Insurance clause division method and device, electronic device, and storage medium

Abstract

The embodiment of the present application provides a method and device for dividing insurance clauses, an electronic device, and a storage medium, which belongs to the field of financial technology. The method includes: obtaining an insurance clause data set, the insurance clause data set including initial insurance clauses and initial clause content data; performing clause similarity detection on any two initial clause content data to obtain a clause similarity value; determining an initial clause pair from the initial insurance clauses based on the clause similarity value; performing edge betweenness detection on the initial clause pair to obtain an initial edge betweenness; performing clause division on the first connected clause of the initial insurance clause pair based on the initial edge betweenness and a preset number of groups to determine a candidate clause group, the candidate clause group including the second connected clause; obtaining a group feedback score for the candidate clause group; performing group clause optimization on the second connected clause based on a reinforcement learning model and a group feedback score to obtain a target clause group. The embodiment of the present application can improve the accuracy and efficiency of the division of insurance clauses.

Description

Technical Field

[0001] The present application relates to the field of financial technology, and in particular to a method and device for dividing insurance clauses, an electronic device, and a storage medium. Background Art

[0002] Insurance clauses refer to the various terms and conditions specified in an insurance contract and are an essential component of the contract. They can include basic clauses, additional clauses, and special agreements. However, the number of insurance clauses is relatively large. By scientifically combining and adjusting similar and highly related insurance clauses, multiple objectives can be achieved, including improving the competitiveness of insurance products and addressing customer needs. For example, in FinTech insurance sales scenarios, by better categorizing clause combinations and offering bundled discounts, multiple customer needs can be met simultaneously, improving service quality.

[0003] However, related technologies typically employ two approaches to categorizing insurance clauses. One approach involves simply classifying auto insurance clauses by category. However, this approach lacks a deep understanding of the underlying connections and complex relationships between clauses, resulting in low accuracy in categorizing insurance clauses. The other approach relies on the experience of experts or salespeople to manually combine clauses. However, this approach is inefficient and prone to missing subtle connections between clauses, resulting in low accuracy in categorizing insurance clauses. Therefore, improving the accuracy and efficiency of insurance clause categorization has become a pressing technical challenge. Summary of the Invention

[0004] The main purpose of the embodiments of the present application is to propose a method and device for dividing insurance clauses, an electronic device, and a storage medium, aiming to improve the accuracy and efficiency of dividing insurance clauses.

[0005] To achieve the above objectives, a first aspect of an embodiment of the present application provides a method for dividing insurance clauses, the method comprising:

[0006] Acquire an insurance clause dataset, the insurance clause dataset including initial insurance clauses and initial clause content data of the initial insurance clauses, the initial insurance clauses being used to represent nodes of a preset clause topology network;

[0007] Performing a clause similarity test on any two of the initial clause content data to obtain a clause similarity value;

[0008] Determining an initial clause pair from the initial insurance clauses based on the clause similarity value, wherein the initial clause pair is used to represent an edge of the clause topology network, the initial insurance clause pair includes a first connection clause, and the edge of the clause topology network is used to represent a path between two connected first connection clauses;

[0009] Performing an edge betweenness test on the initial clause pair to obtain an initial edge betweenness, where the initial edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two of the initial insurance clauses;

[0010] Dividing the first connecting clause into clauses based on the initial edge betweenness and the preset number of groups to determine a candidate clause group, wherein the candidate clause group includes the second connecting clause;

[0011] Obtaining a group feedback score for the candidate clause group;

[0012] The second connection terms are group-optimized based on a preset reinforcement learning model and the group feedback score to obtain a target term group.

[0013] In some embodiments, the performing group clause optimization on the second connection clause based on the preset reinforcement learning model and the group feedback score to obtain a target clause group includes:

[0014] extracting a demand feedback sub-score, an object feedback sub-score, and a group application sub-score from the group feedback score;

[0015] Performing an action decision on the demand feedback sub-score, the object feedback sub-score, the group application sub-score, and the clause content data of the second connection clause based on the reinforcement learning model to obtain clause action decision data, wherein the clause action decision data includes optimization feedback data;

[0016] Calculating similarity between the optimized feedback data and the preset target feedback data to obtain a feedback data similarity value;

[0017] Adjusting the second connection clause based on the feedback data similarity value and the clause action decision data to obtain an optimized connection clause;

[0018] The target term group is constructed based on the optimized connection terms of the candidate term group.

[0019] In some embodiments, performing an action decision on the demand feedback sub-score, the object feedback sub-score, the group application sub-score, and the clause content data of the second connection clause based on the reinforcement learning model to obtain clause action decision data includes:

[0020] determining a candidate combination reward value for the candidate clause group based on the demand feedback sub-score, the object feedback sub-score, and the group application sub-score;

[0021] An action decision is made on the candidate combination reward value and the clause content data of the second connection clause based on the reinforcement learning model to obtain the clause action decision data.

[0022] In some embodiments, dividing the first connection terms based on the initial edge betweenness and the preset number of groups to determine candidate term groups includes:

[0023] Determining a central term pair from the initial term pairs based on the initial edge betweenness, wherein the central term pair is used to indicate the initial term pair with the largest initial edge betweenness;

[0024] The first connection clause is divided into clauses based on the central clause pair and the preset number of groups to determine the candidate clause group.

[0025] In some embodiments, dividing the first connection clauses based on the central clause pair and the preset number of groups to determine the candidate clause groups includes:

[0026] Dividing the first connecting clauses into clauses based on the central clause pair to determine a first candidate clause group and a second candidate clause group;

[0027] determining a first candidate group weight based on the initial edge betweenness included in the first candidate term group;

[0028] determining a second candidate group weight based on the initial edge betweenness included in the second candidate term group;

[0029] Dividing the first connection clause into clauses based on the first candidate group weight and the second candidate group weight to obtain an initial clause group;

[0030] The initial clause group is updated based on the preset number of groups to determine the candidate clause group.

[0031] In some embodiments, dividing the first connection terms based on the first candidate group weight and the second candidate group weight to obtain an initial term group includes:

[0032] determining a preset weight difference threshold based on the number of combinations of the first candidate clause group and the second candidate clause group;

[0033] If the weight difference between the first candidate group weight and the second candidate group weight is greater than the preset weight difference threshold, splitting the central clause pair to obtain a split central clause;

[0034] The first connection clause is divided into clauses based on the first candidate group weight, the second candidate group weight and the split center clause to obtain the initial clause group.

