Car insurance claim settlement intelligent supervision method and system based on big data
By constructing a claims relationship graph and a trust score update mechanism, the problem of low technical efficiency in traditional auto insurance claims supervision has been solved, and intelligent management and risk identification of the auto insurance claims process have been achieved.
Patent Information
- Application Number
- CN202511499095.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-20
- Publication Date
- 2026-01-20
AI Technical Summary
Traditional auto insurance claims monitoring technology is inefficient, lacks flexibility, cannot effectively identify fraudulent activities in complex and ever-changing business scenarios, and fails to fully utilize the network structure information in the claims ecosystem.
Construct a claims relationship graph, generate claims feature vectors and vehicle feature vectors, dynamically monitor the behavior of participants through overall deviation, trust score and volatility index, and conduct intelligent supervision by combining basic trust score and propagation trust score.
It enables precise identification of high-risk nodes in the auto insurance claims process and optimizes the claims process at low-risk nodes, thereby improving risk management efficiency and business management level, and reducing fraud risk.
Smart Images
Figure CN121366045A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of knowledge graph. More particularly, the present application relates to a big data-based intelligent supervision method and system for vehicle insurance claims. BACKGROUND
[0002] Vehicle insurance claims are an important part of the insurance industry, involving multiple parties such as vehicle owners, loss assessors and repair shops. With the continuous expansion of vehicle insurance business, fraud and abnormal operations in the claims process have gradually become one of the main challenges faced by insurance companies. Traditional vehicle insurance claims supervision mainly relies on manual review and fixed rule setting, such as setting a maximum claim amount or reviewing specific types of cases to identify potential risks. However, traditional vehicle insurance claims supervision techniques have deficiencies in efficiency, flexibility and comprehensiveness.
[0003] Firstly, manual review is inefficient and difficult to cope with the growth of massive claims. When faced with a large amount of data, manual review not only consumes time and effort, but also is prone to misjudgment or omission due to subjective judgment. Secondly, fixed rule setting lacks flexibility and cannot adapt to complex and changing business scenarios.
[0004] In addition, traditional methods pay insufficient attention to the interaction between parties and fail to fully utilize the network structure information in the claims ecosystem. For example, a loss assessor frequently cooperates with multiple high-risk repair shops, which may imply potential joint fraud, but traditional methods usually cannot effectively capture such potential joint fraud information. At the same time, existing technologies also have shortcomings in dynamic monitoring, which cannot timely discover and respond to changes in behavior patterns of parties, which may lead to risks accumulating to an uncontrollable level.
[0005] Therefore, there is an urgent need for a big data-based intelligent supervision method for vehicle insurance claims. SUMMARY
[0006] To solve the technical problems of the deficiencies of traditional vehicle insurance claims supervision techniques in efficiency, flexibility and comprehensiveness, the present application provides solutions in the following aspects.
[0007] In a first aspect, the present application provides a big data-based intelligent supervision method for vehicle insurance claims, comprising: constructing a claim settlement relationship graph, generating a claim settlement feature vector corresponding to each historical claim settlement case of each node in the claim settlement relationship graph and a vehicle feature vector; classifying the historical claim settlement cases of each node according to the vehicle feature vector, and constructing a historical behavior inertia vector of each category according to the claim settlement feature vectors of the historical claim settlement cases in each category; taking each node corresponding to the current claim settlement case as a participant node, determining the overall deviation of the current claim settlement case to the participant node according to the claim settlement feature vector of the current claim settlement case, the vehicle feature vector and the historical behavior inertia vector of each category of the participant node; determining the volatility index of the participant node according to the overall deviation and the trust score of the participant node; for any node, determining the basic trust score of the node according to the trust score of the node before the occurrence of the current claim settlement case and the volatility indexes of each participant node, determining the propagation trust score of all nodes pointing to the node to the node according to the basic trust scores of the nodes pointing to the node, and updating the trust score of the node according to the basic trust score and the propagation trust score; and performing intelligent supervision of vehicle insurance claim settlement according to the updated trust scores of the nodes.
[0008] The application can accurately capture behavior patterns in different scenarios by classifying historical claim settlement cases and modeling feature vectors, and can effectively identify abnormal behaviors by quantifying the overall deviation based on the comparison between the feature vector of the current claim settlement case and the historical behavior inertia vector. The application adjusts the trust score by introducing the volatility index, comprehensively considers the behavior change of the node itself and its historical trust level, and avoids misjudgment caused by simply relying on behavior deviation or historical records. In the trust score updating process, the basic trust score and the propagation trust score are combined, and the direct influence of the node's own behavior and the indirect influence of the neighbor node's behavior are dynamically balanced through the damping coefficient, ensuring the comprehensiveness and rationality of the evaluation results. The application implements intelligent supervision based on the updated trust scores, which not only accurately locates high-risk nodes and takes targeted control measures, but also optimizes the claim settlement process of low-risk nodes to improve efficiency, thereby reducing fraud risk while improving overall business management level. The application enhances the intelligence of the vehicle insurance claim settlement system and improves the adaptive ability and risk management effect of the vehicle insurance claim settlement system.
