Insurance policy anti-fraud detection method and device, electronic equipment and storage medium
The method constructs a graph-based model to integrate and analyze policy information from multiple entities, enhancing the accuracy of fraud detection in insurance claims by leveraging edge weight adjustment and time series capture layers.
Patent Information
- Application Number
- CN202411381081.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2024-12-31
AI Technical Summary
Existing insurance policy fraud detection systems struggle to accurately identify fraudulent claims due to the complexity of integrating information from multiple third-party entities such as policy holders, assessors, and repair locations in auto insurance claims.
A method involving constructing a graph based on policy information, training an anti-fraud prediction model using sample policy labels and graph data, and merging this data with current policy information to perform accurate fraud detection using an edge weight adjustment, time series capture, and classification layers.
Enhances the accuracy of fraud detection by comprehensively analyzing multi-party information, capturing time-related patterns, and identifying fraudulent claims effectively.
Smart Images
Figure CN119228560A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of text processing technology, and in particular to an insurance policy anti-fraud detection method and device, an electronic device and a storage medium. Background Art
[0002] Policy anti-fraud prediction technology is used to detect whether the policy in the insurance claim involves fraud. For example, in the auto insurance claim scenario, it is necessary to detect whether the policy is false and fraudulent. However, since the auto claim policy contains information of multiple third-party entities, such as the policy holder, assessor, repair location, and repairman, it is difficult to make an accurate judgment by integrating multiple information when identifying whether the auto insurance policy is fraudulent. Therefore, how to accurately identify whether there is a false claim in the auto insurance policy has become an urgent problem to be solved. Summary of the invention
[0003] The main purpose of the embodiments of the present application is to propose a policy anti-fraud detection method and device, electronic equipment and storage medium, aiming to accurately identify whether there is a false claim in a motor vehicle insurance policy.
[0004] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a policy anti-fraud detection method, the method comprising:
[0005] Obtaining a policy sample and a sample policy label of the policy sample; wherein the sample policy label is used to indicate whether the policy sample is a false policy or not;
[0006] Constructing a graph based on the sample policy information of the policy sample to obtain sample policy graph data;
[0007] The preset anti-fraud prediction model is trained by using the sample policy labels and the sample policy graph data to obtain a target anti-fraud prediction model;
[0008] Acquire the current insurance policy information of the target object, and merge the graphs based on the sample insurance policy graph data and the current insurance policy information to obtain the target insurance policy graph data;
[0009] Anti-fraud prediction is performed on the current policy information through the target anti-fraud prediction model and the target policy map data to obtain anti-fraud prediction information; wherein the anti-fraud prediction information is used to indicate whether the current policy is a false policy or not.
[0010] In some embodiments, the sample policy information includes a policy number, driver information, accident report, assessor information, and beneficiary information; the sample policy information based on the policy sample is graph-constructed to obtain sample policy graph data, including:
[0011] Filter the driver information according to the insurance policy number to obtain insurance policy driver related information;
[0012] Perform node connection on the insurance policy number, the driver information and the insurance policy driver association information to obtain initial insurance policy graph structure data; the initial insurance policy graph structure data includes at least two main nodes and main node edges between any two of the main nodes;
[0013] The weight of each of the main node edges is updated according to the preset main node edge weight to obtain updated insurance policy graph structure data;
[0014] The sample policy graph data is obtained by performing random node connections based on the updated policy graph structure data, the accident report, the assessor information and the beneficiary information.
[0015] In some embodiments, the performing random node connection based on the updated insurance policy graph structure data, the accident report, the assessor information, and the beneficiary information to obtain the sample insurance policy graph data includes:
[0016] Randomly select a master node according to the updated policy graph structure data to obtain a target node;
[0017] According to the target node, the accident report, the assessor information and the beneficiary information are associated and screened to obtain at least one candidate node;
[0018] Randomly select a node from the candidate nodes to obtain a secondary node;
[0019] The updated policy graph structure data is updated according to the secondary node and the target node to obtain the sample policy graph data.
[0020] In some embodiments, the updating of the updated insurance policy graph structure data according to the secondary node and the target node to obtain the sample insurance policy graph data includes:
[0021] If there is a secondary connection edge between the secondary node and the target node, the secondary connection edge is updated according to the preset secondary update weight to obtain an updated connection edge, and the updated insurance policy graph structure data is updated based on the updated connection edge to obtain the sample insurance policy graph data;
[0022] If there is no secondary connection edge between the secondary node and the target node, the secondary node and the target node are connected according to the preset secondary connection edge weight to obtain an updated connection edge, and the updated policy graph structure data is updated based on the updated connection edge to obtain the sample policy graph data.