[0035] In some embodiments, updating the initial clause group based on the preset number of groups to determine the candidate clause group includes:

[0036] If the number of the initial clause groups is less than the preset number of groups, performing an edge betweenness test on the initial clause pairs in the initial clause group to obtain a candidate edge betweenness, where the candidate edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two third connected clauses in the initial clause group;

[0037] Performing clause division on the third connection clause in the initial clause group based on the candidate edge betweenness to obtain a third candidate clause group;

[0038] The candidate clause group is determined based on the third candidate clause group and the undivided initial clause group.

[0039] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides an insurance clause division device, the device comprising:

[0040] A first acquisition module is configured to acquire an insurance clause dataset, wherein the insurance clause dataset includes initial insurance clauses and initial clause content data of the initial insurance clauses, wherein the initial insurance clauses are used to represent nodes of a preset clause topology network;

[0041] A similarity detection module is used to perform a clause similarity detection on any two of the initial clause content data to obtain a clause similarity value;

[0042] a clause pair determination module, configured to determine an initial clause pair from the initial insurance clauses based on the clause similarity value, wherein the initial clause pair is used to represent an edge of the clause topology network, the initial insurance clause pair includes a first connection clause, and the edge of the clause topology network is used to represent a path between two connected first connection clauses;

[0043] an edge betweenness detection module, configured to perform edge betweenness detection on the initial clause pair to obtain an initial edge betweenness, wherein the initial edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two of the initial insurance clauses;

[0044] a clause division module, configured to divide the first connecting clause into clauses based on the initial edge betweenness and a preset number of groups, and determine a candidate clause group, wherein the candidate clause group includes the second connecting clause;

[0045] A second acquisition module is used to obtain a group feedback score of the candidate clause group;

[0046] A group clause optimization module is used to perform group clause optimization on the second connection clause based on a preset reinforcement learning model and the group feedback score to obtain a target clause group.

[0047] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the method described in the first aspect when executing the computer program.

[0048] To achieve the above-mentioned purpose, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method described in the first aspect.

[0049] The present application proposes an insurance clause division method and device, electronic device, and storage medium. The method comprises: obtaining an insurance clause dataset, wherein the insurance clause dataset includes initial insurance clauses and initial clause content data of the initial insurance clauses, wherein the initial insurance clauses are used to represent nodes of a preset clause topology network; performing clause similarity detection on any two initial clause content data to obtain a clause similarity value; determining an initial clause pair from the initial insurance clauses based on the clause similarity value, wherein the initial clause pair is used to represent an edge of a clause topology network, wherein the initial insurance clause pair includes a first connection clause, and the edge of the clause topology network is used to represent a path between the two connected first connection clauses; further, performing edge betweenness detection on the initial clause pair to obtain an initial edge betweenness, wherein the initial edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two initial insurance clauses; further, performing clause division on the first connection clause based on the initial edge betweenness and a preset number of groups to determine a candidate clause group, wherein the candidate clause group includes a second connection clause; further, obtaining a group feedback score for the candidate clause group; and performing group clause optimization on the second connection clause based on a preset reinforcement learning model and the group feedback score to obtain a target clause group. Compared to related technologies that simply classify clauses based on category or rely on human experience to combine clauses, this application can deeply understand the potential connections and complex relationships between clauses. It first determines initial clause pairs with similarity, then automatically classifies insurance clauses based on edge betweenness and the preset number of clauses. It further optimizes the grouping of insurance clauses based on group feedback scoring and reinforcement learning models, avoiding missing subtle connections between clauses and effectively improving the accuracy of insurance clause classification. Therefore, the embodiments of this application can effectively improve the accuracy and efficiency of insurance clause classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flow chart of the insurance clause division method provided in the embodiment of the present application;

[0051] Figure 2 This is a specific schematic diagram of a clause topology network provided in an embodiment of the present application;

[0052] Figure 3 yes Figure 1 A flowchart of step S150 in FIG.

[0053] Figure 4 yes Figure 3 A flowchart of step S320 in FIG.

[0054] Figure 5 yes Figure 4 A flowchart of step S440 in FIG.

[0055] Figure 6 yes Figure 4 A flowchart of step S450 in FIG.

[0056] Figure 7 yes Figure 1 A flowchart of step S170 in FIG.

[0057] Figure 8 yes Figure 7 A flowchart of step S720 in FIG.

[0058] Figure 9 This is a structural diagram of the insurance clause division device provided in an embodiment of the present application;

[0059] Figure 10 This is a hardware structure diagram of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0061] It should be noted that although the device schematics illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the device or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0063] First, let’s analyze some of the terms used in this application:

[0064] Artificial Intelligence (AI) is a new technical discipline that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence. A branch of computer science, AI seeks to understand the essence of intelligence and create new intelligent machines that can respond in a manner similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thinking. It also encompasses theories, methods, technologies, and application systems that use digital computers or digital computer-controlled machines to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.

[0065] Natural Language Processing (NLP): NLP uses computers to process, understand, and apply human languages ​​(such as Chinese and English). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is commonly used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis, and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0066] A knowledge graph is an information system used to store and represent knowledge. It organizes entities and their relationships in a graph format. A knowledge graph typically consists of nodes (representing entities such as people, places, and things) and edges (representing relationships between entities). This structure enables knowledge to be represented and searched in an intuitive manner.

[0067] Network topology: A term used to describe the connections between network elements. It involves the arrangement and layout of nodes (representing network entities, such as devices and items) and edges (representing the connections between nodes). Network topology focuses not only on physical connections but also on logical connections, forming the foundation for building and managing networks.

[0068] Girvan-Newman algorithm: It is a network analysis algorithm used for community discovery. It identifies the community structure in the network by calculating the edge betweenness centrality (EBC) of each edge in the network.

[0069] Betweenness Centrality: This refers to the number of shortest paths in a network that pass through an edge. Specifically, for each pair of nodes in the network, the shortest path is the shortest path between the two nodes, and the betweenness centrality of an edge is the number of these shortest paths that pass through the edge.

[0070] Deep Q-Learning (DQL) is a model that combines deep learning with the Q-Learning reinforcement learning algorithm. The DQL model is designed to solve decision-making and control problems. The main goal of the DQL model is to approximate the Q function using deep neural networks, thereby overcoming the computational difficulties of traditional Q-Learning when dealing with large or continuous state spaces.