[0009] Preferably, the generating of the claim settlement feature vector corresponding to each historical claim settlement case of each node in the claim settlement relationship graph and the vehicle feature vector comprises: the claim settlement feature vector contains claim settlement amount, repair project complexity, high-priced accessory proportion, claim settlement processing time length and loss assessment deviation rate; and the vehicle feature vector contains vehicle type and accident type.
[0010] Preferably, the method comprises: classifying historical claim cases of each node according to the vehicle feature vector; constructing a historical behavior inertia vector of each category according to the claim feature vectors of historical claim cases in each category, comprising: for any node, classifying historical claim cases with the same vehicle feature vector into the same category; determining a focus weight of historical claim cases in the category according to the difference between the current time and the occurrence time of the historical claim cases, wherein the focus weight is negatively correlated with the difference; and performing weighted summation on the claim feature vectors of all historical claim cases in the category according to the focus weight of the historical claim cases in the category to obtain the historical behavior inertia vector of the corresponding category.
[0011] The application classifies historical claim cases with the same vehicle feature vector into the same category, can effectively distinguish the behavior characteristics in different scenarios, and avoids the problem of ambiguous behavior patterns caused by mixed categories. The application introduces the difference between the current time and the occurrence time of historical claim cases to determine the focus weight, and makes the focus weight negatively correlated with the time difference, so as to ensure that the historical behavior inertia vector can reflect both long-term trends and recent performance, thereby dynamically adapting to the changes of node behavior. The application generates the historical behavior inertia vector of the corresponding category by performing weighted summation on the claim feature vectors of all historical claim cases in the category, and further enhances the quantification ability of the model to the behavior patterns of the node in a specific scenario.
[0012] Preferably, the method for obtaining the overall deviation degree comprises: determining a time deviation degree of the current claim case relative to each category of the participant node; taking the Hamming distance between the vehicle feature vector of the current claim case and the vehicle feature vector corresponding to the historical claim case in each category of the participant node as a category deviation degree of the current claim case relative to each category of the participant node; determining a behavior deviation degree of the current claim case relative to each category of the participant node according to the cosine similarity between the claim feature vector of the current claim case and the historical behavior inertia vector of each category of the participant node; determining a reference weight of each category according to the category deviation degree and the proportion of historical claim cases in each category; and determining the overall deviation degree of the current claim case relative to the participant node according to the reference weight and the time deviation degree and the behavior deviation degree.
[0013] In this invention, time deviation reflects the impact of the time interval between the current claim case and the historical behavior of a node, effectively capturing the potential risk when a similar case suddenly appears after a long period of inactivity. This invention calculates behavioral deviation based on the cosine similarity between the claim feature vector and the historical behavior inertia vector, objectively measuring the consistency between the current claim case and the historical behavior pattern of a node, and capturing the changing trends of behavioral characteristics. This invention combines category deviation and the proportion of historical claim cases to determine reference weights, ensuring a reasonable contribution ratio of each category to the overall deviation, highlighting the impact of major business scenarios while also considering the special characteristics of rare scenarios. By comprehensively calculating the overall deviation using time deviation, behavioral deviation, and reference weights, this invention achieves a comprehensive assessment of the abnormality of the current claim case, providing a scientific and reliable basis for the subsequent generation of volatility indices and dynamic adjustment of trust scores, thereby effectively improving the accuracy and effectiveness of intelligent supervision of auto insurance claims.
[0014] Preferably, the method for obtaining the reference weight is as follows: for any category of the participating node, the category deviation of the current claim case relative to that category of the participating node is negatively correlated and normalized; the reference weight of the category is obtained by multiplying the result of the negative correlation normalization with the proportion of historical claim cases of the category.
[0015] This invention comprehensively considers the representativeness of the category and the matching degree of current claims cases, ensuring that the reference weight can reflect the main business model of the node in the category, and will not be excessively affected by rare categories. It avoids the bias caused by simply relying on the number of historical claims cases or the matching degree of categories, making the contribution ratio of each category to the overall deviation more balanced and closer to the actual business scenario.
[0016] Preferably, the overall deviation satisfies the expression: In the formula, This indicates the overall deviation of the current claim case from the participating party's node; The first node of the participating node Reference weights for each category; This indicates the current claim case relative to the node of the participating parties. Deviation of behavior in each category; This indicates the current claim case's status regarding the participating party's node. Time deviation of each category; This indicates the number of categories corresponding to the participating nodes.