[0023] In some embodiments, the target anti-fraud prediction model includes an edge weight adjustment layer, a time series capture layer, and a classification layer; performing anti-fraud prediction on the current policy information through the target anti-fraud prediction model and the target policy graph data to obtain anti-fraud prediction information includes:
[0024] The target policy graph data is weighted by the edge weight adjustment layer to obtain weighted updated policy graph data;
[0025] Capturing the time series characteristics of the weight updated insurance policy graph data through the time series capture layer to obtain insurance policy time series characteristic data;
[0026] Anti-fraud prediction is performed on the current policy information through the classification layer and the policy time series feature data to obtain the anti-fraud prediction information.
[0027] In some embodiments, the edge weight adjustment layer includes multiple multi-head self-attention layers and a weight fusion layer; the target policy graph data includes multiple target node edges, and the target policy graph data is weighted by the edge weight adjustment layer to obtain weight-updated policy graph data, including:
[0028] The target node edge is weighted by a plurality of the multi-head self-attention layers to obtain a plurality of candidate weight data;
[0029] Perform weight fusion on the plurality of candidate weight data according to the weight fusion layer to obtain target weight data;
[0030] The target policy diagram data is weight-adjusted according to the target weight data to obtain the weight-updated policy diagram data.
[0031] In some embodiments, the classification layer includes a fully connected layer and an activation layer; performing anti-fraud prediction on the current policy information through the classification layer and the policy time series feature data to obtain the anti-fraud prediction information includes:
[0032] Performing feature aggregation on the policy time series feature data through the fully connected layer to obtain policy aggregate feature data;
[0033] The policy aggregate feature data is classified through the activation layer to obtain the anti-fraud prediction information.
[0034] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application provides an insurance policy anti-fraud detection device, the device comprising:
[0035] A data acquisition module, used to acquire a policy sample and a sample policy label of the policy sample; wherein the sample policy label is used to indicate whether the policy sample is a false policy or not;
[0036] A graph construction module, used to construct a graph based on the sample policy information of the policy sample to obtain sample policy graph data;
[0037] A model training module, used to train a preset anti-fraud prediction model through the sample policy labels and the sample policy graph data to obtain a target anti-fraud prediction model;
[0038] A graph merging module, used to obtain the current policy information of the target object, and perform graph merging based on the sample policy graph data and the current policy information to obtain the target policy graph data;
[0039] The policy prediction module is used to perform anti-fraud prediction on the current policy information through the target anti-fraud prediction model and the target policy map data to obtain anti-fraud prediction information; wherein the anti-fraud prediction information is used to indicate whether the current policy is a false policy or not.
[0040] To achieve the above objectives, a 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 when the processor executes the computer program, the method described in the first aspect is implemented.
[0041] To achieve the above-mentioned purpose, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method described in the first aspect is implemented.
[0042] The present application proposes a policy anti-fraud detection method and device, electronic device and storage medium, which obtains policy samples and sample policy labels, and then constructs a graph based on the sample policy information of the policy samples to obtain sample policy graph data, and then trains a preset anti-fraud model through the sample policy labels and the sample policy graph data, thereby improving the preset anti-fraud model's perception of policy information and improving the accuracy of prediction of whether there is false claims in the policy, thereby obtaining a target anti-fraud model; further, based on, obtaining the current policy information of the target object, and merging the graph based on the sample policy graph data and the current policy information to obtain the target policy graph data, and finally performing anti-fraud prediction on the policy information through the target anti-fraud prediction model and the target policy graph data, thereby accurately predicting whether the current policy information of the target object is a false policy. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 is a flow chart of the insurance policy anti-fraud detection method provided in an embodiment of the present application;
[0044] Figure 2 yes Figure 1 Flow chart of step S102 in FIG.
[0045] Figure 3 yes Figure 2 Flow chart of step S204 in FIG.
[0046] Figure 4 yes Figure 3 Flow chart of step S304 in FIG.
[0047] Figure 5 yes Figure 1 Flow chart of step S105 in FIG.
[0048] Figure 6 yes Figure 5 Flow chart of step S501 in FIG.
[0049] Figure 7 yes Figure 5 Flow chart of step S503 in FIG.
[0050] Figure 8 It is a structural schematic diagram of the insurance policy anti-fraud detection device provided in an embodiment of the present application;
[0051] Fig. 9 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION
[0052] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0053] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.
[0054] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.
[0055] First, some nouns involved in this application are analyzed:
[0056] Graph Data Structure: It is a data model used to model complex relationships between peer entities and is part of discrete mathematics. Graph data structures are mainly used to represent bidirectional or unidirectional relationships between entities. Graph data structures are widely used in various scientific, engineering, and business fields, such as social network analysis, network traffic analysis, routing algorithms, and recommendation systems. Algorithms related to graph data structures include graph creation, traversal, search, shortest path, loop detection, and optimization problems.