[0071] Insurance clauses refer to the various terms and conditions specified in an insurance contract and are an essential component of the contract. They can include basic clauses, additional clauses, and special agreements. However, the number of insurance clauses is relatively large. By scientifically combining and adjusting similar and highly correlated insurance clauses, we can better achieve multiple goals, such as improving the competitiveness of insurance products and addressing customer needs. For example, in the FinTech insurance sales scenario, the number of insurance clauses is very large. By better classifying clause combinations and offering combined discounts, we can simultaneously meet the diverse needs of customers and improve service quality.

[0072] However, related technologies typically employ two approaches to categorizing insurance clauses. One approach involves simply classifying auto insurance clauses by category. However, this approach lacks a deep understanding of the underlying connections and complex relationships between clauses, resulting in low accuracy in categorizing insurance clauses. The other approach relies on the experience of experts or salespeople to manually combine clauses. However, this approach is inefficient and prone to missing subtle connections between clauses, resulting in low accuracy in categorizing insurance clauses. Therefore, improving the accuracy and efficiency of insurance clause categorization has become a pressing technical challenge.

[0073] Based on this, the embodiments of the present application provide a method and device for dividing insurance clauses, an electronic device, and a storage medium, aiming to improve the accuracy and efficiency of dividing insurance clauses.

[0074] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0075] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0076] The insurance clause division method provided in the embodiment of the present application relates to the field of artificial intelligence technology. The recommended method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, or can be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, or as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the insurance clause division method, etc., but is not limited to the above forms.

[0077] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0078] It should be noted that in each specific embodiment of the present application, when it comes to the need to perform relevant processing based on data related to the identity or characteristics of the object, such as object operation information, object behavior data, and object feedback data, the permission or consent of the object will be obtained first, and the collection, use and processing of such data will comply with relevant laws, regulations and standards. In addition, when the embodiment of the present application needs to obtain the sensitive personal information of the object, the separate permission or consent of the object will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the separate permission or consent of the object, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0079] See also Figure 1 , Figure 1 This is an optional flow chart of the insurance clause division method provided in the embodiment of the present application. In some embodiments of the present application, Figure 1 The method may specifically include but is not limited to steps S110 to S170.

[0080] Step S110, obtaining an insurance clause dataset;

[0081] Step S120, performing a clause similarity test on any two initial clause content data to obtain a clause similarity value;

[0082] Step S130, determining an initial clause pair from the initial insurance clauses based on the clause similarity value;

[0083] Step S140, performing edge betweenness detection on the initial clause pair to obtain initial edge betweenness;

[0084] Step S150 , dividing the first connection clause into clauses based on the initial edge betweenness and the preset number of groups to determine candidate clause groups;

[0085] Step S160, obtaining group feedback scores of the candidate clause group;

[0086] Step S170 , performing group clause optimization on the second connection clause based on a preset reinforcement learning model and group feedback score to obtain a target clause group.

[0087] In steps S110 to S170 of this application, in the insurance sales scenario of financial technology, the number of insurance clauses is very large. By better dividing the clause combinations and providing combination discounts, it is possible to meet the various needs of customers at the same time and improve the quality of service. For example, the car insurance clauses in the insurance industry can usually be divided into multiple types, such as basic insurance, third-party liability insurance, vehicle occupant liability insurance, etc. The existing clauses are very similar or even the same, but because we don’t know what clauses there are, we often have to add new clauses or how to combine the existing few clauses. Based on this, by dividing the insurance clauses into clause combinations and providing combination discounts, we can better determine the insurance coverage, insurance amount and deductible information under different insurance combinations, thereby improving the convenience of the insured. Compared to related technologies that simply classify clauses based on category or rely on human experience to combine clauses, this application can deeply understand the potential connections and complex relationships between clauses. It first determines initial clause pairs with similarity, then automatically classifies insurance clauses based on edge betweenness and the preset number of clauses. It further optimizes the grouping of insurance clauses based on group feedback scoring and reinforcement learning models, avoiding missing subtle connections between clauses and effectively improving the accuracy of insurance clause classification. Therefore, the embodiments of this application can effectively improve the accuracy and efficiency of insurance clause classification.

[0088] In step S110 of some embodiments, the insurance clause data set refers to a data set used to store all insurance clauses that currently need to be divided. The insurance clause data set includes multiple initial insurance clauses and initial clause content data of each initial insurance clause. The initial clause content data is used to characterize the specific content of the corresponding initial insurance clause. For example, in the fintech auto insurance clause division combination scenario, the insurance clause data set may include all clauses related to auto insurance, such as basic insurance (such as vehicle loss insurance), third-party liability insurance, vehicle occupant liability insurance, spontaneous combustion insurance, etc., without specific limitation. In addition, the data in the insurance clause data set of this application can be reasonably obtained from publicly available insurance company databases, public policy documents or industry reports, etc., to ensure the integrity and accuracy of the data, without specific limitation.

[0089] It should be noted that to gain a deeper understanding of the potential connections and complex relationships between insurance clauses, this application can incorporate graph theory, specifically the Girvan-Newman algorithm, to identify the association structure between insurance clauses. In this case, each initial insurance clause can be used to represent a node in a pre-defined clause topology network. This pre-defined clause topology network describes the relationships and layout of each element in the network by connecting nodes and edges, thereby enabling targeted optimization.

[0090] In step S120 of some embodiments, in order to first determine whether there is a correlation between any two initial insurance clauses, the present application may perform a clause similarity test on the content data of any two initial clauses to obtain a clause similarity value. This clause similarity value is used to represent the degree of similarity between the corresponding two initial insurance clauses. The higher the clause similarity value, the higher the degree of similarity between the corresponding two initial insurance clauses. The degree of clause similarity can be reflected in the fact that the two clauses are usually purchased together, have similarities that meet certain conditions, have the same number of insurance claims, etc., without specific limitation.

[0091] It should be noted that clause similarity detection can use natural language processing techniques, such as cosine similarity and word embedding models, to quantify the similarity between any two initial clauses. Specifically, by comparing the language structure, keywords, topics, purchase history, and claims usage of the two initial clauses, a similarity value can be derived to indicate the degree of similarity between the two initial insurance clauses.

[0092] It should be noted that, since there may be initial insurance terms that are not related to any initial insurance terms when determining the initial clause pair, they can be directly used as candidate clause groups that need to be optimized later.

[0093] In step S130 of some embodiments, after determining the clause similarity values ​​of multiple pairs of initial insurance clauses, at least one pair of initial insurance clauses whose clause similarity value is greater than or equal to a preset similarity threshold can be used as an initial clause pair. If a graph structure of a clause topology network is constructed, the initial clause pairs are used to characterize the edges of the clause topology network, the initial insurance clause pairs include first connection clauses, and the edges of the clause topology network are used to characterize the paths between the two connected first connection clauses. In this way, initial insurance clauses with similarities can be clustered together to facilitate subsequent analysis. For example, if the initial clause pair includes clause A and clause B, then clause A and clause B are first connection clauses belonging to the same initial clause pair.