[0017] This invention introduces reference weights to ensure that the contribution ratio of each category to the overall deviation is commensurate with its representativeness and matching degree, avoiding the excessive influence of a single category on the results. This invention combines behavioral deviation with temporal deviation, dynamically adjusting the impact of behavioral deviation: when the temporal deviation is large (i.e., historical behavior is long ago), the effect of behavioral deviation is amplified, thus more sensitively capturing potential risks; conversely, when the temporal deviation is small, the impact of behavioral deviation is relatively weakened, avoiding misjudgment due to the neglect of recent normal behavior. This invention performs a weighted summation of all categories, comprehensively considering the behavioral characteristics of nodes in different scenarios and their changing trends over time, enabling the overall deviation to fully reflect the true degree of abnormality of the current claim case relative to the participating node, providing a more reliable foundation for subsequent trust score adjustments and intelligent supervision.
[0018] Preferably, determining the volatility index of the participating nodes includes: In the formula, This represents the volatility index of the participating nodes; This indicates the overall deviation of the current claim case from the participating party's node; This indicates the trust score of the participating nodes before the current claim occurred; This represents the trust sensitivity coefficient.
[0019] In this invention, the overall deviation reflects the specific differences between the current claim case and the behavioral patterns of participating nodes, accurately capturing abnormal behavioral characteristics. The trust score reflects the historical credibility of a node before the current claim case occurred. This invention uses the node's historical trust level as a regulating factor to dynamically suppress or amplify the impact of the overall deviation, achieving differentiated monitoring that is "lenient towards those with high credibility and strict towards those with low credibility." This effectively balances the relationship between abnormal behavior and historical trust, avoiding misjudgments or omissions caused by a single indicator, thus providing a more accurate risk quantification basis for subsequent trust score updates and intelligent supervision.
[0020] Preferably, the basic trust score satisfies the expression: ,in, Represents a node Basic trust score; This indicates the node before the current claim case occurred. Trust score; This represents the set of participating nodes in the current claims case. Indicates when node When a node is a participant in the current claims case, the node... The volatility index; the propagation trust score satisfies the expression: , denotes the propagation trust score of all nodes pointing to node denotes the set of all nodes pointing to node in the claim relationship graph; denotes the base trust score of node denotes the out-degree of node is a minimum function.
[0021] Preferably, the updating of the trust score of the node comprises: weighted sum of the base trust score of the node and the propagation trust score of all nodes pointing to the node to the node to obtain the updated trust score of the node.
[0022] The base trust score in the application reflects the behavior characteristics of the node itself and its performance in the current claim case, and can accurately capture the abnormal or normal behavior of the node when directly participating in the case. The propagation trust score reflects the trust level of all neighbor nodes pointing to the node, and embodies the quantitative result of the indirect influence of other nodes in the claim relationship graph on the node. The trust score of the node is updated by comprehensively considering the base trust score and the propagation trust score, which not only retains the characteristics of the node itself, but also fully considers the interaction mode and overall behavior trend of the node in the claim relationship graph, thereby avoiding the one-sidedness caused by simply relying on direct behavior or indirect propagation. The finally generated updated trust score can more accurately reflect the real credibility of the node, provide a reliable decision basis for subsequent intelligent supervision of vehicle insurance claims, effectively reduce potential risks and improve overall business management efficiency.
[0023] In a second aspect, the application provides a vehicle insurance claim intelligent supervision system based on big data, comprising a processor and a memory, and the memory stores computer program instructions. When the computer program instructions are executed by the processor, the above-mentioned vehicle insurance claim intelligent supervision method based on big data is realized.
[0024] By adopting the above technical scheme, the above-mentioned vehicle insurance claim intelligent supervision method based on big data is generated into a computer program and stored in the memory to be loaded and executed by the processor, so as to manufacture a terminal device according to the memory and the processor, and facilitate use.
[0025] The application has the following advantages: (1) The application realizes fine evaluation and dynamic supervision of the behaviors of each participant in the vehicle insurance claim process by constructing a claim relationship graph and combining multi-dimensional data analysis, which significantly improves the scientificity, accuracy and efficiency of risk identification and management.
[0026] (2) The application accurately describes the behavior mode of the node in different scenarios by classifying historical claim cases and constructing historical behavior inertia vectors of each category, avoiding the problem of blurred behavior characteristics caused by mixed categories or single global evaluation. At the same time, the application introduces a time decay mechanism to calculate the attention weight, so that the historical behavior inertia vector can not only reflect the long-term trend, but also focus on the recent performance, thereby dynamically adapting to the changes of node behavior.
[0027] (3) The application calculates the overall deviation of the current claim case relative to the participant node by comprehensively considering the time deviation, category deviation and behavior deviation, comprehensively reflects the abnormal degree of the case in the time, category and behavior three dimensions, effectively captures the potential risk signal, and provides a reliable quantitative basis for the generation of subsequent volatility index.