[0057] Graph Neural Networks (GNN): A neural network architecture that specifically processes graph-structured data. Graph neural networks use deep learning techniques directly on the nodes and edges of a graph, enabling the model to capture complex relationships and dependencies between nodes. This network structure is particularly suitable for solving problems in the fields of social network analysis, bioinformatics, recommendation systems, knowledge graphs, etc., where data is naturally represented as a graph structure. The core of graph neural networks is to integrate the feature information of locally connected nodes through node representation learning and edge information transmission mechanisms, thereby performing effective feature extraction and prediction.
[0058] Node: It is the basic unit in the graph data structure, also known as vertex. In computer science and mathematics, nodes are used to represent various entities in the network, such as people, places, objects or concepts. In the graph data structure, nodes are connected to each other through edges to form a complex relationship network.
[0059] Edge: An edge is a line segment connecting two nodes in a graph data structure, used to represent the relationship or interaction between nodes. In graph theory in computer science and mathematics, edges can be directed or undirected, representing the directionality or non-directionality of the relationship, respectively. The weight of an edge represents the strength or cost of the connection between nodes. For example, in a traffic network, the weight of an edge can represent the length of a road segment or the difficulty of traveling.
[0060] Policy anti-fraud prediction technology is used to detect whether the policy in the insurance claim involves fraud. For example, in the auto insurance claim scenario, it is necessary to detect whether the policy is false and fraudulent. However, since the auto claim policy contains information of multiple third-party entities, such as the policy holder, assessor, repair location, and repairman, it is difficult to make an accurate judgment by integrating multiple information when identifying whether the auto insurance policy is fraudulent. Therefore, how to accurately identify whether there is a false claim in the auto insurance policy has become an urgent problem to be solved.
[0061] Based on this, the embodiments of the present application provide a policy anti-fraud detection method and device, an electronic device and a storage medium, which are intended to accurately identify whether there are false claims in a motor vehicle insurance policy.
[0062] The embodiments of the present application provide a policy anti-fraud detection method and device, electronic device and storage medium, which are specifically illustrated through the following embodiments. First, the policy anti-fraud detection method in the embodiments of the present application is described.
[0063] 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 obtain the best results.
[0064] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0065] The insurance policy anti-fraud detection method provided in the embodiment of the present application relates to the field of text processing technology. The insurance policy anti-fraud detection method provided in the embodiment of the present application can be applied to a terminal, can be applied to a server side, and can also 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 a server cluster or a distributed system composed of multiple physical servers, or 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 policy anti-fraud detection method, etc., but is not limited to the above forms.
[0066] 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, etc. 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, etc. 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 through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0067] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.
[0068] Figure 1 is an optional flow chart of the insurance policy anti-fraud detection method provided in the embodiment of the present application. Figure 1 The method may include but is not limited to steps S101 to S105.
[0069] Step S101, obtaining a sample insurance policy and a sample insurance policy label of the sample insurance policy; wherein the sample insurance policy label is used to indicate whether the sample insurance policy is a false insurance policy or not;
[0070] Step S102, constructing a graph based on the sample policy information of the policy sample to obtain sample policy graph data;
[0071] Step S103, training the preset anti-fraud prediction model through the sample policy labels and sample policy graph data to obtain a target anti-fraud prediction model;
[0072] Step S104, obtaining the current policy information of the target object, merging the graphs based on the sample policy graph data and the current policy information, and obtaining the target policy graph data;
[0073] Step S105, performing anti-fraud prediction on the current policy information through the target anti-fraud prediction model and the target policy graph data to obtain anti-fraud prediction information; wherein the anti-fraud prediction information is used to indicate whether the current policy is a false policy or not.
[0074] In steps S101 to S105 shown in the embodiment of the present application, by obtaining policy samples and sample policy labels, and then constructing a graph based on the sample policy information of the policy samples to obtain sample policy graph data, and then training the preset anti-fraud model through the sample policy labels and the sample policy graph data, thereby improving the preset anti-fraud model's perception of policy information and improving the accuracy of prediction of whether there is false claims in the policy, thereby obtaining a target anti-fraud model; further, based on, obtaining the current policy information of the target object, and merging the graph based on the sample policy graph data and the current policy information to obtain the target policy graph data, and finally performing anti-fraud prediction on the policy information through the target anti-fraud prediction model and the target policy graph data, thereby accurately predicting whether the current policy information of the target object is a false policy.
[0075] In step S101 of some embodiments, the policy sample refers to a historical auto insurance claim policy, which may be a historical auto insurance claim policy composed of different accidents, different drivers, different appraisers, etc. The policy sample contains a policy label confirmed by manual or computer means, namely, a sample policy label. The sample policy label is used to indicate whether the policy sample is a false policy or not.
