Insurance product recommendation method and device, electronic equipment and storage medium

Abstract

The invention provides an insurance product recommendation method and device, electronic equipment and a storage medium, and the method comprises the steps: extracting an explicit feature vector of a to-be-recommended user according to the user information of the to-be-recommended user; based on the social relation of the to-be-recommended user in the community network, extracting an implicit feature vector of the to-be-recommended user; splicing the explicit feature vector and the implicit feature vector of the to-be-recommended user to generate a fused feature vector of the to-be-recommended user; determining a plurality of target reference users based on the similarity between the fusion feature vector of the to-be-recommended user and the fusion feature vector of the reference user; and based on the insurance type purchase list of the target reference user, determining an insurance type recommendation list of the to-be-recommended user.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method, apparatus, electronic device, and storage medium for recommending insurance products. Background Technology

[0002] With the continuous development of the insurance market and the increasing variety of insurance products, users face numerous difficulties when choosing insurance products. Current insurance recommendation systems, drawing on methods from the field of intelligent internet recommendations, rely on users' basic information, browsing history, and purchasing behavior to provide personalized recommendations. However, for insurance companies that still primarily rely on offline marketing methods, user behavioral data is relatively sparse, or even lacks online behavioral data records. This is especially true for new users, where relevant data is essentially nonexistent, resulting in low recommendation accuracy. Summary of the Invention

[0003] The purpose of this application is to provide a method, apparatus, electronic device, and storage medium for recommending insurance products, so as to solve the technical problem of inaccurate insurance product recommendations in existing applications.

[0004] In a first aspect, the present invention provides a method for recommending insurance products. The method includes: extracting an explicit feature vector of a user to be recommended based on the user information of the user to be recommended; extracting an implicit feature vector of the user to be recommended based on the social relationships of the user to be recommended in a community network; concatenating the explicit and implicit feature vectors of the user to be recommended to generate a fused feature vector of the user to be recommended; determining multiple target reference users based on the similarity between the fused feature vector of the user to be recommended and the fused feature vectors of reference users; and determining a list of recommended insurance products for the user to be recommended based on the insurance product purchase lists of the target reference users.

[0005] In an optional implementation, the implicit feature vector is extracted in the following way: Construct a community network graph, which includes a second-order path formed by multiple first-user nodes, community nodes, and second-user nodes; Based on the community network graph, random walks generate multiple node path sequences of preset lengths; The node path sequence is input into the Skip-Gram model, which outputs the corresponding implicit feature vector to train and generate an implicit embedding model.

[0006] In an optional implementation, the implicit feature vector is extracted in the following way: Construct a community network graph, which includes a second-order path formed by multiple first-user nodes, community nodes, and second-user nodes; Based on the community network graph, random walks generate multiple node path sequences of preset lengths; Generate positive and negative samples based on node path sequences; Positive and negative samples are input into the Skip-Gram model, which outputs the corresponding implicit feature vectors to train and generate an implicit embedding model.

[0007] In an optional implementation, user information includes at least the user's basic personal information, policy behavior information, browsing behavior information, and medical claims information, and explicit feature vectors are extracted using the following methods: The system encodes and fuses user basic personal information, policy behavior information, browsing behavior information, and medical claims information to generate explicit feature vectors of specified dimensions.

[0008] In an optional implementation, prior to the step of encoding the user's basic personal information, policy behavior information, browsing behavior information, and medical claims information, the method further includes: Fill in or remove missing values ​​for users' basic personal information, policy behavior information, browsing behavior information, and medical claims information; It also standardizes users' basic personal information, policy behavior information, browsing behavior information, and medical claims information.

[0009] In an optional implementation, reference users are sorted based on similarity, and the top K reference users are selected as target reference users.

[0010] In an optional implementation, the recommended user is a user who has browsed the insurance products purchased by the reference user within a preset time period but has not purchased the insurance product.

[0011] Secondly, the present invention provides an insurance product recommendation device, the device comprising: The first vector extraction module is used to extract the explicit feature vector of the user to be recommended based on the user information of the user to be recommended; The second vector extraction module extracts the implicit feature vectors of the users to be recommended based on their social relationships in the community network. The fusion module is used to concatenate the explicit and implicit feature vectors of the users to be recommended to generate a fused feature vector of the users to be recommended. The search module is used to identify multiple target reference users based on the similarity between the fused feature vector of the user to be recommended and the fused feature vector of the reference user. The recommendation module is used to determine a list of recommended insurance products for users to be recommended, based on the insurance product purchase list of the target reference user.