[0028] (4) The application determines the volatility index in combination with the overall deviation and the historical trust score of the node, realizes the differentiated monitoring of "tolerance to high credit, strictness to low credit".
[0029] (5) The application updates the trust score of the node by comprehensively considering the basic trust score and the propagation trust score, retains the characteristics of the node itself, and also considers the interaction mode and overall behavior trend of the node in the claim relationship graph, improves the comprehensiveness and accuracy of the trust score evaluation result.
[0030] (6) The application implements intelligent supervision based on the updated trust score, takes targeted measures according to different risk levels, can reduce the fraud risk, improve the speed of claim processing and customer satisfaction, and realizes the dual goals of risk control and business efficiency. BRIEF DESCRIPTION OF DRAWINGS
[0031] Figure 1 is a flowchart schematically showing a vehicle insurance claim intelligent supervision method based on big data in the application. DETAILED DESCRIPTION
[0032] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the application.
[0033] The specific embodiments of the application will be described in detail below with reference to the drawings.
[0034] The embodiments of the application disclose a vehicle insurance claim intelligent supervision method based on big data, referring to Figure 1 , comprising steps S1-S6: S1, construct a claim settlement relationship graph, generate a claim settlement feature vector corresponding to each historical claim case of each node in the claim settlement relationship graph and a vehicle feature vector.
[0035] Specifically, the insurance company's claim settlement database is obtained, the core participating entities in the claim settlement case are extracted, including repair factories, loss assessors, vehicle owners, etc., which are taken as nodes in the graph. The business interaction relationships between entities such as "loss assessor A investigates the case submitted by repair factory B" and "vehicle owner C's vehicle is repaired at repair factory B" are extracted as edges connecting the nodes, thereby realizing the construction of the claim settlement relationship graph.
[0036] Meanwhile, for each node in the claim settlement relationship graph, all historical claim settlement cases thereof are sorted in chronological order, each historical claim settlement case is represented by a multi-dimensional feature vector, and the multi-dimensional feature vector is denoted as the claim settlement feature vector of the historical claim settlement case. The claim settlement feature vector includes claim settlement amount, repair item complexity, high-priced accessory proportion, claim settlement processing time, loss assessment deviation rate, etc. In order to ensure the dimensional consistency between different dimensions, the present application adopts Z-score standardization to perform standardization processing on each dimension respectively, and the data of each dimension included in the claim settlement feature vector is data after standardization processing.
[0037] For each node in the claim settlement relationship graph, the vehicle involved in each historical claim settlement case is represented by a multi-dimensional feature vector, and the multi-dimensional feature vector is denoted as the vehicle feature vector of the historical claim settlement case. The vehicle feature vector includes vehicle type, accident type, etc. The vehicle type includes small car, micro car, compact car, medium car, high-end car, luxury car, three-door car, CDV car, MPV car, SUV car, etc., and the accident type includes collision, scratch, spontaneous combustion, etc.
[0038] S2, classify the historical claim settlement cases of each node according to the vehicle feature vector, and construct a historical behavior inertia vector for each category according to the claim settlement feature vectors of the historical claim settlement cases in each category.
[0039] Specifically, the classification of the historical claim settlement cases of each node according to the vehicle feature vector includes: For all historical claim settlement cases of each node in the claim settlement relationship graph, historical claim settlement cases with the same vehicle feature vector are divided into the same category.
[0040] It should be noted that since a participant's historical behavior is not static, recent behavior is more representative of its current state than long-term behavior. Therefore, this invention adopts a weighted average method and uses a time decay function to calculate and determine the historical behavior inertia vector of each category. The historical behavior inertia vector represents the long-term behavior pattern of a node under a category, so that the historical behavior inertia vector can reflect both long-term trends and recent performance.
[0041] Specifically, the historical behavior inertia vector satisfies the expression:
[0042] In the formula, Represents a node The Historical behavior inertia vectors for each category; Represents a node The The first in the category Claims feature vector of a historical claims case; Represents a node The The first in the category The time of occurrence of each historical claim case; Indicates the current time; Represents a node The The number of historical claims cases included in each category; Representing the natural constant, this invention uses the natural constant. exponential function with base Perform negative correlation normalization; Indicates the time decay factor. It is a positive number.
[0043] In the formula, Represents a node The The first in the category The weight given to historical claims cases Represents a node The The first in the category The greater the time difference between a historical claim case and the current time, the more significant the milestone. The The first in the category The lower the weight given to historical claims cases, the less weight they receive. (Time decay factor) The time decay factor is set by the implementers based on the actual implementation situation. When the setting is larger, for example , the attention weight of historical claim cases with larger time difference significantly decreases, meaning that the historical behavior inertia vector will be more defined by recent historical behaviors. When the time decay factor is set smaller, for example , historical claim cases with larger time difference still retain certain weight, meaning that the historical behavior inertia vector tends to reflect long-term trends more.
