An interactive intelligent insurance recommendation method and system based on a knowledge graph
By constructing an insurance knowledge graph and using entity recognition technology, the problems of inaccurate and unexplainable acquisition of user needs in existing insurance recommendation methods have been solved, enabling more accurate and explainable insurance product recommendations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGSU DAOTAI INFORMATION TECH CO LTD
- Filing Date
- 2023-11-22
- Publication Date
- 2026-07-24
AI Technical Summary
Existing insurance recommendation methods cannot explicitly obtain users' specific needs, rely on the user data owned by the enterprise platform, resulting in inaccurate prediction results and a lack of interpretability, making it difficult to modify them for specific factors.
Construct a knowledge graph of information fields, insurance types, insurance products, and insurers. Extract demand information from users' natural language descriptions through entity recognition and field standardization. Calculate the weights of insurance types and products using the knowledge graph to generate recommendation reasons.
It achieves more accurate insurance product recommendations, is interpretable, suitable for cold start situations, can be improved in a targeted manner, and generates persuasive recommendation reasons.
Smart Images
Figure CN117743589B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of artificial intelligence technology, and more specifically, relates to an interactive intelligent insurance recommendation method and system based on knowledge graph fusion. Background Technology
[0002] As the economy develops and people's living standards rise, their awareness of insurance also increases. Therefore, how to provide personalized insurance product recommendations tailored to individual needs has become an important issue. However, current common insurance recommendation methods often rely on the user's personal information on the platform and employ machine learning to recommend insurance products. This method has the following drawbacks: (1) Passively inferring users’ insurance intentions through the user data owned by the enterprise platform, it is impossible to explicitly and proactively obtain users’ specific needs, and there may be biases in the understanding of users’ needs; (2) The accuracy of insurance product forecasts heavily depends on the user data owned by the enterprise platform. This is very unfriendly to enterprises with little or no user data, and may lead to inaccurate forecast results. (3) Machine learning methods are often black box models, which are not interpretable and make it difficult to verify the accuracy of recommendation methods in principle. When the recommendation is wrong, it is also difficult to make targeted improvements. (4) Machine learning methods are data-driven, uncontrollable, and difficult to make targeted modifications to certain specific elements. Summary of the Invention
[0003] To address the aforementioned problems in existing technologies, this application aims to provide an interactive intelligent insurance recommendation method and system based on knowledge graphs.
[0004] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows: Firstly, this application provides an interactive intelligent insurance recommendation method based on knowledge graphs, the method comprising: Construct a knowledge graph of information fields - insurance type - insurance product - insurer; From the natural language description text input by the user, the requirement information is identified through entity recognition and field standardization processes. The requirement information includes a set of required information fields, a set of optional information fields, and a set of possible insurers and insurance types. Calculate the insurance type set and the weight of each insurance type based on the aforementioned demand information and knowledge graph; The insurance product and its corresponding score are calculated using the knowledge graph, the insurance type set, and the weight of each insurance type. Based on the calculated insurance products and their corresponding rankings, recommended insurance products are determined, and reasons for the recommendations are generated based on the constructed knowledge graph.
[0005] In a possible embodiment, the construction of the knowledge graph of information fields - insurance type - insurance product - insurer specifically includes: Define a schema layer for the knowledge graph of information fields, insurance types, insurance products, and insurers. The schema layer includes four concepts: information fields, insurance types, insurance products, and insurers; and four relationships: information fields-insurance types, insurance types-insurance products, information fields-insurance products, and insurance products-insurers. The information fields-insurance types and information fields-insurance products relationships have weighted attributes. The information fields include attributes such as field name, field alias, field type, whether the field is required, field value type, and field value. The insurance products include attributes such as product name, product alias, product description, and product recommendation level. The insurers include attributes such as insurer name, insurer alias, insurer description, and insurer address. The field types of the information fields include basic information and feature information. Based on the defined schema layer, fill in the information fields, insurance type, insurance product, and insurer as data nodes of the knowledge graph, and construct the relationships between the nodes as the final knowledge graph; For the weighted attributes of relations in a knowledge graph, if the nodes on both sides of the relation do not involve feature information, the weight is set to 1; if one of the nodes on both sides belongs to feature information, the weights of the two relations, information field-insurance product and information field-insurance type, are calculated using the following algorithm: There are m experts, and the feature information of the information field is t. i The insurance product or type of insurance is c. j The relationship between information fields and insurance products or information fields and insurance types. i,j The weight w i,j The calculation formula is:
[0006] Among them, rec i,j It is an expert's view on relationship r i,j The level of recognition, imp i For feature information t i Importance, n i,j It is believed that c j Having feature information t i The number of experts, N i It contains feature information t i The number of relationships between information field - insurance product or information field - insurance type, where N is the total number of relationships between all information fields - insurance products or information fields - insurance types. It is a smoothing factor with a value greater than or equal to 0. It is a selection factor.