[0076] See also Figure 2 In some embodiments, the sample policy information includes the policy number, driver information, accident report, assessor information, and beneficiary information. Step S102 may include, but is not limited to, steps S201 to S204:
[0077] Step S201, filtering the driver information according to the insurance policy number to obtain the insurance policy driver related information;
[0078] Step S202, the insurance policy number, the driver information and the insurance policy driver association information are node-connected to obtain initial insurance policy graph structure data; the initial insurance policy graph structure data includes at least two main nodes and main node edges between any two main nodes;
[0079] Step S203, updating the weight of each main node edge according to the preset main node edge weight, and obtaining updated insurance policy graph structure data;
[0080] Step S204, randomly connect nodes based on the updated insurance policy graph structure data, accident report, assessor information and beneficiary information to obtain sample insurance policy graph data.
[0081] In steps S201 to S204 shown in the embodiment of the present application, the driver information is screened according to the policy number to obtain the policy-driver associated information, and then the policy number, driver information and the policy-driver associated information are node-connected to obtain initial policy graph structure data; the initial policy graph structure data includes at least two main nodes and main node edges between any two main nodes, and then the weight of each main node edge is updated according to the preset main node edge weight to obtain updated policy graph structure data, and finally random node connections are made based on the updated policy graph structure data, accident report, assessor information and beneficiary information to obtain sample policy graph data, thereby integrating various information to construct a graph data structure, providing a comprehensive data foundation for subsequent anti-fraud prediction based on the graph data structure, and improving the prediction accuracy for subsequent anti-fraud prediction.
[0082] In step S201 of some embodiments, the policy number is the serial number of the claim policy, and each claim policy corresponds to a unique policy number. The claim policy includes driver information corresponding to the policy number. The driver information is screened by the policy number to obtain the driver information corresponding to each policy number, that is, the policy-driver association information. For example, the policy number includes the first policy and the second policy. The drivers corresponding to the first policy are the first driver and the second driver, and the drivers corresponding to the second policy are the first driver and the third driver. Then the policy-driver association information is (first policy, first driver, second driver), (second policy, first driver, third driver).
[0083] In step S202 of some embodiments, the policy number and driver information are connected by nodes according to the policy-driver association information. For example, if the policy-driver association information is (first policy, first driver, second driver), (second policy, first driver, third driver), then the first policy, the second policy, the first driver, the second driver and the third driver are all master nodes, there is a master node edge between the first policy and the first driver, there is a master node edge between the first policy and the second driver, there is a master node edge between the second policy and the first driver, and there is a master node edge between the second policy and the third driver.
[0084] In step S203 of some embodiments, in the graph data structure, the edge weight represents the strength or cost of the connection between nodes. In the initial insurance policy graph structure data, the edge weight represents the importance from node to node. The preset master node edge weight is a value greater than 0. The preset master node edge weight is used to assign a weight to each master node edge to obtain the updated order graph structure data. It should be noted that the preset master node edge weight can be set according to the actual scenario.
[0085] See also Figure 3In some embodiments, step S204 may include but is not limited to steps S301 to S304:
[0086] Step S301, randomly selecting a main node according to the updated policy graph structure data to obtain a target node;
[0087] Step S302, performing association screening on the accident report, assessor information, and beneficiary information according to the target node to obtain at least one candidate node;
[0088] Step S303, randomly selecting a node from the candidate nodes to obtain a secondary node;
[0089] Step S304, update the insurance policy graph structure data according to the secondary node and the target node to obtain sample insurance policy graph data.
[0090] Steps S301 to S304 shown in the embodiment of the present application randomly select the main node according to the sample policy graph data to obtain the target node, then perform association screening on the accident report, assessor information and beneficiary information according to the target node to obtain at least one candidate node, and then randomly select nodes from the candidate nodes to obtain sub-nodes, and finally update the updated policy graph structure data according to the sub-nodes and the target nodes to obtain the sample policy graph data, thereby randomly selecting from the accident report, assessor information and benefit objects based on the random walk model, and selecting sub-nodes with higher frequency of occurrence in terms of probability, thereby avoiding excessive computing power requirements due to too large a sample and focusing on nodes with higher probability of occurrence, ignoring secondary information, thereby improving the prediction accuracy of anti-fraud;
[0091] In step S301 of some embodiments, a main node is randomly selected from the main nodes in the updated policy graph structure data as the target node.
[0092] In step S302 of some embodiments, at least one candidate node is obtained by performing association screening from the accident report, assessor information and beneficiary according to the target node. For example, the target node is the first driver, the accident report includes the first report, the second report and the third report, the assessor information includes the first assessor and the second assessor, the beneficiary information includes the first beneficiary and the second beneficiary, the accident report corresponding to the first driver includes the first report and the second report, the assessor information includes the first assessor and the second assessor, and the beneficiary includes the first beneficiary, then the candidate nodes include all nodes corresponding to the first driver, namely the first report, the second report, the first assessor, the second assessor and the first beneficiary.