[0044] S3, determining the overall deviation degree of the current claim case to the participant node according to the claim feature vector of the current claim case, the vehicle feature vector, and the historical behavior inertia vectors of each category of the participant node.
[0045] Specifically, each node corresponding to the current claim case is taken as a participant node. For each participant node, the claim feature vector of the current claim case and the vehicle feature vector are constructed, and the construction method is the same as that of the claim feature vector and the vehicle feature vector of the historical claim case in step S1.
[0046] For any participant node, in response to the participant node having no historical claim case, the overall deviation degree is set to 0; in response to the participant node having historical claim cases, the time deviation degree of the current claim case to each category of the participant node is determined according to the difference between the occurrence time of the historical claim case with the closest occurrence time to the current time and the occurrence time of the current claim case. The category deviation degree of the current claim case to each category of the participant node is determined according to the difference between the vehicle feature vector of the current claim case and the vehicle feature vector corresponding to the historical claim case of each category of the participant node. The behavior deviation degree of the current claim case to each category of the participant node is determined according to the difference between the claim feature vector of the current claim case and the historical behavior inertia vector of each category of the participant node. The reference weight of each category of the participant node is determined according to the category deviation degree of the current claim case to each category of the participant node and the proportion of the historical claim case of each category. The overall deviation degree of the current claim case to the participant node is determined according to the reference weight of each category of the participant node, the time deviation degree of the current claim case, and the behavior deviation degree of the participant node.
[0047] Specifically, the time deviation degree satisfies the expression:
[0048] In the formula, denotes the time deviation degree of the current claim case to the th category of the participant node; denotes the occurrence time of the current claim case; The first node of the participating node The time of occurrence of the most recent historical claim case in each category; This indicates the time of occurrence of the historical claim case that is furthest from the current time among the historical claim cases of the participating node; This represents the maximum value function, used to prevent the denominator from being 0; Used for Normalization is performed. If a participating node has not processed a certain type of case for a long time, and a similar case suddenly appears, that case deserves more attention.
[0049] It should be noted that, since this invention classifies historical claims cases with the same vehicle feature vector into the same category, all historical claims cases in each category have the same vehicle feature vector. Therefore, this invention determines the category deviation of the current claim case relative to each category of the participating node based on the difference between the vehicle feature vector of the current claim case and the vehicle feature vectors corresponding to the historical claims cases in each category of the participating node.
[0050] Specifically, the category deviation satisfies the expression:
[0051] In the formula, This indicates the current claim case relative to the node of the participating parties. Category deviation of each category; This represents the vehicle feature vector for the current claim case; The first node of the participating node Vehicle feature vectors corresponding to historical claims cases in each category; This represents the Hamming distance function, where the Hamming distance between two vectors is the number of distinct elements at corresponding positions in the two vectors. If the category deviation is 0, it means that the vehicle features of the current claim case perfectly match a certain category of the participating node; otherwise, the greater the category deviation, the lower the degree of matching.
[0052] Furthermore, the behavioral deviation satisfies the expression:
[0053] In the formula, This indicates the current claim case relative to the node of the participating parties. Deviation of behavior in each category; This represents the claim feature vector of the current claim case; The first node of the participating node Historical behavior inertia vectors for each category; Represents the dot product symbol; Indicates the modulus symbol; This represents the claim feature vector of the current claim case. The first node of the participating party Historical behavioral inertia vectors for each category The cosine similarity between them; the greater the cosine similarity, the closer the claim feature vector of the current claim case is to the first node of the participating party. The more similar the historical behavior inertia vectors of each category, the more likely the current claim case is to be related to the first node of the participating parties. The smaller the behavioral deviation of each category, the better, since the cosine similarity ranges from [-1, 1]. Therefore, this invention... Multiply The deviation of behavior is limited to the range of [0,1].
[0054] Furthermore, the reference weights for each category of the participating nodes satisfy the expression:
[0055] In the formula, The first node of the participating node Reference weights for each category; This indicates the current claim case relative to the node of the participating parties. Category deviation of each category; The first node of the participating node The ratio of the number of historical claims cases in each category to the total number of historical claims cases corresponding to the participating node; This represents the class decay factor, used to control the degree to which class deviation affects the weights. It is a positive number in this invention. In other embodiments, The implementation personnel shall determine this based on the actual implementation situation; This indicates the number of categories corresponding to the participating nodes; Representing the natural constant, this invention uses the natural constant. exponential function with base Perform negative correlation normalization.
[0056] In the formula, when the category deviation is 0, the vehicle feature vector of the current claim case and the first node of the participating party are... The fact that the vehicle feature vectors corresponding to historical claims cases in all categories are the same indicates that the vehicle features of the current claim case are the same as those of the participating party node. The first category is a perfect match, the second... The historical behavioral inertia vectors corresponding to each category are more valuable for reference in current claims cases. The reference weights of each category will increase significantly; when the category deviation is not zero, the reference weights will decrease significantly as the category deviation increases.