[0007] Specifically, the process of identifying requirement information from the user-input natural language description text through entity recognition and field standardization includes: Construct an insurance type dictionary, an insurer dictionary, and a field information component dictionary; The requirement information in the natural language description text is identified using an entity recognition method based on bidirectional maximum character matching. The non-standard insurance type names, insurer names, and information field component names identified are mapped to standard names using an insurance type dictionary, an insurer dictionary, and a field information component dictionary. Align the standardized names to the corresponding nodes in the constructed knowledge graph.
[0008] Specifically, the method for constructing the field information component dictionary is as follows: Define the components of the information field, including: object terms, object modifiers, limiting terms, quantitative limiting values, and qualitative limiting values; Based on the components of the defined information fields, a dictionary of information field components is constructed using the collected customer demand description data.
[0009] Specifically, the step of calculating the insurance type set and the weight of each insurance type based on the demand information and the knowledge graph includes: Establish a set of possible insurance types T and a set of information field-insurance type relationships RT, both initialized as empty sets. Iterate through each information field in the information field set. i Traverse each field in the knowledge graph i The insurance type is part of the information field related to the insurance type. j Let the relation be rt. i,j The corresponding relation weight is wrt. i,j , type j Add T, and rt i,j If we include RT, then each insurance type in the final T will be... k weight wt k for:
[0010] Set an insurance type threshold th, and iterate through T for each insurance type type in T. k If its corresponding weight wt k If it is less than the threshold th, then type k Delete from T; If T does not contain the insurance type t predicted by the user. p Then t p Add T; regardless of whether T originally contained t. p , all will t p weight wt p Set the weight to the maximum value of the weights of all insurance types in T, i.e., the new weight wt. p The calculation method for ' is as follows:
[0011] Specifically, the step of calculating insurance products and corresponding scores using the knowledge graph, insurance type set, and weights for each insurance type includes: Establish a set of possible insurance products P and a set of information fields-insurance products relation RP, both initialized as empty sets. Iterate through each information field in the information field set. i Traverse each field in the knowledge graph i Composition Information Fields - Insurance Product Relationships p j Let the relationship be rp i,j The corresponding relation weight is WRP. i,j Additionally, p can be obtained from the knowledge graph. j Corresponding insurance type j Let the weight calculated for this type of insurance be wt. k If type j If it belongs to T, then p j Add P, and rp i,j Add RP. Each insurance product p in the final P k weight wp k for:
[0012] rec k For insurance products p in the knowledge graph k Product recommendation rate; Iterate through each insurance product p in the information field set. k For each p k Iterate through each information field in the information field set. i If field i The "Is field required?" attribute is true and p k Does not conform to field i The constraints, or p k If the insurer's prediction in the knowledge graph does not match the customer's prediction, then p will be... k Remove from P and use the final P as the recommended set of insurance products; If P is ultimately empty, and an insurance product was previously removed from P because the corresponding insurer did not match the customer's prediction, then add that insurance product back into P.
[0013] Specifically, the process involves determining recommended insurance products based on the calculated insurance product ranking and corresponding scores, and generating recommendation reasons based on the constructed knowledge graph, including: For each insurance product p in the calculated set of insurance products P k According to its corresponding insurance type k Group the insurance products, keeping only the one with the highest weight (wp) in each group. Then, select the top k groups of insurance products with the highest weights, based on each information field in the information field set. i Based on the corresponding node information in the knowledge graph, as well as other general information about the product in the knowledge graph, recommendation text is generated according to predefined rule templates.