[0012] Thirdly, the present invention provides an electronic device, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the recommended method of any of the insurance products described in the foregoing embodiments.

[0013] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the recommended method for any of the insurance products described in the foregoing embodiments.

[0014] This application provides a method, apparatus, electronic device, and storage medium for recommending insurance products. The method includes: extracting an explicit feature vector of the user to be recommended based on the user information of the user to be recommended; extracting an implicit feature vector of the user to be recommended based on the user's social relationships in a community network; concatenating the explicit and implicit feature vectors of the user to be recommended to generate a fused feature vector of the user to be recommended; determining multiple target reference users based on the similarity between the fused feature vector of the user to be recommended and the fused feature vectors of reference users; and determining a list of recommended insurance products for the user to be recommended based on the insurance product purchase lists of the target reference users. By leveraging social relationships and mining commonalities within a group, more accurate recommendations and coverage can be achieved even with sparse data. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the steps of a method for recommending an insurance product, as provided in this application embodiment; Figure 2 A schematic diagram of the structure of an insurance product recommendation device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0018] Example 1 One embodiment of this application provides a method for recommending insurance products based on social relationships. First, the original full user information can be obtained. Users who made purchases within a given time period are designated as high-quality reference users, while other users are considered potential customers for recommendation.

[0019] The user information here may include, but is not limited to, user registration information, historical policy data, App browsing logs (product page browsing time, App login frequency, etc.), claims records, customer service interaction records, and third-party authorized data (such as credit reports, health data, etc.).

[0020] Specifically, the reference users here can be those who have purchased high-value products, made multiple repeat purchases, or completed the full insurance application process in the past 6 months or 1 year, in order to improve the quality of the sample.

[0021] The users to be recommended here are those who have browsed the insurance products purchased by the reference users within a preset time period but have not purchased those insurance products. For example, users who have not made a purchase in the past 6 months or 1 year but have potential intention to purchase insurance (e.g., they have browsed, inquired, compared prices, etc.).

[0022] For any user, explicit and implicit feature vectors can be extracted in advance.

[0023] User information includes at least the user's basic personal information, policy behavior information, browsing behavior information, and medical claims information.

[0024] Basic personal information includes at least static demographic information such as age, gender, occupation, education level, marital status, income range, and region.

[0025] Policy behavior information should include at least the number of historical policies, type of insurance (life insurance / health insurance / auto insurance, etc.), premium amount, renewal rate, and surrender records.

[0026] Browsing behavior information includes at least the duration of page stay, the types of products clicked, search keywords, the number of times items were added to the cart but not paid for, and the frequency of visits.

[0027] Medical claims information should include at least the number of historical claims, claim amount, type of disease, level of hospital treated, and chronic disease records (which must be anonymized in compliance with regulations).

[0028] Explicit feature vectors can include basic features, policy behavior features, browsing behavior features, medical claims features, etc.

[0029] Here, basic user information, policy behavior information, browsing behavior information, and medical claim information can be encoded and feature fused to generate explicit feature vectors of specified dimensions. Before encoding these information, the process includes filling in or removing missing values ​​from the user's basic user information, policy behavior information, browsing behavior information, and medical claim information. The collected user data is then vectorized, including missing value handling, normalization, and binning, to create a display vector representation of the user.

[0030] In one feasible implementation, missing fields in user information can be filled with the mean, mode, or marked as a special category (such as "unknown"). Features with high missing rates can be considered for removal.

[0031] For continuous variables (such as age and premium), use Min-Max normalization or Z-score standardization to eliminate the influence of dimensions. Discretize continuous variables (e.g., age is divided into [18-25, 26-35, …]) to enhance model robustness and capture nonlinear relationships. One-Hot encoding, Label Encoding, or Embedding mapping can be used for discrete features.

[0032] Time-series behaviors such as browsing / clicking are aggregated into statistical metrics (such as the number of times health insurance was clicked in the most recent month, the most frequently browsed product categories, etc.).

[0033] Finally, the user (Including reference users and target customers) is represented as a fixed dimension. explicit eigenvectors .