[0094] For example, Figure 2 As shown, Figure 2It is a specific schematic diagram of the clause topology network provided in the embodiment of the present application. Assuming that the insurance clause data set includes five initial insurance clauses, namely Clause A (basic insurance), Clause B (third party liability insurance), Clause C (vehicle personnel liability insurance), Clause D (whole vehicle theft and robbery insurance), and Clause E (spontaneous combustion insurance), after performing clause similarity detection on any two initial clause content data, 10 clause similarity values ​​can be obtained. Further, by comparing the clause similarity value with the preset similarity threshold, it can be determined that the initial clause pairs with similar associations include clause pair (A, B) (i.e., clause A and clause B), clause pair (A, C), clause pair (B, C), clause pair (B, D), clause pair (B, E), clause pair (C, D), and clause pair (D, E).

[0095] In step S140 of some embodiments, after a plurality of initial clause pairs are determined, edge betweenness detection can be performed on the edges corresponding to the initial clause pairs to obtain the initial edge betweenness corresponding to each initial clause pair. The initial edge betweenness of the present application also represents the betweenness centrality of the initial clause pair, that is, it is used to characterize the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two initial insurance clauses. Specifically, the shortest path of the node pair formed by any two initial insurance clauses can be determined first, and then the initial edge betweenness can be determined based on the number of times the edges corresponding to the initial clause pairs appear in these shortest paths.

[0096] For example, Figure 2 As shown, the shortest path of the clause pair (A, B) is (A, B) (i.e., from the node corresponding to clause A to the node corresponding to clause B), the shortest paths of the clause pair (A, C), clause pair (B, C), clause pair (B, D), clause pair (B, E), clause pair (C, D) and clause pair (D, E) are all themselves, the shortest path of the clause pair (A, D) is (A, B, D) (i.e., from the node corresponding to clause A to the node corresponding to clause B and then to the node corresponding to clause D) or (A, C, D), the shortest path of the clause pair (A, E) is (A, B, E), and the shortest path of the clause pair (C, E) is (C, B, E) or (C, D, E). In this way, we can get the initial edge betweenness of the initial clause pair (A,B) is 3, that is, the shortest path of the clause pair (A,B), the shortest path of the clause pair (A,D) and the shortest path of the clause pair (A,E); the initial edge betweenness of the initial clause pair (A,C) is 2, the initial edge betweenness of the initial clause pair (B,C) is 1, the initial edge betweenness of the initial clause pair (B,D) is 1, the initial edge betweenness of the initial clause pair (B,E) is 3, the initial edge betweenness of the initial clause pair (C,D) is 3, and the initial edge betweenness of the initial clause pair (D,E) is 2.

[0097] In some embodiments, in step S150, after obtaining the initial edge betweenness corresponding to each initial clause pair, the first connecting clauses can be divided into clauses based on the initial edge betweenness and the preset number of groups to determine candidate clause groups. Furthermore, the initial insurance clauses included in the candidate clause groups can be used as second connecting clauses. The number of second connecting clauses included in the candidate clause groups can be one, two, three, etc., without limitation.

[0098] See also Figure 3 , Figure 3 This is a specific flow chart of step S150 provided in an embodiment of the present application. In some embodiments of the present application, step S150 may specifically include but is not limited to steps S310 to S320.

[0099] Step S310, determining a central term pair from the initial term pairs based on the initial edge betweenness;

[0100] Step S320 : dividing the first connection clause into clauses based on the central clause pair and the preset number of groups to determine candidate clause groups.

[0101] In step S310 of some embodiments, the central term pair is used to indicate the initial term pair with the largest initial edge betweenness, and is used to represent the edge that has the greatest impact on the graph structure. Figure 2 As shown, after determining that the initial edge betweenness of the initial clause pair (A, B) is 3, the initial edge betweenness of the initial clause pair (A, C) is 2, the initial edge betweenness of the initial clause pair (B, C) is 1, the initial edge betweenness of the initial clause pair (B, D) is 1, the initial edge betweenness of the initial clause pair (B, E) is 3, the initial edge betweenness of the initial clause pair (C, D) is 3, and the initial edge betweenness of the initial clause pair (D, E) is 2, by comparing the initial edge betweennesses, it can be determined that the largest initial clause pairs include the initial clause pair (A, B), the initial clause pair (B, E) and the initial clause pair (C, D), and then it can be determined that the central clause pairs in the current clause topology network include the clause pair (A, B), the clause pair (B, E) and the clause pair (C, D).

[0102] In step S320 of some embodiments, the preset number of groups refers to the pre-set number of candidate clause groups that need to be divided, which can also be referred to as the number of subgraphs that need to be divided. After determining the central clause pair, the first connection clause can be further divided into clauses based on the central clause pair and the preset number of groups to determine the preset number of candidate clause groups. It should be noted that the subgraph mentioned in this application is used to represent a clause combination method.

[0103] In some specific embodiments, after determining the central clause pair, any high betweenness centrality edge can be removed, such as the edge where clause pair (A, B), clause pair (B, E) and clause pair (C, D) are located. In this case, the entire graph can be divided into two subgraphs of various cases. For example, removing the edge where clause pair (A, B) is located, two subgraphs can be obtained, including the subgraphs of clause A and clause B. Figure 1 , and a sub-clause containing clauses C, D, and E Figure 2 Alternatively, by removing the edge where the clause pair (B, E) is located, we can obtain two subgraphs, one containing clause B and the other containing clause E. Figure 3 , and a sub-clause containing clauses A, C, and D Figure 4 Alternatively, by removing the edge where the clause pair (C, D) is located, we can obtain two subgraphs, one containing clauses C and one containing clauses D. Figure 5 , and a sub-clause containing clauses A, B, and E Figure 6 The specific division method can be adjusted according to the specific parameter settings of the Girvan-Newman algorithm on which it is based, and is not limited.

[0104] In other specific embodiments, the present application may consider the weight of the divided subgraphs when performing term division to improve the balance of the division. The weight of the divided subgraphs may be determined by the sum of the initial edge betweennesses of the edges corresponding to all initial term pairs contained in the subgraphs.