[0014] Secondly, an interactive intelligent insurance recommendation system based on knowledge graphs includes: The knowledge graph construction module constructs the corresponding nodes and relationships of the knowledge graph of information fields, insurance types, insurance products, and insurers based on the survey information from experts, and calculates the weights on the relationships; The knowledge graph storage and query module is used to store the constructed information fields - insurance type - insurance product - insurer knowledge graph and provide query services; The dialogue module is used to ask users for their needs and collect natural language description text input by the users. The entity recognition and name standardization module identifies the requirement information based on the natural language description text collected by the dialogue module and performs standardization operations on the requirement information. The requirement information includes a set of required information fields, a set of optional information fields, and a set of possible insurers and insurance types. The insurance product recommendation module calculates the insurance type set and the weight of each insurance type, the optional insurer set, and the constructed knowledge graph based on the demand information and knowledge graph to recommend insurance products. The management module is responsible for calling the knowledge graph storage and query module, dialogue module, entity recognition and name standardization module, and insurance product recommendation module to make insurance recommendations.
[0015] Specifically, the entity recognition and name standardization module includes: The dictionary storage and query module is used to store the constructed insurance type dictionary, insurer dictionary, and field information component dictionary, and provides query services. The entity recognition module uses an entity recognition method based on bidirectional maximum character matching to identify the requirement information in the natural language description text; The standardization module uses an insurance type dictionary, an insurer dictionary, and a field information component dictionary to convert the identified non-standard insurance type names, insurer names, and information field component names into standard names. The mapping module maps the standardized names to the corresponding nodes in the information field - insurance type - insurance product - insurer knowledge graph.
[0016] Specifically, the insurance product recommendation module includes: The insurance type calculation module uses the user's information field set, the predicted insurance type set, and the constructed knowledge graph to calculate the possible insurance type set and the weight of each insurance type; The insurance product calculation module uses the user's information field set, the possible insurance type set and its weight, the optional insurer set, and the constructed knowledge graph to calculate the recommended insurance products and their corresponding scores. The recommendation reason generation module calculates the insurance products and their corresponding rankings, determines the recommended insurance products, and generates recommendation reasons based on the constructed knowledge graph.
[0017] Compared with the prior art, this application has the following advantages: (1) Receiving insurance needs described by users in natural language is more natural and in line with human communication habits; (2) It can obtain more detailed information about the user's needs through interactive methods, making the recommendations more accurate; (3) It does not rely on the enterprise platform's own user data, making it more user-friendly and suitable for cold start situations; (4) Knowledge graphs are constructed based on expert knowledge for the recommendation of insurance products, making the recommendation method interpretable. The accuracy and reliability of the recommendation method can be directly tested, and it can be modified in a targeted manner for specific elements. (5) Based on the corresponding node information in the knowledge graph of the information fields in the user's needs description, the recommendation reasons can be generated in a targeted manner to respond to the user's concerns and make the recommendation reasons more persuasive. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating an embodiment of the interactive smart insurance recommendation method of this application; Figure 2 This is a flowchart of the entity identification and name standardization process for this application; Figure 3 A flowchart illustrating the process of constructing a dictionary of field information components for this application; Figure 4 This is a schematic diagram of an embodiment of the interactive intelligent insurance recommendation system of this application; Figure 5 This is a schematic diagram illustrating the structure of an embodiment of the entity identification and name standardization module of this application; Figure 6 This is a schematic diagram of an embodiment of the insurance product recommendation module in this application; Figure 7 This application provides a schematic diagram of the structure of an embodiment of an interactive intelligent insurance recommendation system. Figure 8 This is a schematic diagram of the structure of an embodiment of a computer storage medium provided in this application. Detailed Implementation
[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to embodiments and accompanying drawings. The content mentioned in the embodiments is not intended to limit the present invention.
[0020] like Figure 1 As shown, this embodiment provides an interactive intelligent insurance recommendation method based on knowledge graphs, including the following steps: S100: Construct an information field - insurance type - insurance product - insurer knowledge graph.
[0021] Building a knowledge graph first requires collecting survey information from experts to construct the corresponding nodes, relationships, and weights within those relationships. The specific methods for constructing a knowledge graph are as follows: First, we define the schema layer of the knowledge graph for information fields, insurance types, insurance products, and insurers. The schema layer includes four concepts: information fields, insurance types, insurance products, and insurers. There are four relationships: information fields-insurance types, insurance types-insurance products, information fields-insurance products, and insurance products-insurers. The relationships between information fields and insurance types and between information fields and insurance products have weighted attributes. Information fields include attributes such as field name, field alias, field type, whether the field is required, field value type, and field value. Insurance products include attributes such as product name, product alias, product description, and product recommendation level. Insurers include attributes such as insurer name, insurer alias, insurer description, and insurer address. The field types of information fields include two main categories: basic information (such as coverage amount, coverage period, payment method, etc.) and feature information (such as suitability for the elderly, lenient underwriting, simple claims process, etc.). Basic information is a description of objective facts, while feature information tends to be subjective evaluation.