[0034] In one feasible embodiment, the implicit feature vector can be extracted in the following way: A community network graph is constructed, consisting of second-order paths formed by multiple first-user nodes, community nodes, and second-user nodes. Based on the community network graph, a random walk generates multiple node path sequences of preset lengths. These node path sequences are then input into a Skip-Gram model, which outputs corresponding implicit feature vectors to train and generate an implicit embedding model.

[0035] Specifically, based on the home address or work address filled in by the user during the insurance application process, the address can be mapped to a community to build a social network between the user and the community. The constructed network graph contains two different entities: users (P) and communities (C). Using the metapath2vec method, higher-order social relationships between users are captured and transformed into implicit vector representations of users.

[0036] Implicit feature vectors can also be extracted in the following ways: A community network graph is constructed, consisting of second-order paths formed by multiple first-user nodes, community nodes, and second-user nodes. Based on the community network graph, a random walk is performed to generate multiple node path sequences of preset lengths. Positive and negative samples are generated based on the node path sequences. These positive and negative samples are then input into a Skip-Gram model, causing the Skip-Gram model to output corresponding implicit feature vectors, thereby training and generating an implicit embedding model.

[0037] In one specific embodiment, social connections can be built by leveraging the aggregation of users in physical space (such as the same community or the same unit) to explore "indirect social influence" and supplement the deficiencies of explicit behavioral characteristics.

[0038] Community network diagrams can be constructed in the following ways: Entity types (nodes) include user nodes and community nodes. A user node (P) represents an individual user. A community node (C) represents a geographically or organizationally significant cluster, such as "a residential community in Chaoyang District, Beijing" or "a technology company in Pudong New Area, Shanghai".

[0039] It should be noted that the community nodes listed here are "family + company," which is only the minimum feasible set of community types. The concept of "community" can be expanded to include richer, more multi-dimensional semantic clustering units to more comprehensively depict users' potential social connections and common risks, such as medical hospital communities, interest / consumption communities, etc.

[0040] If user A enters a home address that belongs to the community Then add edges to the graph. — Similarly, if user B's home address is also... ,but — This creates a heterogeneous network graph containing two types of nodes (users and communities). Users sharing the same community are indirectly connected through community nodes, forming a second-order path of "user-community-user".

[0041] In insurance scenarios, users are often unwilling or unable to provide their actual social friend lists. However, address / workplace information is routinely filled out when applying for insurance, making the data readily available. Geographical or workplace proximity often implies similar lifestyles, risk perceptions, and spending power, demonstrating a strong correlation.

[0042] Taking this scenario as an example, the Metapath2Vec learning algorithm operates as follows: Define the meta-path as "user-community-user" (PCP), indicating that two users belong to the same community. Define the node sequence length as L. Second, generate path sequences according to random walks, for example... - - - - .

[0043] Specifically, for each user node P in the graph, it is used as the starting point for the walk. A neighbor of P is randomly selected, and P moves to that neighbor. Then, a neighbor of P is randomly selected, and P moves to that neighbor, and so on, until the length reaches L. For each user node, the above process is repeated r times to obtain multiple context-rich sequences.

[0044] All generated sequences are treated as a "virtual corpus," where each node is considered a "word." The Skip-Gram model is trained on this corpus to learn low-dimensional vector representations of each node. Specifically, the model traverses each sequence through a sliding window, designating the node at the center of the window as the "center node" and its neighboring nodes as "context nodes."

[0045] The training objective is to adjust the vector parameters of each node so that the vector of the center node maximizes the probability of predicting the occurrence of its context nodes. To improve training efficiency, a negative sampling strategy is adopted: for each (center node, context node) positive sample pair, several randomly sampled negative sample nodes are introduced simultaneously, and the overall likelihood function is optimized. After multiple rounds of iterative training, each user node P obtains a real-valued vector of fixed dimensions (e.g., 128 dimensions), called the user's implicit embedding vector. This vector implicitly encodes the user's structural role in the social network—even if two users have no direct interaction, as long as they share similar community environments (such as living in the same neighborhood or working for the same company), their implicit vectors will tend to be similar in the vector space.