[0105] See also Figure 4 , Figure 4 This is a specific flow chart of step S320 provided in an embodiment of the present application. In some embodiments of the present application, step S320 may specifically include but is not limited to steps S410 to S450.

[0106] Step S410 , dividing the first connection clause into clauses based on the central clause pair to determine a first candidate clause group and a second candidate clause group;

[0107] Step S420 , determining the weight of the first candidate group based on the initial edge betweenness included in the first candidate term group;

[0108] Step S430 , determining the weight of the second candidate group based on the initial edge betweenness included in the second candidate term group;

[0109] Step S440 , dividing the first connection clause based on the weight of the first candidate group and the weight of the second candidate group to obtain an initial clause group;

[0110] Step S450 : updating the initial clause groups based on the preset number of groups to determine candidate clause groups.

[0111] In step S410 of some embodiments, the first connection clauses are divided into clauses based on the central clause pair, and the clause group containing only the central clause pair can be used as the first candidate clause group, and the clause group not containing the central clause pair can be used as the second candidate clause group. For example, by removing the edge where the central clause pair (A, B) is located, two subgraphs can be obtained, the subgraph containing clause A and clause B. Figure 1 , and a sub-clause containing clauses C, D, and E Figure 2 , at this time Figure 1 The clause group formed is equivalent to the first candidate clause group. Figure 2 The formed clause group is equivalent to the second candidate clause group.

[0112] In step S420 and step S430 of some embodiments, further, since the edge betweenness can reflect the importance and intermediary role of the clause in the clause topology network, the higher the edge betweenness usually means that the clause is more important to the connectivity and information flow of other clauses. Therefore, by considering these edge betweenness, a weight value can be assigned to the candidate clause group so that its importance can be considered in subsequent optimization and evaluation. Specifically, the initial edge betweenness contained in the first candidate clause group can be added to obtain the weight of the first candidate group. The initial edge betweenness contained in the second candidate clause group is added to obtain the weight of the second candidate group. Figure 2 As shown, if the first candidate clause group includes clause A and clause B, and the initial edge betweenness of the initial clause pair (A, B) is 3, then the weight of the first candidate group can be determined to be 3. Similarly, if the second candidate clause group includes clause C, clause D, and clause E, and the initial edge betweenness of the initial clause pair (C, D) is 3, and the initial edge betweenness of the initial clause pair (D, E) is 2, then the weight of the second candidate group can be determined to be 5.

[0113] In some embodiments, in step S440, the first connected clauses may be further divided based on the weight of the first candidate group and the weight of the second candidate group to obtain an initial clause group. In other words, the present application can determine which clauses can be combined into a clause combination by comparing the weights of each clause group, emphasizing the importance of finding complementary and interconnected clause combinations to enhance overall performance and relevance.

[0114] In some embodiments, in step S450, after determining the initial clause group, it is necessary to consider whether the number of subgraphs currently divided has reached the preset number of groups. If not, it is necessary to continue dividing the subgraph to obtain more accurate and detailed clause combinations. This ensures that the final candidate clause group meets the preset number of groups and reaches an optimal state.

[0115] In the above embodiment, by considering the weights of the divided candidate clause groups to determine whether to perform balanced division on the graph corresponding to the clause topology network, the accuracy and comprehensiveness of the combination allocation can be improved.

[0116] See also Figure 5 , Figure 5 This is a specific flow chart of step S440 provided in an embodiment of the present application. In some embodiments of the present application, step S440 may specifically include but is not limited to steps S510 to S530.

[0117] Step S510, determining a preset weight difference threshold based on the number of combinations of the first candidate clause group and the second candidate clause group;

[0118] Step S520: If the weight difference between the first candidate group weight and the second candidate group weight is greater than a preset weight difference threshold, the central clause pair is split to obtain a split central clause;

[0119] Step S530 : dividing the first connection clause based on the first candidate group weight, the second candidate group weight and the split center clause to obtain an initial clause group.

[0120] In step S510 of some embodiments, when performing the specific division of clauses, the present application may first determine a preset weight difference threshold based on the number of combinations of the first candidate clause group and the second candidate clause group. The preset weight difference threshold is used as a benchmark value for measuring the difference in importance between the two candidate clause groups. To ensure its scientificity and rationality, the setting of this threshold is usually based on data statistical analysis, and factors may include: the average weight of the candidate clause group, the number and complexity of the clauses in the group, industry standards or historical data in related fields, etc. When the number of combinations of candidate clause groups is large and involves a wide variety of clauses, the weight threshold should also be appropriately adjusted to reflect the complexity of the data.

[0121] In step S520 of some embodiments, the weight difference between the weight of the first candidate group and the weight of the second candidate group can be further calculated. If the weight difference is greater than or equal to the preset weight difference threshold, it means that the current division level is insufficient. In this case, the central clause pair can be split to obtain the split central clause. In this way, the functions and responsibilities of different clause groups can be clearly divided and potential confusion caused by the difference in weights between clauses can be eliminated. For example, Figure 2As shown, if the first candidate clause group includes clause A and clause B, and the initial edge betweenness of the initial clause pair (A, B) is 3, then the weight of the first candidate group can be determined to be 3. Similarly, if the second candidate clause group includes clause C, clause D, and clause E, and the initial edge betweenness of the initial clause pair (C, D) is 3, and the initial edge betweenness of the initial clause pair (D, E) is 2, then the weight of the second candidate group can be determined to be 5. If the preset weight difference threshold is 1, the central clause pair (C, D) can be split to obtain the split central clause C and the split central clause D.

[0122] In step S530 of some embodiments, further, the first connection clause is divided based on the first candidate group weight, the second candidate group weight and the split center clause to obtain an initial clause group. As in the above example, the first candidate group weight of the initial clause pair (A, B) at this time is 3, the split center clause C can be directly used as a candidate clause group, and the second candidate group weight of the second candidate clause group composed of the split center clause D and clause E is 2. In this way, if the weight difference between the re-divided candidate clause groups meets the requirement of being less than or equal to the preset weight difference threshold, these candidate clause groups that meet the requirements can be used as the initial clause group. If the weight difference between the re-divided candidate clause groups still does not meet the requirement of being less than or equal to the preset weight difference threshold, the central clause pair can continue to be split, or the preset weight difference threshold can be re-determined without specific limitation.

[0123] In the above embodiment, the present application can flexibly respond to different situations and improve the overall effect by dynamically optimizing the terms in the candidate term group and considering the preset weight difference threshold and weight difference value.