[0022] The construction of the knowledge graph for information fields, insurance types, insurance products, and insurers adopted a top-down strategy. Questionnaires were distributed to multiple experts, who filled in instance data such as information fields, insurance types, insurance products, and insurers as data nodes of the knowledge graph based on the defined knowledge graph pattern layers. They also constructed relationships between nodes. The union of the experts' results was taken, and after resolving conflicts, it became the final knowledge graph data. The conflict resolution process adopted a negotiation and voting strategy. If a consensus could not be reached through negotiation, the content to be resolved was voted on, and the one with the most votes became the final result.
[0023] For the weight attribute of relations in a knowledge graph, if the nodes on both sides of the relation do not involve feature information, the weight is set to 1; if one of the nodes on both sides belongs to feature information, the weight of the two relations, information field-insurance product and information field-insurance type, is calculated using the following algorithm. The following example is to calculate the weight of the information field-insurance product relation (the same operation is performed on the information field-insurance type relation).
[0024] There are m experts, and the feature information of the information field is t. i The insurance product is C. j The relationship between information fields and insurance products. i,j The weight w on i,j The calculation formula is:
[0025] Among them, rec i,j It is an expert's view on relationship r i,j The level of recognition, imp i For feature information t i Importance, n i,j It is believed that c j Having feature information t i The number of experts, N i It contains feature information t i The number of relationships between information fields and insurance products, where N is the total number of relationships between all information fields and insurance products. It is a smoothing factor with a value greater than or equal to 0. It is a selection factor.
[0026] A value of 0 indicates that all feature information has the same importance. A value greater than 0 indicates that the more common the feature information, the more important it is. A value less than 0 indicates that the more distinctive the feature, the more important it is. In this embodiment, we take... = =0.
[0027] S200: From the natural language description text input by the user, through entity recognition and field standardization processes, identify the requirement information, which includes a set of required information fields, a set of optional information fields, and a set of possible insurers and insurance types.
[0028] like Figure 2 As shown, the specific process of this step is as follows: S210: Construct an insurance type dictionary, an insurer dictionary, and a field information component dictionary; S220: Identify the demand information in the natural language description text, such as insurance type, insurer, and field information, using an entity recognition method based on bidirectional maximum character matching.
[0029] S230: Using the insurance type dictionary, insurer dictionary, and field information component dictionary, map the identified non-standard insurance type names, insurer names, and information field component names to standard names; S240: Align the standardized standard names to the corresponding nodes in the constructed knowledge graph.
[0030] In step S210, the process of constructing the insurance type dictionary specifically involves: defining the concept of insurance types; utilizing collected customer demand description data and insurance product data; constructing insurance type terms and their standard expressions; and thus forming the insurance type dictionary. An example dictionary is shown in Table 1. Table 1. Examples of Insurance Type Dictionary Step 210, the process of constructing the insurer dictionary, specifically involves: defining the concept of an insurer; utilizing collected customer demand description data and insurance product data; constructing insurer terms and their standard expressions; and thus forming the insurer dictionary. An example dictionary is shown in Table 2. Table 2 Examples of Insurer's Dictionary In step S210, the process of constructing the field information component dictionary is as follows: Figure 3 As shown, specifically: S211: Define the components of the information field, including: object words, object modifiers, limiting words, quantitative limiting values, and qualitative limiting values.
[0031] The definition and scope of the information field are as follows: Object term: The object being described is usually a related attribute of an insurance product, such as "premium", "coverage", "waiting period", etc.
[0032] Modifiers for objects: Modifiers for objects usually further refine the objects. For example, "monthly" and "annual" are modifiers for the object "premium".
[0033] Restricting words: Conditional restrictions on object words, usually numerical or set operations, such as "maximum", "not exceeding", "above", etc.
[0034] Quantitative limit: The specific quantitative limit corresponding to the limiting words, such as "60 yuan" or "30 days".
[0035] Qualitative limiting value: The specific qualitative limiting value corresponding to the limiting word, such as "cheap" or "mid-range".
[0036] S212: Based on the components of the set information fields, construct an information field component dictionary using the collected customer demand description data.