[0093] In step S303 of some embodiments, a node is randomly selected from the candidate nodes as the backup node. For example, the candidate nodes include all nodes corresponding to the first driver, namely the first report, the second report, the first assessor, the second assessor and the first beneficiary. The randomly selected node is the first beneficiary, and the first beneficiary serves as the secondary node.
[0094] See also Figure 4 In some embodiments, step S304 may include but is not limited to steps S401 to S403:
[0095] Step S401: if there is a secondary connection edge between the secondary node and the target node, the secondary connection edge is updated according to the preset secondary update weight to obtain an updated connection edge, and the updated insurance policy graph structure data is updated based on the updated connection edge to obtain sample insurance policy graph data;
[0096] Step S402, if there is no secondary connection edge between the secondary node and the target node, the secondary node and the target node are connected according to the preset secondary connection edge weight to obtain the updated connection edge, and the updated policy graph structure data is updated based on the updated connection edge to obtain the sample policy graph data.
[0097] In steps S401 to S402 shown in the embodiment of the present application, when there is a secondary connection edge between the secondary node and the target node, the secondary connection edge is updated according to the preset secondary update weight to obtain the updated connection edge, and the updated policy graph structure data is updated based on the updated connection edge to obtain the sample policy graph data; when there is no secondary connection edge between the secondary node and the target node, the secondary node and the target node are edge-connected according to the preset secondary connection edge weight to obtain the updated connection edge, and the updated policy graph structure data is updated based on the updated connection edge to obtain the sample policy graph data, thereby confirming the edge weight, so that the edge weight between frequently appearing nodes is greater than the edge weight between infrequently appearing nodes, thereby providing a data basis for the edge weight for the subsequent anti-fraud model.
[0098] In step S401 of some embodiments, if there is a secondary connection edge between the secondary node and the target node, it means that in the process of establishing the graph data structure, the secondary node has established a connection edge in the previous cycle. When the secondary node is selected again, the secondary connection edge is updated based on the preset secondary update weight, and the weight of the secondary connection edge is increased, and then the updated graph data structure is obtained, that is, the sample policy graph data.
[0099] In step S402 of some embodiments, if there is no connecting edge between the secondary node and the target node, it means that the secondary node is selected for the first time in the previous cycle, and the secondary node and the target node are directly connected based on the preset secondary connection edge weight to obtain an updated graph data structure, i.e., the sample policy graph data.
[0100] It should be noted that the establishment of the sample insurance policy graph data is a cyclic iterative process. At the beginning of the loop, the graph structure data only includes the edges between the main nodes. During the loop, a node in the graph data structure is randomly selected as the current node, and the entity information of the current node is taken out. For example, the current node is the beneficiary information, and one is randomly selected as the next node based on the policy number, driver information, assessor information and accident report associated with the beneficiary information, and the edge weight between the next node and the current node is confirmed. The cycle is repeated until the loop condition is met, thereby determining the edge weights between each main node, each secondary node and adjacent nodes in the graph data structure, that is, determining the graph data structure based on the random walk model to obtain the sample insurance policy graph data.
[0101] In step S103 of some embodiments, the sample policy image data is input into a preset anti-fraud model for anti-fraud prediction, and it is determined whether each sample policy is a false policy to obtain a predicted policy label. Then, a mean square error calculation is performed based on the predicted policy label and the sample policy label to obtain loss value data. Then, based on the loss value data, the preset anti-fraud prediction model is optimized in parameters to obtain a target anti-fraud prediction model.
[0102] In step S104 of some embodiments, the current policy information is a policy that needs to be predicted for anti-fraud, and a graph is merged based on the sample policy graph data and the current policy information. The graph is merged in a manner of obtaining the current policy number, the current driver information, the current accident report, the current assessor information, and the current beneficiary from the current policy information, and based on the current policy number, the current driver information, the current accident report, the current assessor information, and the current beneficiary as the current node, if there is already a node identical to the current node in the sample policy graph data, the node is merged, and if there is no node identical to the current node, a new node is created in the sample policy graph data to obtain the initial policy graph data. Then, according to the connection relationship between the current nodes, the initial policy graph data is weighted, specifically, if a new connection edge is created due to the connection relationship between the current nodes, the edge weight of the new connection edge is used according to the preset current weight, and if the existing connection edge in the initial policy graph data needs to be updated due to the connection relationship between the current nodes, the existing connection edge that needs to be updated is updated according to the preset current weight to obtain the target policy graph data. For example: the sample policy graph data includes the first node, the second node and the third node. There is a connection edge between the first node and the second node with a weight of 1, and there is a connection edge between the second node and the third node with a weight of 1. According to the current policy information, the nodes are converted to obtain the second node, the third node and the fourth node. The second node and the third node are connected, and the third node and the fourth node are connected. Then the nodes are merged to obtain the initial policy graph data, which includes the first node, the second node, the third node and the fourth node. Then the weight is updated. There is a connection edge between the first node and the second node with a weight of 1, there is a connection edge between the second node and the third node with a weight of 2, there is a connection board between the third node and the fourth node with a weight of 1, and the target policy graph data is obtained.