[0057] In the formula, the ratio Reflects the first The representativeness of each category among the participating nodes, if Larger, indicating that the first The [number] category accounts for a large proportion of the behavioral patterns of the participating nodes, therefore the [number]th category of the participating nodes... Reference weights for each category It should be larger, if Smaller, indicating the first The [number] category accounts for a small proportion of the behavioral patterns of the participating nodes, therefore the [number]th category of the participating nodes... Reference weights for each category The value should be relatively small. This invention introduces... This can effectively prevent certain rare categories from having an excessive impact on the overall deviation calculation. For example, suppose a repair shop mainly handles minor car scratches. Luxury car collision cases differ significantly from the repair shop's main business (minor car scratches) in terms of vehicle type, repair complexity, and compensation amount. If the repair shop suddenly starts handling luxury car collision cases, this behavior might be considered abnormal. If we simply ignore... This could lead to abnormal behavior (luxury car collision cases) having an unreasonable dominant effect on the overall deviation calculation, introducing... Subsequently, the reference weight for small car scratch cases will be significantly higher than that for luxury car collision cases, thus enabling the overall deviation calculation to more accurately reflect the main behavioral patterns of the participating nodes and avoid misjudgment due to rare categories.
[0058] Furthermore, the overall deviation of the current claim case relative to the participating node satisfies the expression:
[0059] In the formula, This indicates the overall deviation of the current claim case from the participating party's node; The first node of the participating node Reference weights for each category; This indicates the current claim case relative to the node of the participating parties. Deviation of behavior in each category; This indicates the current claim case's status regarding the participating party's node. Time deviation of each category; This indicates the number of categories corresponding to the participating nodes.
[0060] When time deviation A larger value indicates that the most recent historical claim case of the participating node in that category is from a longer period of time. This suggests that the participating node has not processed similar cases for a long time, and the sudden appearance of similar cases may be more noteworthy. Therefore, this invention utilizes the time deviation. Deviation from behavior Magnify, when time deviation When it is larger, The smaller the value, the lower the deviation in behavior. The greater the magnification, the greater the time deviation. The smaller, The closer a value is to 1, the more recent a historical claim case of the participating node in that category is to the current time, indicating a more continuous behavioral pattern for the participating node. In this case, the behavioral deviation is considered more accurate. The smaller the amplification, the more the overall deviation depends on the behavioral deviation.
[0061] This invention assigns reference weights to each category of participating nodes. Weighting is applied to ensure that the overall deviation reflects the degree of deviation of the current claims case from the main behavioral patterns of the participating parties, thus providing a basis for risk identification. When the overall deviation... The larger the value, the more likely the current claim case is to involve abnormal behavior or potential risks.
[0062] S4. Determine the volatility index of the participating node based on the overall deviation of the current claims case relative to the participating node and the trust score of the participating node.
[0063] It's important to note that mere deviation does not equate to fraud risk. A highly reputable repair shop handling a rare luxury car accident will inevitably deviate from its historical patterns, but this is normal business practice. Conversely, even minor deviations from the behavior of a repair shop with a poor track record warrant caution. Therefore, this invention introduces a node trust score to adjust the risk interpretation of the overall deviation, obtaining a volatility index for trust adjustment.
[0064] Specifically, for any participating node, the volatility index of the participating node satisfies the expression:
[0065] In the formula, This represents the volatility index of the participating nodes; This indicates the overall deviation of the current claim case from the participating party's node; This represents the trust score of the participating nodes before the current claim case occurred, with a value range of [0,1]. In the claim relationship graph, the initial trust score of each node can be set according to actual needs, such as 0.5. represents a trust sensitivity coefficient, used to control the suppression intensity of the trust score, is a positive number in the present application , in other embodiments, is determined by the implementer according to the actual implementation.
[0066] In line with the principle of being lenient to high-reputation parties and strict to low-reputation parties, the present application uses an exponential function to non-linearly suppress the overall deviation of the current claim case with respect to the participating nodes based on the trust scores of the participating nodes before the occurrence of the current claim case. As an adjustment factor, the overall deviation is non-linearly suppressed when the trust score of the participating node is higher, is smaller, thereby significantly suppressing the influence of the overall deviation , so that the resulting volatility index is also smaller, thereby achieving "trust inertia" for high-reputation nodes, allowing them to have certain behavioral fluctuations in business. Conversely, when the trust score of the participating node is lower, is larger, and tends to 1, almost no suppression effect is produced , and the volatility index will directly reflect the original size of the overall deviation, thereby achieving high sensitivity to the behavior of low-reputation nodes.