[0124] It should be noted that due to the Girvan-Newman algorithm-based clause partitioning method, some edge cases may occur during the partitioning process, causing nodes included in node edges with high betweenness centrality to be easily assigned to subgraphs that do not conform to the actual partitioning algorithm. This is also due to the subsequent edge removal leading to splitting and boundary fuzziness in the algorithm. Based on this, this application will introduce a reinforcement learning model in subsequent steps to optimize the clauses of each combination to improve the accuracy of the clause combination.

[0125] See also Figure 6 , Figure 6 This is a specific flow chart of step S450 provided in an embodiment of the present application. In some embodiments of the present application, step S450 may specifically include but is not limited to steps S610 to S630.

[0126] Step S610: If the number of initial clause groups is less than the preset number of groups, edge betweenness detection is performed on the initial clause pairs in the initial clause groups to obtain candidate edge betweenness;

[0127] Step S620 , dividing the third connected clause in the initial clause group based on the candidate edge betweenness to obtain a third candidate clause group;

[0128] Step S630 : determining a candidate clause group based on the third candidate clause group and the undivided initial clause group.

[0129] In step S610 of some embodiments, the candidate edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two third-connected clauses in the initial clause group. It is understandable that if the number of initial clause groups is less than the preset number of groups, it indicates that the number of current clause groupings is insufficient to meet the target requirements. At this time, subgraph partitioning can be continued for each initial clause group, that is, the initial clause pairs in the initial clause group are first tested for edge betweenness to obtain candidate edge betweenness. The specific detection of candidate edge betweenness can refer to the detection of initial edge betweenness in the above embodiment, and will not be repeated here.

[0130] In some embodiments, in step S620, the third connecting clause refers to an insurance clause included in the initial clause group. Furthermore, the third connecting clauses in the initial clause group are partitioned based on the candidate edge betweenness to obtain a third candidate clause group. This can be done by referring to step S150 above, i.e., using the candidate edge betweenness as a basis to better identify the relationships between the third connecting clauses in the initial clause group and form a new clause grouping.

[0131] In some embodiments, in step S630, the third candidate clause group is merged with the undivided initial clause group to ensure that all relevant clauses are considered. The undivided initial clause group refers to initial insurance clauses that have no significant correlation with other initial insurance clauses, as determined based on clause similarity values. In this manner, a predetermined number of candidate clause groups are obtained.

[0132] It should be noted that when determining the candidate clause group based on the third candidate clause group and the undivided initial clause group, if the determined candidate clause group is still less than the preset number of groups, the clause division of the determined candidate clause group will be repeated until the determined candidate clause group is equal to the preset number of groups.

[0133] In step S160 of some embodiments, after determining a preset number of candidate clause groups, each candidate clause combination obtained is equivalent to a clause combination that is preliminarily divided and that is more in line with the needs. However, whether these clause combinations are more applicable in actual applications still needs to be optimized for each candidate clause group, that is, to optimize the clause content and structure to ensure consistency and coherence between the clauses. Based on this, the present application will obtain a group feedback score for each candidate clause group, and the group feedback score is derived from the group feedback data for the candidate clause group. Among them, the group feedback data is used to characterize the data combined with actual application feedback, and may include market demand data, customer feedback data, clause purchase frequency table and other data fed back by business objects. In order to facilitate the model to learn specific data, it can be quantified into demand feedback sub-scores (corresponding to market demand data), object feedback sub-scores (corresponding to customer feedback data) and group application sub-scores (corresponding to clause purchase frequency table).

[0134] In step S170 of some embodiments, the present application may further perform group optimization on the second connected clauses based on a preset reinforcement learning model and group feedback scores to obtain a target clause group. Group clause optimization includes adjusting the content of a clause, adding or removing clauses, or combining new clause combinations. The reinforcement learning model of the present application may be constructed based on a DQL model, i.e., combining DQL with reinforcement learning to achieve automated optimization of decision management.

[0135] See also Figure 7 , Figure 7 This is a specific flow chart of step S170 provided in an embodiment of the present application. In some embodiments of the present application, step S170 may specifically include but is not limited to steps S710 to S750.

[0136] Step S710, extracting the demand feedback sub-score, the object feedback sub-score, and the group application sub-score from the group feedback score;

[0137] Step S720 , performing an action decision on the demand feedback sub-score, the object feedback sub-score, the group application sub-score, and the clause content data of the second connected clause based on the reinforcement learning model to obtain clause action decision data;

[0138] Step S730, performing similarity calculation on the optimized feedback data and the preset target feedback data to obtain a feedback data similarity value;

[0139] Step S740 , adjusting the clause data of the second connection clause based on the feedback data similarity value and the clause action decision data to obtain an optimized connection clause;

[0140] Step S750 : constructing a target clause group based on the optimized connection clauses of the candidate clause group.

[0141] In steps S710 and S720 of some embodiments, the clause action decision data includes optimized feedback data, which is used to characterize and guide the action taken by the agent next. That is, the present application can use the various attributes and historical data in the clause topology network as states, the object purchase decision and purchase results as reward functions, and use the DQL algorithm to learn how to take the best combination decision in different situations, thereby achieving faster and more accurate product combination sales. In this process, the entities and relationships in the graph can be regarded as a decision problem, and the attributes and association relationships of each entity can be regarded as a decision. Because the agent uses a deep neural network to estimate the Q-value function, the network takes the state as input and outputs the Q-value of each action. By randomly selecting actions and using experience replay technology to collect data, the agent can learn from the data how to take the best action in different states. At the same time, the value of taking each action in a specific state. By continuously iteratively updating the Q-value function, the agent can learn how to take different actions in different states to maximize the cumulative reward. Decision optimization is performed through deep neural networks and reinforcement learning algorithms.

[0142] Specifically, in the reinforcement learning model, the present application can set a memory D with an initialization capacity of N and a Q network using random parameters θ. The state s of the reinforcement learning model can include the clause content data of the second connected clause in the candidate clause group (i.e., the current state of the insurance clause and its combination), demand feedback sub-score, object feedback sub-score, group application sub-score and other data. The action a of the reinforcement learning model defines possible actions, such as adjusting the content of a clause, adding or reducing clauses, combining new clause packages, etc. The reward rt of the reinforcement learning model can be based on customer satisfaction, market response, sales data, etc., and a reward function can be designed to evaluate the effect of each action. Among them, the reward obtained based on the previous action can help the model make action decisions and obtain clause action decision data.

[0143] See also Figure 8 , Figure 8 This is a specific flow chart of step S720 provided in an embodiment of the present application. In some embodiments of the present application, step S720 may specifically include but is not limited to steps S810 to S820.