[0103] See also Figure 5 In some embodiments, the target anti-fraud prediction model includes an edge weight adjustment layer, a time series capture layer, and a classification layer. Step S105 includes but is not limited to steps S501 to S503:
[0104] Step S501, weighting the target policy graph data through the edge weight adjustment layer to obtain weighted updated policy graph data;
[0105] Step S502, capturing the time series features of the weight update policy graph data through the time series capture layer to obtain the policy time series feature data;
[0106] Step S503, performing anti-fraud prediction on the current policy information through the classification layer and the policy time series feature data to obtain anti-fraud prediction information.
[0107] In steps S501 to S503 shown in the embodiment of the present application, the target policy graph data is weighted adjusted through the edge weight adjustment layer, which effectively re-evaluates and strengthens the relationship strength between nodes, so that the model can better understand and utilize the mutual influence between various entities, thereby improving the ability to comprehensively analyze multi-party information, and obtaining weight-updated policy graph data, and then the time series feature of the weight-updated policy graph data is captured through the time series capture layer to extract time-related behavior patterns and trends, which helps to explain potential fraud from a time perspective and obtain policy time series feature data, and then the current policy information is subjected to anti-fraud prediction through the classification layer and the policy time series feature data to obtain anti-fraud prediction information, thereby accurately identifying whether there is false claim in the motor vehicle insurance policy.
[0108] See also Figure 6 In some embodiments, the edge weight adjustment layer includes multiple multi-head self-attention layers, a weight fusion layer, the target policy graph data includes multiple target node edges, and step S501 includes but is not limited to steps S601 to S603:
[0109] Step S601, adjusting the weight of the target node edge through multiple multi-head self-attention layers to obtain multiple candidate weight data;
[0110] Step S602, weight fusion is performed on a plurality of candidate weight data according to a weight fusion layer to obtain target weight data;
[0111] Step S603, weight adjustment is performed on the target insurance policy diagram data according to the target weight data to obtain weight-updated insurance policy diagram data.
[0112] Steps S601 to S603 shown in the embodiment of the present application can finely adjust the relationship weights between each node through a multi-head self-attention layer to strengthen the relationship features related to anti-fraud prediction; further, the weight fusion layer integrates the alternative weight data to form more comprehensive target weight data, thereby reducing information loss and ensuring the effective capture of key signals, so that the target anti-fraud model can more accurately identify potential fraudulent behaviors, and finally adjust the weights of the target policy graph data according to the target weight data to obtain weight-updated policy graph data, so that the weights in the weighted policy graph data can enable the target anti-fraud model to make anti-fraud predictions more accurately.
[0113] In step S601 of some embodiments, the edge weight adjustment layer is the GAT Layer in the Hierarchical Attention Network, and the GAT Layer includes multiple multi-head self-attention layers. The weight of the target node edge is input into the multiple multi-head self-attention layers, and each multi-head self-attention layer adjusts the weight of the target node edge and outputs alternative weight data, that is, multiple self-attention layers adjust the weight of the target node edge and output multiple alternative weight data.
[0114] In step S602 of some embodiments, the candidate weight data are weighted averaged to obtain target weight data.
[0115] In step S603 of some embodiments, weights corresponding to the target insurance policy diagram data are assigned based on the target weight data to obtain weight-updated insurance policy diagram data.
[0116] In step S502 of some embodiments, the timing capture layer is an LSTM layer, which extracts the timing features in the weight update order graph data through the timing capture layer, assigns timing-level features on top of the graph data structure, and obtains policy timing feature data, thereby improving the accuracy of anti-fraud prediction by the target anti-fraud model.
[0117] See also Figure 7 In some embodiments, the classification layer includes a fully connected layer and an activation layer, and step S503 may include but is not limited to steps S701 to S702:
[0118] Step S701, performing feature aggregation on the policy time series feature data through a fully connected layer to obtain policy aggregate feature data;
[0119] Step S702, classify the policy aggregate feature data through the activation layer to obtain anti-fraud prediction information.