[0037] In this embodiment, examples of dictionaries are shown in Table 3: Table 3. Examples of Field Information Components Dictionary Based on the constructed insurance type dictionary, insurer dictionary, and field information component dictionary, the insurance type, insurer, and field information in user description text can be extracted, identified, and standardized, thereby mapping them to corresponding nodes in the knowledge graph. For example, for the user description text "20 years old, hoping to recommend a moderate medical insurance plan, with a maximum monthly premium of no more than 30 yuan," the mapped knowledge graph nodes are shown in Table 4: Table 4 Examples of dictionary-based identification extraction, name standardization, and knowledge graph node mapping This means that the insurance product must have a "mid-range" attribute value for the "tier" attribute in the knowledge graph, the insurance type must be "medical insurance," and the attribute value for the "monthly premium for a 20-year-old" must be less than or equal to 30 yuan.
[0038] In some embodiments, for required information fields not mentioned by the user, data can be collected by prompting the user through human-computer interaction. The specific process can be as follows: The system iterates through the information fields in the constructed knowledge graph that are required fields. If the corresponding field is not collected, the system asks the user for information about the field through human-computer interaction and performs entity recognition and field standardization on the user's response text. If there are still required information fields that have not been mentioned, the system continues to ask until all required information fields have been collected.
[0039] S300: Calculate the insurance type set and the weight of each insurance type based on the required information and the knowledge graph.
[0040] The specific process for this step is as follows: Establish a set of possible insurance types T and a set of information field-insurance type relationships RT, both initialized as empty sets. Iterate through each information field in the information field set. i Traverse each field in the knowledge graph i The insurance type is part of the information field related to the insurance type. j Let the relation be rt. i,j The corresponding relation weight is wrt. i,j , type j Add T, and rt i,j If we include RT, then each insurance type in the final T will be... k weight wt k for:
[0041] Set an insurance type threshold th, and iterate through T for each insurance type type in T. k If its corresponding weight wt k If it is less than the threshold th, then type k Delete from T.
[0042] If T does not contain the insurance type t predicted by the user. p Then t p Add T; regardless of whether T originally contained t. p , all will t p weight wt p Set the weight to the maximum value of the weights of all insurance types in T, i.e., the new weight wt. p The calculation method for ' is as follows:
[0043] S400: Calculate the insurance product and its corresponding score using the knowledge graph, the insurance type set, and the weight of each insurance type.
[0044] Specifically, by utilizing the user's information field set, the predicted set of insurers, the calculated set of possible insurance types and their weights, and the constructed knowledge graph, the insurance product and its corresponding score are calculated. The specific process is as follows: Establish a set of possible insurance products P and a set of information fields-insurance products relation RP, both initialized as empty sets. Iterate through each information field in the information field set. i Traverse each field in the knowledge graph i Composition Information Fields - Insurance Product Relationships p j Let the relationship be rp i,j The corresponding relation weight is WRP. i,j Additionally, p can be obtained from the knowledge graph.j Corresponding insurance type j Let the weight calculated for this type of insurance be wt. k If type j If it belongs to T, then p j Add P, and rp i,j Add RP. Each insurance product p in the final P k weight wp k for:
[0045] rec k For insurance products p in the knowledge graph k Product recommendation rate.
[0046] Iterate through each insurance product p in the information field set. k For each p k Iterate through each information field in the information field set. i If field i The "Is field required?" attribute is true and p k Does not conform to field i The constraints, or p k If the insurer in the knowledge graph does not match the customer's prediction (if the customer described the insurer), then p will be... k Remove from P and use the final P as the recommended set of insurance products.
[0047] If P is ultimately empty, and an insurance product was previously removed from P because the corresponding insurer did not match the customer's prediction, then add that insurance product back into P.
[0048] S500: Based on the calculated insurance products and their corresponding rankings, determine the recommended insurance products and generate recommendation reasons based on the constructed knowledge graph.
[0049] In a specific embodiment, this step specifically includes: for each insurance product p in the calculated insurance product set P k According to its corresponding insurance type k Group the insurance products, keeping only the one with the highest weight (wp) in each group. Then, select the top k groups of insurance products with the highest weights, based on each information field in the information field set. i Based on the corresponding node information in the knowledge graph, as well as other general information about the product in the knowledge graph, recommendation text is generated according to predefined rule templates.
[0050] Reference Figure 4This embodiment provides an interactive intelligent insurance recommendation system based on knowledge graphs. The system includes a knowledge graph construction module, a knowledge graph storage and query module, a dialogue module, an entity recognition and name standardization module, an insurance product recommendation module, and a management module.