[0067] S5, for any node, the basic trust score of the node is determined according to the volatility index and the trust score of the node before the occurrence of the current claim case, the propagation trust score of all nodes pointing to the node is determined according to the basic trust scores of the nodes pointing to the node, and the trust score of the node is updated according to the basic trust score and the propagation trust score of the node.
[0068] Specifically, for any node in the claim relationship graph, the basic trust score of the node is determined according to the trust score of the node before the occurrence of the current claim case and the volatility index of the node:
[0069] In the formula, represents the basic trust score of the node; represents the trust score of the node before the occurrence of the current claim case; represents the set of participating nodes of the current claim case, represents the volatility index of the node when the node is a participating node of the current claim case. When the node is not a participating node of the current claim case, the volatility index of the node Belongs to set When a node is in the middle, the node As a participating node in the current claim case, the specific claim processing actions in the current claim case affect the node. The trust score has a direct impact, therefore, the node... The volatility index is related to the time point before the occurrence of the current claim. Trust score Make corrections to obtain nodes. The basic trust score; when the node Not a set When the node is in the current claim case, the node is... The trust score has no direct impact, therefore the current claim settlement point before the incident is directly considered. Trust score As a node The basic trust score.
[0070] Furthermore, for any node in the claims relationship graph, based on the basic trust scores of each node pointing to that node, the propagation trust scores of all nodes pointing to that node are determined:
[0071] In the formula, Indicates pointing to a node All node pairs Trust score in dissemination; Represents all nodes pointed to in the claims relationship graph. The set of nodes; Represents a node Basic trust score; Represents a node The out-degree of the node. The number of outgoing edges, This is a minimum value function used to prevent the propagation trust score from exceeding 1.
[0072] In the formula, Represents a node To the node The amount of trust in the propagation, when the node When a node has high credibility and its behavior is stable, The basic trust score is high, and the node The level of trust propagated to the nodes it points to is relatively high. When a node While possessing high credibility, a company's base trust score decreases when its behavior in current claims cases fluctuates drastically compared to its historical behavior. The ability to spread trust outwards is significantly weakened. When nodes... When the node has a low reputation, The basic trust score is low, and the node The trust level propagated to the nodes it points to is relatively high. This invention integrates all the nodes it points to. node to node The amount of trust in the propagation is used to obtain the pointer to the node. All node pairs Trust score of dissemination When all pointing to the node When nodes have high credibility and stable behavior, the trust score is propagated. The larger.
[0073] Furthermore, for any node in the claims relationship graph, the updated trust score of that node is determined based on its base trust score and propagation trust score:
[0074] In the formula, Indicates the point in time after the current claim case occurred. Updated trust score; Represents a node Basic trust score; Indicates pointing to a node All node pairs Trust score in dissemination; The damping coefficient is used to control the proportion of trust scores propagated from neighboring nodes. This invention considers the behavior of a node itself to be just as important as the combined behavioral patterns of its neighboring nodes; therefore, the damping coefficient is used to... The damping coefficient is set to 0.5. In other embodiments, the implementer can set the damping coefficient according to the actual implementation situation. .
[0075] If node Behavioral stability (i.e.) Larger), and points to the node The behavior of each node is also stable (i.e. If the node is larger, then the node Updated trust score Larger. If the node Unstable behavior (i.e.) (smaller), or pointing to a node The behavior of each node is unstable (i.e.) If the node is smaller, then the node Updated trust score Smaller.
[0076] S6. Intelligent supervision of auto insurance claims based on the updated trust scores of each node.
[0077] Specifically, the nodes are divided into different risk levels according to the updated trust scores of the nodes, in the embodiment of the application, the nodes with trust scores less than 0.3 are regarded as high-risk nodes, the nodes with trust scores greater than or equal to 0.3 and less than 0.7 are regarded as medium-risk nodes, and the nodes with trust scores greater than or equal to 0.7 are regarded as high-risk nodes. In other embodiments, the implementers can set the risk levels and the division rules of the risk levels according to the actual implementation conditions.
[0078] Further, for the high-risk nodes, a first early warning of the car insurance claim system is triggered, the auditing frequency of the high-risk nodes is automatically increased, the permission of the high-risk nodes for rapid claim is limited, and manual intervention is notified for verification.
[0079] For the medium-risk nodes, a second early warning of the car insurance claim system is triggered, and the claim cases submitted by the medium-risk nodes are sampled and verified.
[0080] For the low-risk nodes, no early warning of the car insurance claim system is triggered, the claim process is optimized, and the auditing steps are reduced to improve the efficiency.
[0081] The embodiment of the application also discloses a car insurance claim intelligent supervision system based on big data, comprising a processor and a memory, and the memory stores computer program instructions, when the computer program instructions are executed by the processor, a car insurance claim intelligent supervision method based on big data according to the application is realized.