[0120] In steps S701 to S702 shown in the embodiment of the present application, feature aggregation is performed on the policy time series feature data through a fully connected layer to obtain policy aggregated feature data, and then the policy aggregated feature data is classified through an activation layer to output final anti-fraud prediction information, thereby completing the anti-fraud prediction of the current policy information of the target object.
[0121] In step S701 of some embodiments, the fully connected layer includes at least one linear layer, the input and output of the linear layer are connected in sequence, and the policy time series feature data is input into the fully connected layer to obtain the policy aggregate feature data.
[0122] In step S702 of some embodiments, the activation layer is a Sigmoid function, and the policy aggregation feature data is input into the activation layer to obtain the final classification, that is, whether the current policy information of the target object belongs to a false policy or a real policy.
[0123] See also Figure 8 The embodiment of the present application also provides an insurance policy anti-fraud detection device, which can implement the above-mentioned insurance policy anti-fraud detection method, and the device includes:
[0124] The data acquisition module 801 is used to acquire a sample insurance policy and a sample insurance policy label of the sample insurance policy; wherein the sample insurance policy label is used to indicate whether the sample insurance policy is a false insurance policy or not;
[0125] A graph construction module 802, for constructing a graph based on the sample policy information of the policy sample to obtain sample policy graph data;
[0126] Model training module 803, used to train the preset anti-fraud prediction model through sample policy labels and sample policy graph data to obtain a target anti-fraud prediction model;
[0127] The graph merging module 804 is used to obtain the current policy information of the target object, and merge the graphs based on the sample policy graph data and the current policy information to obtain the target policy graph data;
[0128] The policy prediction module 805 is used to perform anti-fraud prediction on the current policy information through the target anti-fraud prediction model and the target policy graph data to obtain anti-fraud prediction information; wherein the anti-fraud prediction information is used to indicate whether the current policy is a false policy or not.
[0129] The specific implementation of the insurance policy anti-fraud detection device is basically the same as the specific implementation of the above-mentioned insurance policy anti-fraud detection method, and will not be repeated here.
[0130] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned insurance policy anti-fraud detection method when executing the computer program. The electronic device can be any smart terminal including a tablet computer, a car computer, etc.
[0131] See also Fig. 9 , Fig. 9 The hardware structure of an electronic device of another embodiment is illustrated, and the electronic device includes:
[0132] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (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;
[0133] The memory 902 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 902 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the insurance policy anti-fraud detection method of the embodiment of this application;
[0134] Input / output interface 903, used to implement information input and output;
[0135] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WI FI, Bluetooth, etc.);
[0136] A bus 905 that transmits information between various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);
[0137] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .
[0138] 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, the above-mentioned insurance policy anti-fraud detection method is implemented.
[0139] 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 disposed 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.
[0140] The policy anti-fraud detection method, policy anti-fraud detection device, electronic device and storage medium provided in the embodiments of the present application obtain policy samples and sample policy labels, and then construct a graph based on the sample policy information of the policy samples to obtain sample policy graph data, and then train a preset anti-fraud model through the sample policy labels and the sample policy graph data, thereby improving the preset anti-fraud model's perception of policy information and improving the accuracy of prediction of whether there is false claims in the policy, thereby obtaining a target anti-fraud model; further, based on, obtaining the current policy information of the target object, and merging the graph based on the sample policy graph data and the current policy information to obtain the target policy graph data, and finally performing anti-fraud prediction on the policy information through the target anti-fraud prediction model and the target policy graph data, thereby accurately predicting whether the current policy information of the target object is a false policy.
[0141] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present 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 the present application are also applicable to similar technical problems.
[0142] Those skilled in the art will appreciate 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.
[0143] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0144] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.
[0145] 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 sequence. 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 of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising 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.
[0146] It should be understood that in the present 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 objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single 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.
[0147] In the several embodiments provided in the present 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 only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, 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.
[0148] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0149] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.
[0150] 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 is essentially 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, and the computer software product is stored in a storage medium, including multiple instructions to enable 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: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.
[0151] The preferred embodiments of the present application are described above with reference to the accompanying drawings, but the scope of the rights of the present application is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present application should be within the scope of the rights of the present application.
Claims
1. A method for detecting insurance policy fraud, characterized in that: The method comprises: Obtaining a policy sample and a sample policy label of the policy sample; wherein the sample policy label is used to indicate whether the policy sample is a false policy or not; Constructing a graph based on the sample policy information of the policy sample to obtain sample policy graph data; The preset anti-fraud prediction model is trained by using the sample policy labels and the sample policy graph data to obtain a target anti-fraud prediction model; Acquire the current insurance policy information of the target object, and merge the graphs based on the sample insurance policy graph data and the current insurance policy information to obtain the target insurance policy graph data; Anti-fraud prediction is performed on the current policy information through the target anti-fraud prediction model and the target policy map data to obtain anti-fraud prediction information; wherein the anti-fraud prediction information is used to indicate whether the current policy is a false policy or not.