[0144] Step S810 , determining a candidate combination reward value for the candidate clause group based on the demand feedback sub-score, the object feedback sub-score, and the group application sub-score;

[0145] Step S820: Perform an action decision on the candidate combination reward value and the clause content data of the second connection clause based on the reinforcement learning model to obtain clause action decision data.

[0146] In steps S810 and S820 of some embodiments, the present application may update the demand feedback sub-score, object feedback sub-score, and group application sub-score after each action, thereby determining the candidate combination reward value for the current action. Furthermore, an action decision is made based on the candidate combination reward value and the clause content data of the second connected clause based on the reinforcement learning model to obtain clause action decision data, i.e., what to do in the next action.

[0147] It should be noted that the reinforcement learning model of the present application will undergo a cyclic training. For each candidate clause group, based on the current state s of the current clause combination, the agent uses random parameters θ after wandering the attributes and associations of the entity to derive the action a to be executed, and determines whether the sales result is successfully sold. The reward is rt and is recorded. Each subsequent action is based on the previous state s, action a, and reward rt. At the same time, in order to obtain the best effect, except for the first data in the initial state, all other actions are taken according to the optimal decision, that is, a=arg maxQ(s). In addition, the above information can be stored in the memory pool Memory D, and the state that is most similar to the state can be extracted at any time. The network in this state can be used to reduce the deviation of action a caused by environmental influences to stabilize the environment. In addition, the deviation of random parameters caused by training can be reduced by using fixed parameters to maintain network stability. After the cycle ends, the parameters and all records can be viewed, and at the same time, the model's a can be checked to see if it is better. The parameters can be adjusted to continue training.

[0148] In steps S730 to S750 of some embodiments, in order to verify whether the clause content of the candidate clause group after the action adjustment reaches the predicted target, the optimized feedback data in the clause action decision data and the preset target feedback data can be calculated for similarity to obtain a feedback data similarity value, which characterizes the gap between the next action of the action decision and the expected target. Furthermore, based on the feedback data similarity value and the clause action decision data, the clause data of the second connection clause is adjusted, that is, the content of a certain clause is adjusted, clauses are added or reduced, new clause packages are combined, etc., to obtain the optimized connection clauses contained in the adjusted candidate clause group. Moreover, the optimized connection clauses at this time represent clauses whose feedback data similarity values ​​are less than or equal to the preset feedback similarity threshold, that is, the content contained in the currently adjusted candidate clause group is more in line with actual needs, so as to obtain the target clause group.

[0149] It is understandable that in actual applications, this application can adjust the term combination strategy in real time through the DQL model to respond to market changes and customer feedback.

[0150] It should be noted that the non-Company's software tools or components that appear in the embodiments of this application are merely examples and do not represent actual use.

[0151] The insurance clause division method provided in the embodiment of the present application reveals the potential correlation between insurance clauses through graph theory methods, helps to discover hidden clause groups, and performs targeted optimization. The subgraph division algorithm is used to perform clause association analysis to reduce manual workload and improve analysis efficiency. In addition, the optimized clause structure is clearer, which can help actual application objects to more easily understand the role and benefits of different types of insurance and improve customer satisfaction. Based on this, the present application can deeply understand the potential correlation and complex relationship between clauses, and combined with subgraph division and reinforcement learning methods, it can better understand the complex relationship between clauses and make more informed decisions. Therefore, the embodiment of the present application can effectively improve the accuracy and efficiency of the division of insurance clauses.

[0152] See also Figure 9 The present application also provides an insurance clause division device that can implement the above-mentioned insurance clause division method. The device includes:

[0153] A first acquisition module 910 is configured to acquire an insurance clause dataset, the insurance clause dataset including initial insurance clauses and initial clause content data of the initial insurance clauses, the initial insurance clauses being used to represent nodes of a preset clause topology network;

[0154] A similarity detection module 920 is used to perform a clause similarity detection on any two initial clause content data to obtain a clause similarity value;

[0155] A clause pair determination module 930 is configured to determine an initial clause pair from the initial insurance clauses based on the clause similarity value, wherein the initial clause pair is used to represent an edge of a clause topology network, the initial insurance clause pair including a first connection clause, and the edge of the clause topology network is used to represent a path between two connected first connection clauses;

[0156] An edge betweenness detection module 940 is configured to perform edge betweenness detection on the initial clause pair to obtain an initial edge betweenness, where the initial edge betweenness represents the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two initial insurance clauses.

[0157] A clause partitioning module 950 is configured to partition the first connecting clause based on the initial edge betweenness and the preset number of groups, and determine a candidate clause group, where the candidate clause group includes the second connecting clause;

[0158] The second acquisition module 960 is used to obtain the group feedback score of the candidate clause group;

[0159] The group clause optimization module 970 is used to perform group clause optimization on the second connection clause based on a preset reinforcement learning model and group feedback score to obtain a target clause group.

[0160] The specific implementation of the insurance clause division device in the embodiment of the present application is basically the same as the specific implementation of the above-mentioned insurance clause division method, and will not be repeated here.

[0161] The present application also provides an electronic device comprising a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described insurance clause classification method. The electronic device can be any smart terminal, such as a tablet computer or an in-vehicle computer.

[0162] See also Figure 10 , Figure 10 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:

[0163] The processor 1010 can be implemented as a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0164] The memory 1020 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1020 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1020 and is called by the processor 1010 to execute the insurance clause division method of the embodiments of this application.

[0165] Input / output interface 1030, used to implement information input and output;

[0166] Communication interface 1040, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0167] bus 1050 , which transmits information between various components of the device (e.g., processor 1010 , memory 1020 , input / output interface 1030 , and communication interface 1040 );

[0168] The processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 are connected to each other in communication within the device via a bus 1050 .

[0169] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned insurance clause division method.