[0046] The core idea of ​​the Skip-Gram model is to maximize the joint probability between a node and its context nodes. Specifically, the objective function can be expressed as:

[0047] Where V is the set of all nodes in the graph. viIt is the target node. N ( vi ) is a node vi The set of context nodes, which are generated by random walks. P ( vj | vi ) is at a given node vi Under the condition, node vj The probability of occurrence.

[0048] Figure 1 This is a flowchart illustrating the steps of a method for recommending an insurance product, as provided in an embodiment of this application. Figure 1 As shown in the embodiment of this application, a method for recommending an insurance product includes: S1. Extract the explicit feature vector of the user to be recommended based on the user information of the user to be recommended.

[0049] S2. Based on the social relationships of the users to be recommended in the community network, extract the implicit feature vectors of the users to be recommended.

[0050] In steps S1 and S2, feature extraction can be performed in accordance with the manner described in the foregoing embodiments.

[0051] S3. Concatenate the explicit and implicit feature vectors of the users to be recommended to generate a fused feature vector for the users to be recommended.

[0052] Here, you can either directly splice or use weighted splicing.

[0053] S4. Based on the similarity between the fused feature vector of the user to be recommended and the fused feature vector of the reference user, multiple target reference users are identified.

[0054] In step S4, the top K similar reference users are found for the user to be recommended. The reference users can be ranked based on the similarity, and the top K reference users are selected as target reference users.

[0055] Specifically, similarity can be calculated using cosine similarity, Mahalanobis distance, etc.

[0056] S5. Based on the insurance purchase list of the target reference users, determine the insurance recommendation list for the users to be recommended.

[0057] Based on a list of insurance products purchased by a reference user within a given time period, recommend appropriate insurance products to the target user.

[0058] This section can be used to identify the N most frequently purchased insurance products by target users and recommend them to potential users.

[0059] For new users, most basic characteristics, policy behavior characteristics, browsing behavior, and medical claim characteristics are empty. This application provides a method for recommending insurance products that, compared to existing technologies, effectively mines commonalities derived from real-world social relationships such as family and workplace by adding social relationship networks, thus alleviating data sparsity issues and improving the accuracy and coverage of recommendations.

[0060] Example 2 Figure 2 This is a schematic diagram of the structure of an insurance product recommendation device provided in an embodiment of this application. Figure 2 As shown, based on the same inventive concept, this application also provides an insurance product recommendation device 20, which includes: The first vector extraction module 210 is used to extract the explicit feature vector of the user to be recommended based on the user information of the user to be recommended. The second vector extraction module 220 extracts the implicit feature vector of the user to be recommended based on the social relationship of the user in the community network; The fusion module 230 is used to concatenate the explicit feature vector and the implicit feature vector of the user to be recommended in order to generate the fused feature vector of the user to be recommended. Search module 240 is used to determine multiple target reference users based on the similarity between the fused feature vector of the user to be recommended and the fused feature vector of the reference user; The recommendation module 250 is used to determine the insurance recommendation list for the target reference user based on the insurance purchase list of the target reference user.

[0061] In an optional implementation, the second vector extraction module 220 extracts the implicit feature vector in the following manner: Construct a community network graph, which includes a second-order path formed by multiple first-user nodes, community nodes, and second-user nodes; Based on the community network graph, random walks generate multiple node path sequences of preset lengths; The node path sequence is input into the Skip-Gram model, which outputs the corresponding implicit feature vector to train and generate an implicit embedding model.

[0062] In an optional implementation, the second vector extraction module 220 extracts the implicit feature vector in the following manner: Construct a community network graph, which includes a second-order path formed by multiple first-user nodes, community nodes, and second-user nodes; Based on the community network graph, random walks generate multiple node path sequences of preset lengths; Generate positive and negative samples based on node path sequences; Positive and negative samples are input into the Skip-Gram model, which outputs the corresponding implicit feature vectors to train and generate an implicit embedding model.

[0063] In an optional implementation, the user information includes at least the user's basic personal information, policy behavior information, browsing behavior information, and medical claims information. The first vector extraction module 210 extracts explicit feature vectors in the following manner: The system encodes and fuses user basic personal information, policy behavior information, browsing behavior information, and medical claims information to generate explicit feature vectors of specified dimensions.