[0051] The knowledge graph construction module is used to collect survey information from experts and construct the corresponding nodes and relationships of the knowledge graph of information fields, insurance types, insurance products and insurers, and calculate the weights on the relationships.
[0052] The knowledge graph storage and query module is used to store the constructed information fields - insurance type - insurance product - insurer knowledge graph and provide query services.
[0053] The dialogue module is used to ask users for their needs and collect natural language descriptions of user feedback.
[0054] The entity recognition and name standardization module identifies the requirement information based on the natural language description text collected by the dialogue module, and performs standardization operations on the requirement information. The requirement information includes a set of required information fields, a set of optional information fields, and a set of possible insurers and insurance types.
[0055] The insurance product recommendation module calculates the insurance type set and the weight of each insurance type, the set of possible insurers, and the constructed knowledge graph based on the demand information and the knowledge graph to recommend insurance products.
[0056] The management module, when actually recommending insurance products, is responsible for calling the knowledge graph storage and query module, the dialogue module, the entity recognition and name standardization module, and the insurance product recommendation module to make insurance recommendations.
[0057] In some embodiments, such as Figure 5 As shown, the entity recognition and name standardization module specifically includes a dictionary storage and query module, an entity recognition module, a standardization module, and a mapping module.
[0058] The dictionary storage and query module stores the constructed insurance type dictionary, insurer dictionary, and field information component dictionary, and provides query services.
[0059] The entity recognition module uses a bidirectional maximum character matching-based entity recognition method to identify the required information in the natural language description text, such as insurance type, insurer, and field information components.
[0060] The standardization module uses an insurance type dictionary, an insurer dictionary, and a field information component dictionary to convert the identified non-standard insurance type names, insurer names, and information field component names into standard names.
[0061] The mapping module maps the standardized names to the corresponding nodes in the information field - insurance type - insurance product - insurer knowledge graph.
[0062] In some embodiments, such as Figure 6 As shown, the insurance product recommendation module specifically includes an insurance type calculation module, an insurance product calculation module, and a recommendation reason generation module.
[0063] The insurance type calculation module uses the user's information field set, the predicted insurance type set, and the constructed knowledge graph to calculate the possible insurance type set and the weight of each insurance type.
[0064] The insurance product calculation module uses the user's information field set, the possible insurance type set and its weights, the optional insurer set, and the constructed knowledge graph to calculate the recommended insurance products and their corresponding scores.
[0065] The recommendation reason generation module determines the recommended insurance products based on their ranking and corresponding scores, and generates recommendation reasons based on the constructed knowledge graph.
[0066] like Figure 7 As shown, this embodiment also provides an interactive intelligent insurance recommendation system based on a knowledge graph, including at least one processor 71, a communication interface 72, a memory 73, and a communication bus 74. The processor 71, communication interface 72, and memory 73 communicate with each other via the communication bus 74. The processor 71 can call logical instructions in the memory 73 to execute the interactive intelligent insurance recommendation method based on a knowledge graph in any of the above embodiments.
[0067] The logical instructions in the aforementioned memory 73 can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a 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 the present invention. 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.
[0068] In addition, this application also provides a computer storage medium, see [reference]. Figure 8 The schematic diagram of the embodiment shown illustrates that the storage medium 8 includes a computer program 81 stored thereon. This computer program 81 can be executed to implement the interactive smart insurance recommendation method provided by any one or any non-conflicting combination of the above embodiments. The capacity of the storage medium 8 is sufficient to store the computer program.
[0069] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more storage media 10 (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0070] The foregoing has provided a detailed description of an interactive intelligent insurance recommendation method and system based on knowledge graphs provided in this application. The specific embodiments are described only to aid in understanding the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.
