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0132] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An insurance recommendation method based on temporal similarity, characterized in that Including: Obtain the relationship topology network in the current time period in the database. The relationship topology network in the current time period includes relationship topology networks at multiple historical sampling moments. The relationship topology network at the historical sampling moment includes all users in the database, the social relationships between any two users in the database at the historical sampling moment, and the user attribute feature matrix at the historical sampling moment. The historical sampling moment is obtained by sampling the current time period, and the historical sampling moment closest to the end moment in the current time period is used as the current sampling moment; According to the relationship topology network at the historical sampling moment, obtain the user adjacency matrix at the historical sampling moment; According to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment, obtain the user time series feature matrix at the historical sampling moment, and according to the user time series feature matrix at the historical sampling moment, obtain the user similarity matrix at the historical sampling moment; Based on the trained graph convolutional neural network, obtain the optimal low-dimensional representation feature corresponding to the current sampling moment according to the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment. The trained graph convolutional neural network is trained by the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment; Based on the trained support vector machine, obtain the insurance purchase tendency of the users in the database at the current sampling moment according to the optimal low-dimensional representation feature corresponding to the current sampling moment; 2. The insurance recommendation method based on temporal similarity according to claim 1, wherein The obtaining of the user time series feature matrix at the historical sampling moment according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment is realized by the following formula: Among them, B t represents the user time-series feature matrix at the t-th historical sampling moment, and X t represents the user attribute feature matrix at the t-th historical sampling moment, and X t-1 represents the user attribute feature matrix at the (t - 1)-th historical sampling moment. m represents the total number of all historical sampling moments and the current sampling moment, and both t and m are positive integers.

3. The insurance recommendation method based on temporal similarity according to claim 1, wherein The obtaining of the user similarity matrix at the historical sampling moment according to the user time series feature matrix at the historical sampling moment is realized by the following formula: Among them, represents the element in the \(i\)-th row and \(j\)-th column at the \(t\)-th historical sampling moment, represents the \(i\)-th row vector in the user time series feature matrix at the \(t\)-th historical sampling moment, represents the \(j\)-th row vector in the user time series feature matrix at the \(t\)-th historical sampling moment, and \(n\) represents the number of users in the database. Both \(i\), \(j\), and \(n\) are positive integers.

4. The insurance recommendation method based on temporal similarity according to claim 1, characterized in that The obtaining of the optimal low-dimensional representation feature corresponding to the current sampling moment based on the trained graph convolutional neural network according to the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment is realized by the following formula: i≠j, 1≤i≤n, 1≤j≤n, where the current sampling moment is the m-th historical sampling moment, H m represents the optimal low-dimensional representation feature corresponding to the current sampling moment, ε GCN represents the trained graph convolutional neural network, X m represents the user attribute feature matrix at the current sampling moment, A m represents the user adjacency matrix at the current sampling moment, S m represents the user similarity matrix at the current sampling moment, W m represents the preset weight matrix at the current sampling moment, σ represents the rectified linear unit function, represents the degree matrix, I n represents the identity matrix corresponding to the user similarity matrix at the current sampling moment, represents the reference matrix at the current sampling moment, represents the element in the i-th row and j-th column of the degree matrix, represents the element in the i-th row and j-th column of the reference matrix at the current sampling moment, where i, j, m, and n are all positive integers.

5. The insurance recommendation method based on temporal similarity according to any one of claims 1 to 4, characterized in that The trained graph convolutional neural network is obtained through the following steps: Take the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment as the first training sample; Train the initial graph convolutional neural network using the first training sample. If the loss function or the number of training times corresponding to the initial graph convolutional neural network does not meet the target conditions, adjust the parameters of the initial graph convolutional neural network, and re - train the adjusted initial graph convolutional neural network using the training sample until the loss function or the number of training times corresponding to the adjusted initial graph convolutional neural network meets the target conditions, and obtain the trained graph convolutional neural network.

6. The insurance recommendation method based on temporal similarity according to claim 5, wherein The target conditions include that the number of training times is equal to the preset number, and the difference between the loss functions corresponding to two adjacent trainings is less than the preset loss threshold.

7. The insurance recommendation method based on temporal similarity according to any one of claims 1 to 4, characterized in that The trained support vector machine is obtained through the following steps: Based on the trained graph convolutional neural network, obtain the optimal low - dimensional representation features at the historical sampling moment according to the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment. Use the optimal low - dimensional representation features at the historical sampling moment as the second training sample and the insurance purchase intention at the historical sampling moment as the training label to train the initial support vector machine and obtain the trained support vector machine.

8. An insurance recommendation system based on temporal similarity, characterized in that, It includes: A topology module for obtaining the relationship topology network in the current period in the database. The relationship topology network in the current period includes relationship topology networks at multiple historical sampling moments. The relationship topology network at the historical sampling moment includes all users in the database, the social relationships between any two users in the database at the historical sampling moment, and the user attribute feature matrix at the historical sampling moment. The historical sampling moment is obtained by sampling the current period, and the historical sampling moment closest to the end moment in the current period is used as the current sampling moment. An adjacency module for obtaining the user adjacency matrix at the historical sampling moment according to the relationship topology network at the historical sampling moment. A similarity module for obtaining the user temporal feature matrix at the historical sampling moment according to the user attribute feature matrix at the historical sampling moment and the user attribute feature matrix at the previous historical sampling moment, and obtaining the user similarity matrix at the historical sampling moment according to the user temporal feature matrix at the historical sampling moment. A representation module for obtaining the optimal low - dimensional representation features corresponding to the current sampling moment based on the trained graph convolutional neural network according to the user adjacency matrix at the current sampling moment, the user attribute feature matrix at the current sampling moment, and the user similarity matrix at the current sampling moment, where the trained graph convolutional neural network is trained by the user adjacency matrix at the historical sampling moment, the user attribute feature matrix at the historical sampling moment, and the user similarity matrix at the historical sampling moment. A prediction module for obtaining the insurance purchase tendency of the users in the database at the current sampling moment based on the trained support vector machine according to the optimal low - dimensional representation features corresponding to the current sampling moment.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the steps of the insurance recommendation method based on temporal similarity as described in any one of claims 1 to 7 are implemented.

10. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the steps of the insurance recommendation method based on temporal similarity as described in any one of claims 1 to 7 are implemented.

Images