[0170] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0171] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0172] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0173] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0174] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0175] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0176] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0177] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the above-mentioned units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0178] The units described above as separate components may or may not be physically separate, and 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 these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0179] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0180] If the integrated unit is implemented in the form of a software functional unit 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 the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: various media that can store programs, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0181] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A method for dividing insurance clauses, characterized in that: The method comprises: Acquire an insurance clause dataset, the insurance clause dataset including initial insurance clauses and initial clause content data of the initial insurance clauses, the initial insurance clauses being used to represent nodes of a preset clause topology network; Performing a clause similarity test on any two of the initial clause content data to obtain a clause similarity value; determining an initial clause pair from the initial insurance clauses based on the clause similarity value, wherein the initial clause pair is used to represent an edge of the clause topology network, the initial insurance clause pair includes a first connection clause, and the edge of the clause topology network is used to represent a path between two connected first connection clauses; Performing an edge betweenness test on the initial clause pair to obtain an initial edge betweenness, where the initial edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two of the initial insurance clauses; Determining a central clause pair from the initial clause pairs based on the initial edge betweenness, the central clause pair being used to indicate the initial clause pair with the largest initial edge betweenness; performing clause division on the first connection clause based on the central clause pair and a preset number of groups to determine a candidate clause group, wherein the candidate clause group includes a second connection clause; performing clause division on the first connection clause based on the central clause pair and the preset number of groups to determine a candidate clause group, including: performing clause division on the first connection clause based on the central clause pair to determine a first candidate clause group and a second candidate clause group, the first candidate clause group being used to indicate a clause group containing only the central clause pair, and the second candidate clause group being used to indicate a clause group not containing the central clause pair as a second candidate clause group; determining a corresponding candidate group weight based on the initial edge betweenness contained in the candidate clause group, and performing clause division on the first connection clause based on the candidate group weight and the preset number of groups to determine a candidate clause group; Obtaining a group feedback score for the candidate clause group; Extract the demand feedback sub-score, the object feedback sub-score and the group application sub-score from the group feedback score; perform action decisions on the demand feedback sub-score, the object feedback sub-score, the group application sub-score and the clause content data of the second connection clause based on a preset reinforcement learning model to obtain clause action decision data, wherein the clause action decision data includes optimized feedback data; perform similarity calculation on the optimized feedback data and the preset target feedback data to obtain a feedback data similarity value; perform clause data adjustment on the second connection clause based on the feedback data similarity value and the clause action decision data to obtain an optimized connection clause; and construct a target clause group based on the optimized connection clause of the candidate clause group.

2. The method according to claim 1, characterized in that The performing an action decision on the demand feedback sub-score, the object feedback sub-score, the group application sub-score, and the clause content data of the second connection clause based on the reinforcement learning model to obtain clause action decision data includes: determining a candidate combination reward value for the candidate clause group based on the demand feedback sub-score, the object feedback sub-score, and the group application sub-score; An action decision is made on the candidate combination reward value and the clause content data of the second connection clause based on the reinforcement learning model to obtain the clause action decision data.

3. The method according to claim 1, characterized in that The step of determining corresponding candidate group weights according to the initial edge betweenness included in the candidate term group, and dividing the first connection term according to the candidate group weights and a preset number of groups to determine the candidate term group includes: determining a first candidate group weight based on the initial edge betweenness included in the first candidate term group; determining a second candidate group weight based on the initial edge betweenness included in the second candidate term group; Dividing the first connection clause into clauses based on the first candidate group weight and the second candidate group weight to obtain an initial clause group; The initial clause group is updated based on the preset number of groups to determine the candidate clause group.

4. The method according to claim 3, characterized in that The step of dividing the first connection terms based on the first candidate group weight and the second candidate group weight to obtain an initial term group includes: determining a preset weight difference threshold based on the number of combinations of the first candidate clause group and the second candidate clause group; If the weight difference between the first candidate group weight and the second candidate group weight is greater than the preset weight difference threshold, splitting the central clause pair to obtain a split central clause; The first connection clause is divided into clauses based on the first candidate group weight, the second candidate group weight and the split center clause to obtain the initial clause group.

5. The method according to claim 3, characterized in that The updating of the initial clause group based on the preset number of groups to determine the candidate clause group includes: If the number of the initial clause groups is less than the preset number of groups, performing an edge betweenness test on the initial clause pairs in the initial clause group to obtain a candidate edge betweenness, where the candidate edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two third connected clauses in the initial clause group; Performing clause division on the third connection clause in the initial clause group based on the candidate edge betweenness to obtain a third candidate clause group; The candidate clause group is determined based on the third candidate clause group and the undivided initial clause group.

6. An insurance clause division device, characterized in that: The device comprises: A first acquisition module is configured to acquire an insurance clause dataset, wherein the insurance clause dataset includes initial insurance clauses and initial clause content data of the initial insurance clauses, wherein the initial insurance clauses are used to represent nodes of a preset clause topology network; A similarity detection module is used to perform a clause similarity detection on any two of the initial clause content data to obtain a clause similarity value; a clause pair determination module, configured to determine an initial clause pair from the initial insurance clauses based on the clause similarity value, wherein the initial clause pair is used to represent an edge of the clause topology network, the initial insurance clause pair includes a first connection clause, and the edge of the clause topology network is used to represent a path between two connected first connection clauses; an edge betweenness detection module, configured to perform edge betweenness detection on the initial clause pair to obtain an initial edge betweenness, wherein the initial edge betweenness is used to represent the number of times the edge corresponding to the initial clause pair appears in the shortest path formed by any two of the initial insurance clauses; A clause division module, configured to determine a central clause pair from the initial clause pairs based on the initial edge betweenness, the central clause pair being used to indicate the initial clause pair with the largest initial edge betweenness; perform clause division on the first connection clause based on the central clause pair and a preset number of groups to determine a candidate clause group, wherein the candidate clause group includes a second connection clause; the clause division on the first connection clause based on the central clause pair and the preset number of groups to determine the candidate clause group comprises: perform clause division on the first connection clause based on the central clause pair to determine a first candidate clause group and a second candidate clause group, the first candidate clause group being used to indicate a clause group containing only the central clause pair, and the second candidate clause group being used to indicate a clause group not containing the central clause pair as the second candidate clause group; determine a corresponding candidate group weight according to the initial edge betweenness contained in the candidate clause group, and perform clause division on the first connection clause based on the candidate group weight and the preset number of groups to determine a candidate clause group; A second acquisition module is used to obtain a group feedback score of the candidate clause group; A group clause optimization module is used to extract the demand feedback sub-score, the object feedback sub-score and the group application sub-score from the group feedback score; perform action decisions on the demand feedback sub-score, the object feedback sub-score, the group application sub-score and the clause content data of the second connection clause based on a preset reinforcement learning model to obtain clause action decision data, wherein the clause action decision data includes optimized feedback data; perform similarity calculation on the optimized feedback data and preset target feedback data to obtain a feedback data similarity value; perform clause data adjustment on the second connection clause based on the feedback data similarity value and the clause action decision data to obtain an optimized connection clause; and construct a target clause group based on the optimized connection clause of the candidate clause group.

7. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images