Claims
1. An interactive intelligent insurance recommendation method based on knowledge graphs, characterized in that, The method includes: Construct a knowledge graph of information fields - insurance type - insurance product - insurer; From the natural language description text input by the user, through entity recognition and field standardization processes, the requirement information is identified. The requirement information includes a set of required information fields, a set of optional information fields, a set of insurers, and a set of insurance types. Calculate the insurance type set and the weight of each insurance type based on the aforementioned demand information and knowledge graph; The insurance product and its corresponding score are calculated using the knowledge graph, the insurance type set, and the weight of each insurance type. Based on the calculated insurance products and their corresponding rankings, recommended insurance products are determined, and reasons for the recommendations are generated based on the constructed knowledge graph. The construction of the knowledge graph of information fields - insurance type - insurance product - insurer specifically includes: Define a schema layer for the knowledge graph of information fields, insurance types, insurance products, and insurers. The schema layer includes four concepts: information fields, insurance types, insurance products, and insurers; and four relationships: information fields-insurance types, insurance types-insurance products, information fields-insurance products, and insurance products-insurers. The information fields-insurance types and information fields-insurance products relationships have weighted attributes. The information fields include attributes such as field name, field alias, field type, whether the field is required, field value type, and field value. The insurance products include attributes such as product name, product alias, product description, and product recommendation level. The insurers include attributes such as insurer name, insurer alias, insurer description, and insurer address. The field types of the information fields include basic information and feature information. Based on the defined schema layer, fill in the information fields, insurance type, insurance product, and insurer as data nodes of the knowledge graph, and construct the relationships between the nodes as the final knowledge graph; For the weighted attributes of relations in a knowledge graph, if the nodes on both sides of the relation do not involve feature information, the weight is set to 1; if one of the nodes on both sides belongs to feature information, the weights of the two relations, information field-insurance product and information field-insurance type, are calculated using the following algorithm: There are m experts, and the feature information of the information field is t. i The insurance product or type of insurance is c. j The relationship between information fields and insurance products or information fields and insurance types. i,j The weight w on i,j The calculation formula is: Among them, rec i,j It is an expert's view on relationship r i,j The level of recognition, imp i For feature information t i Importance, n i,j It is believed that c j Having feature information t i The number of experts, N i It contains feature information t i The number of relationships between information field - insurance product or information field - insurance type, where N is the total number of relationships between all information fields - insurance products or information fields - insurance types. It is a smoothing factor with a value greater than or equal to 0. It is a selection factor.
2. The interactive intelligent insurance recommendation method according to claim 1, characterized in that, The process of identifying requirement information from user-input natural language description text through entity recognition and field standardization includes: Construct an insurance type dictionary, an insurer dictionary, and a field information component dictionary; The requirement information in the natural language description text is identified using an entity recognition method based on bidirectional maximum character matching. The non-standard insurance type names, insurer names, and information field component names identified are mapped to standard names using an insurance type dictionary, an insurer dictionary, and a field information component dictionary. Align the standardized names to the corresponding nodes in the constructed knowledge graph.
3. The interactive intelligent insurance recommendation method according to claim 2, characterized in that, The method for constructing the field information component dictionary is as follows: Define the components of the information field, including: object terms, object modifiers, limiting terms, quantitative limiting values, and qualitative limiting values; Based on the components of the defined information fields, a dictionary of information field components is constructed using the collected customer demand description data.
4. The interactive intelligent insurance recommendation method according to claim 1, characterized in that, The calculation of the insurance type set and the weight of each insurance type based on the demand information and knowledge graph specifically includes: Establish an insurance type set T and an information field-insurance type relation set RT, both initialized as empty sets. Iterate through each information field in the information field set. i Traverse each field in the knowledge graph i The insurance type is part of the information field related to the insurance type. j Let the relation be rt. i,j The corresponding relation weight is wrt. i,j , type j Add T, and rt i,j If we include RT, then each insurance type in the final T will be... k weight wt k for: Set an insurance type threshold th, and iterate through T for each insurance type type in T. k If its corresponding weight wt k If it is less than the threshold th, then type k Delete from T; If T does not contain the insurance type t predicted by the user. p Then t p Add T; regardless of whether T originally contained t. p , all will t p weight wt p Set the weight to the maximum value of the weights of all insurance types in T, i.e., the new weight wt. p The calculation method for ' is as follows: 。 5. The interactive intelligent insurance recommendation method according to claim 1, characterized in that, The calculation of insurance products and corresponding scores using the knowledge graph, insurance type set, and weights for each insurance type specifically includes: Establish an insurance product set P and an information field-insurance product relationship set RP, both initialized as empty sets. Iterate through each information field in the information field set. i Traverse each field in the knowledge graph i Composition Information Fields - Insurance Product Relationships p j Let the relationship be rp i,j The corresponding relation weight is WRP. i,j Additionally, p can be obtained from the knowledge graph. j Corresponding insurance type j Let the weight calculated for this type of insurance be wt. k If type j If it belongs to T, then p j Add P, and rp i,j Adding RP, each insurance product in P eventually becomes p k weight wp k for: rec k For insurance products p in the knowledge graph k Product recommendation rate; Iterate through each insurance product p in the information field set. k For each p k Iterate through each information field in the information field set. i If field i The "Is field required?" attribute is true and p k Does not conform to field i The constraints, or p k If the insurer's prediction in the knowledge graph does not match the customer's prediction, then p will be... k Remove from P and use the final P as the recommended set of insurance products; If P is ultimately empty, and an insurance product was previously removed from P because the corresponding insurer did not match the customer's prediction, then add that insurance product back into P.