[0064] In an optional implementation, before encoding the user's basic personal information, policy behavior information, browsing behavior information, and medical claim information, the first vector extraction module 210 is also used to fill in or remove missing values ​​from the user's basic personal information, policy behavior information, browsing behavior information, and medical claim information. It also standardizes users' basic personal information, policy behavior information, browsing behavior information, and medical claims information.

[0065] In an optional implementation, the search module 240 is used to sort the reference users based on similarity and select the top K reference users as target reference users.

[0066] In an optional implementation, the recommended user is a user who has browsed the insurance products purchased by the reference user within a preset time period but has not purchased the insurance product.

[0067] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3 As shown, the electronic device 300 includes a processor 310, a memory 320, and a bus 330.

[0068] The memory 320 stores machine-readable instructions executable by the processor 310. When the electronic device 300 is running, the processor 310 and the memory 320 communicate via the bus 330. When the machine-readable instructions are executed by the processor 310, the steps of a method for recommending an insurance product as described in the above method embodiment can be executed. For specific implementation details, please refer to the method embodiment, which will not be repeated here.

[0069] This application also provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it can execute the steps of a method for recommending an insurance product as described in the above method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0070] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0071] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0072] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0073] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0074] It should be noted that if the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0075] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0076] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for recommending insurance products, characterized in that, The method includes: Based on the user information of the users to be recommended, extract the explicit feature vector of the users to be recommended; Based on the social relationships of the users to be recommended in the community network, the implicit feature vectors of the users to be recommended are extracted; The explicit and implicit feature vectors of the users to be recommended are concatenated to generate a fused feature vector of the users to be recommended. Based on the similarity between the fused feature vector of the user to be recommended and the fused feature vector of the reference user, multiple target reference users are identified. Based on the insurance purchase list of the target reference users, determine the insurance recommendation list for the users to be recommended.

2. The method according to claim 1, characterized in that, The implicit feature vector is extracted using the following method: Construct a community network graph, which includes a second-order path formed by multiple first-user nodes, community nodes, and second-user nodes; Based on the community network graph, random walks generate multiple node path sequences of preset lengths; The node path sequence is input into the Skip-Gram model, which outputs the corresponding implicit feature vector to train and generate an implicit embedding model.

3. The method according to claim 1, characterized in that, The implicit feature vector is extracted using the following method: Construct a community network graph, which includes a second-order path formed by multiple first-user nodes, community nodes, and second-user nodes; Based on the community network graph, random walks generate multiple node path sequences of preset lengths; Generate positive and negative samples based on node path sequences; Positive and negative samples are input into the Skip-Gram model, which outputs the corresponding implicit feature vectors to train and generate an implicit embedding model.

4. The method according to claim 1, characterized in that, User information includes at least the user's basic personal information, policy behavior information, browsing behavior information, and medical claims information. Explicit feature vectors are extracted using the following methods: The system encodes and fuses user basic personal information, policy behavior information, browsing behavior information, and medical claims information to generate explicit feature vectors of specified dimensions.

5. The method according to claim 4, characterized in that, Before encoding users' basic personal information, policy behavior information, browsing behavior information, and medical claims information, the following steps are also included: Fill in or remove missing values ​​for users' basic personal information, policy behavior information, browsing behavior information, and medical claims information; It also standardizes users' basic personal information, policy behavior information, browsing behavior information, and medical claims information.

6. The method according to claim 1, characterized in that, The reference users are ranked based on the similarity, and the top K reference users are selected as target reference users.

7. The method according to claim 1, characterized in that, The users to be recommended are those who have browsed the insurance products purchased by the reference users within a preset time period but have not purchased the insurance products.

8. A device for recommending insurance products, characterized in that, The device includes: The first vector extraction module is used to extract the explicit feature vector of the user to be recommended based on the user information of the user to be recommended; The second vector extraction module extracts the implicit feature vectors of the users to be recommended based on their social relationships in the community network. The fusion module is used to concatenate the explicit and implicit feature vectors of the users to be recommended to generate a fused feature vector of the users to be recommended. The search module is used to identify multiple target reference users based on the similarity between the fused feature vector of the user to be recommended and the fused feature vector of the reference user. The recommendation module is used to determine a list of recommended insurance products for users to be recommended, based on the insurance product purchase list of the target reference user.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the recommended method for the insurance product as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method for recommending an insurance product as described in any one of claims 1 to 7.

Images