2. The method according to claim 1, characterized in that The sample policy information includes the policy number, driver information, accident report, assessor information and beneficiary information; the sample policy information based on the policy sample is graph constructed to obtain sample policy graph data, including: Filter the driver information according to the insurance policy number to obtain insurance policy driver related information; Perform node connection on the insurance policy number, the driver information and the insurance policy driver association information to obtain initial insurance policy graph structure data; the initial insurance policy graph structure data includes at least two main nodes and main node edges between any two of the main nodes; The weight of each of the main node edges is updated according to the preset main node edge weight to obtain updated insurance policy graph structure data; The sample policy graph data is obtained by performing random node connections based on the updated policy graph structure data, the accident report, the assessor information and the beneficiary information.
3. The method according to claim 2, characterized in that The step of performing random node connection based on the updated insurance policy graph structure data, the accident report, the assessor information, and the beneficiary information to obtain the sample insurance policy graph data includes: Randomly select a master node according to the updated policy graph structure data to obtain a target node; According to the target node, the accident report, the assessor information and the beneficiary information are associated and screened to obtain at least one candidate node; Randomly select a node from the candidate nodes to obtain a secondary node; The updated policy graph structure data is updated according to the secondary node and the target node to obtain the sample policy graph data.
4. The method according to claim 3, characterized in that The updating of the updated insurance policy graph structure data according to the secondary node and the target node to obtain the sample insurance policy graph data includes: If there is a secondary connection edge between the secondary node and the target node, the secondary connection edge is updated according to the preset secondary update weight to obtain an updated connection edge, and the updated insurance policy graph structure data is updated based on the updated connection edge to obtain the sample insurance policy graph data; If there is no secondary connection edge between the secondary node and the target node, the secondary node and the target node are connected according to the preset secondary connection edge weight to obtain an updated connection edge, and the updated policy graph structure data is updated based on the updated connection edge to obtain the sample policy graph data.
5. The method according to any one of claims 1 to 4, characterized in that: The target anti-fraud prediction model includes an edge weight adjustment layer, a time series capture layer and a classification layer; the anti-fraud prediction is performed on the current policy information through the target anti-fraud prediction model and the target policy graph data to obtain anti-fraud prediction information, including: The target policy graph data is weighted by the edge weight adjustment layer to obtain weighted updated policy graph data; Capturing the time series characteristics of the weight updated insurance policy graph data through the time series capture layer to obtain insurance policy time series characteristic data; Anti-fraud prediction is performed on the current policy information through the classification layer and the policy time series feature data to obtain the anti-fraud prediction information.
6. The method according to claim 5, characterized in that The edge weight adjustment layer includes multiple multi-head self-attention layers and a weight fusion layer; the target policy graph data includes multiple target node edges, and the target policy graph data is weighted by the edge weight adjustment layer to obtain weight-updated policy graph data, including: The target node edge is weighted by a plurality of the multi-head self-attention layers to obtain a plurality of candidate weight data; Perform weight fusion on the plurality of candidate weight data according to the weight fusion layer to obtain target weight data; The target policy diagram data is weight-adjusted according to the target weight data to obtain the weight-updated policy diagram data.
7. The method according to claim 5, characterized in that The classification layer includes a fully connected layer and an activation layer; the anti-fraud prediction is performed on the current policy information through the classification layer and the policy time series feature data to obtain the anti-fraud prediction information, including: Performing feature aggregation on the policy time series feature data through the fully connected layer to obtain policy aggregate feature data; The policy aggregate feature data is classified through the activation layer to obtain the anti-fraud prediction information.
8. An insurance policy anti-fraud detection device, characterized in that: The device comprises: A data acquisition module, used to acquire a policy sample and a sample policy label of the policy sample; wherein the sample policy label is used to indicate whether the policy sample is a false policy or not; A graph construction module, used to construct a graph based on the sample policy information of the policy sample to obtain sample policy graph data; A model training module, used to train a preset anti-fraud prediction model through the sample policy labels and the sample policy graph data to obtain a target anti-fraud prediction model; A graph merging module, used to obtain the current policy information of the target object, and perform graph merging based on the sample policy graph data and the current policy information to obtain the target policy graph data; The policy prediction module is used to perform anti-fraud prediction on the current policy information through the target anti-fraud prediction model and the target policy map data to obtain anti-fraud prediction information; wherein the anti-fraud prediction information is used to indicate whether the current policy is a false policy or not.
9. 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 insurance policy anti-fraud detection method described in any one of claims 1 to 7 when executing the computer program.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the insurance policy anti-fraud detection method described in any one of claims 1 to 7 is implemented.