6. The interactive intelligent insurance recommendation method according to claim 1, characterized in that, The process involves determining recommended insurance products based on the calculated insurance product ranking and corresponding scores, and generating recommendation reasons based on the constructed knowledge graph, specifically including: For each insurance product p in the calculated set of insurance products P k According to its corresponding insurance type k Group the insurance products, keeping only the one with the highest weight (wp) in each group. Then, select the top k groups of insurance products with the highest weights, based on each information field in the information field set. i Based on the corresponding node information in the knowledge graph, as well as other general information about the product in the knowledge graph, recommendation text is generated according to predefined rule templates.
7. An interactive intelligent insurance recommendation system based on knowledge graphs, characterized in that, include: The knowledge graph construction module constructs the corresponding nodes and relationships of the knowledge graph of information fields, insurance types, insurance products, and insurers based on the survey information from experts, and calculates the weights on the relationships; The knowledge graph storage and query module is used to store the constructed information fields - insurance type - insurance product - insurer knowledge graph and provide query services; The dialogue module is used to ask users for their needs and collect natural language description text input by the users. The entity recognition and name standardization module identifies the requirement information based on the natural language description text collected by the dialogue module and performs standardization operations on the requirement information. The requirement information includes a set of required information fields, a set of optional information fields, a set of insurers, and a set of insurance types. The insurance product recommendation module calculates the insurance type set and the weight of each insurance type, the insurer set, and the constructed knowledge graph to recommend insurance products based on the aforementioned demand information and the knowledge graph. The management module is responsible for calling the knowledge graph storage and query module, dialogue module, entity recognition and name standardization module, and insurance product recommendation module to make insurance recommendations; The insurance product recommendation module specifically includes: The insurance type calculation module uses the user's information field set, the predicted insurance type set, and the constructed knowledge graph to calculate the insurance type set and the weight of each insurance type; The insurance product calculation module uses the user's information field set, insurance type set and its weight, insurer set, and constructed knowledge graph to calculate the recommended insurance products and their corresponding scores. The recommendation reason generation module calculates the insurance products and their corresponding rankings, determines the recommended insurance products, and generates recommendation reasons based on the constructed knowledge graph. Based on the defined schema layer, fill in the information fields, insurance type, insurance product, and insurer as data nodes of the knowledge graph, and construct the relationships between the nodes as the final knowledge graph; For the weighted attributes of relations in a knowledge graph, if the nodes on both sides of the relation do not involve feature information, the weight is set to 1; if one of the nodes on both sides belongs to feature information, the weights of the two relations, information field-insurance product and information field-insurance type, are calculated using the following algorithm: There are m experts, and the feature information of the information field is t. i The insurance product or type of insurance is c. j The relationship between information fields and insurance products or information fields and insurance types. i,j The weight w on i,j The calculation formula is: Among them, rec i,j It is an expert's view on relationship r i,j The level of recognition, imp i For feature information t i Importance, n i,j It is believed that c j Having feature information t i The number of experts, N i It contains feature information t i The number of relationships between information field - insurance product or information field - insurance type, where N is the total number of relationships between all information fields - insurance products or information fields - insurance types. It is a smoothing factor with a value greater than or equal to 0. It is a selection factor.
8. The interactive intelligent insurance recommendation system according to claim 7, characterized in that, The entity recognition and name standardization module specifically includes: The dictionary storage and query module is used to store the constructed insurance type dictionary, insurer dictionary, and field information component dictionary, and provides query services. The entity recognition module uses an entity recognition method based on bidirectional maximum character matching to identify the requirement information in the natural language description text; The standardization module uses an insurance type dictionary, an insurer dictionary, and a field information component dictionary to convert the identified non-standard insurance type names, insurer names, and information field component names into standard names. The mapping module maps the standardized names to the corresponding nodes in the information field - insurance type - insurance product - insurer knowledge graph.