[0082] The above system also comprises a communication bus and a communication interface and other components familiar to those skilled in the art, the settings and functions of which are known in the art, and thus will not be described here.
Claims
1. A big data-based intelligent supervision method for vehicle insurance claims, characterized in that, The method comprises the following steps: constructing a claim relationship graph, generating a claim feature vector corresponding to each historical claim case of each node in the claim relationship graph and a vehicle feature vector; classifying the historical claim cases of each node according to the vehicle feature vector, and constructing a historical behavior inertia vector of each category according to the claim feature vectors of the historical claim cases in each category; determining the overall deviation of the current claim case with respect to each participant node according to the claim feature vector, the vehicle feature vector and the historical behavior inertia vector of each category of the participant node; determining the volatility index of the participant node according to the overall deviation and the trust score of the participant node; for any node, determining the basic trust score of the node according to the trust score of the node before the occurrence of the current claim case and the volatility index of each participant node, determining the propagation trust score of all nodes pointing to the node to the node according to the basic trust score of each node pointing to the node, updating the trust score of the node according to the basic trust score and the propagation trust score, and performing intelligent supervision on motor vehicle insurance claims according to the updated trust score of each node. 2.The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 1, characterized in that, The method for generating the claim feature vector corresponding to each historical claim case of each node in the claim relationship graph and the vehicle feature vector comprises: The claim feature vector includes claim amount, repair project complexity, high-priced accessory proportion, claim processing time and loss assessment deviation rate; and the vehicle feature vector includes vehicle type and accident type. 3.The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 1, characterized in that, The method for classifying the historical claim cases of each node according to the vehicle feature vector and constructing the historical behavior inertia vector of each category according to the claim feature vectors of the historical claim cases in each category comprises: For any node, the historical claim cases with the same vehicle feature vector are divided into the same category; the attention weight of the historical claim cases in the category is determined according to the difference between the current time and the occurrence time of the historical claim cases in the category, and the attention weight is negatively correlated with the difference; the claim feature vectors of all the historical claim cases in the category are weighted and summed according to the attention weight of the historical claim cases in the category to obtain the historical behavior inertia vector of the corresponding category. 4.The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 1, characterized in that, The method for obtaining the overall deviation comprises: determining the time deviation of the current claim case with respect to each category of the participant node; taking the Hamming distance between the vehicle feature vector of the current claim case and the vehicle feature vector corresponding to the historical claim cases in each category of the participant node as the category deviation of the current claim case with respect to each category of the participant node; determining the behavior deviation of the current claim case with respect to each category of the participant node according to the cosine similarity between the claim feature vector of the current claim case and the historical behavior inertia vector of each category of the participant node; determining the reference weight of each category according to the category deviation and the proportion of the historical claim cases in each category; determining the overall deviation of the current claim case with respect to the participant node according to the reference weight and the time deviation and the behavior deviation.
5. The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 4, characterized in that, The method for obtaining the reference weight comprises: For any one category of the participant node, the category deviation degree of the current claim case relative to the category of the participant node is negatively correlated and normalized; and a reference weight of the category is obtained according to a product between a result after the negative correlation normalization and an occupation ratio of historical claim cases of the category. 6.The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 4, characterized in that, The overall deviation degree satisfies an expression: ； In the formula, represents the overall deviation degree of the current claim case with respect to the participant node; represents the reference weight of the first category of the participant node; represents the behavior deviation degree of the current claim case with respect to the first category of the participant node; represents the time deviation degree of the current claim case for the first category of the participant node; represents the number of categories corresponding to the participant node. 7.The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 1, characterized in that, The determination of the volatility index of the participant node comprises: ； In the formula, represents the volatility index of the participant node; represents the overall deviation of the current claim case with respect to the participant node; represents the trust score of the participant node before the occurrence of the current claim case; represents the trust sensitivity coefficient. 8.The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 1, characterized in that, The basic trust score satisfies an expression: wherein, represents a node 's base trust score; represents a node 's trust score before the current claim case occurs; represents a set of participant nodes of the current claim case, represents a volatility index of a node when the node is a participant node of the current claim case; The propagation trust score satisfies the expression: , denotes the propagation trust score of all node pairs pointing to node ; denotes the set of all nodes in the claim relationship graph pointing to node ; denotes the base trust score of node ; denotes the out-degree of node ; is a minimum function. 9.The big data-based intelligent supervision method for vehicle insurance claim settlement according to claim 1, characterized in that, The updating of the trust score of the node comprises: The basic trust score of the node and the propagation trust scores of all nodes pointing to the node are weighted and summed to obtain the updated trust score of the node.
10. A big data-based vehicle insurance claim intelligent supervision system, characterized in that, Comprise: A processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, a kind of big data-based vehicle insurance claim intelligent supervision method according to any one of claims 1-9 is realized.