A method for filtering risk control rules in the insurance industry to accelerate execution
By building a rule factor tree and combining case information to predict risk projects, the problem of reduced execution speed of rules in the auto insurance claims system is solved, the processing efficiency and accuracy are improved, and the effective identification and prevention of potential risks are achieved.
Patent Information
- Application Number
- CN202411330301.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-24
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-24
AI Technical Summary
In the case of high concurrency, the existing auto insurance claims system has significantly reduced the execution speed of rules and reduced processing efficiency due to the increase in the number of rules and the complexity of claims cases.
By obtaining a set of rules containing risk item labels and rule factor labels, a rule factor tree is constructed, and risk item prediction is carried out based on case loss determination information, accurate rule statistics and matching are carried out, and selected rule sets are obtained and case detection is performed.
It improves case handling efficiency, improves the accuracy of rule execution, and effectively identifies and prevents potential risks.
Smart Images

Figure CN118864121B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of risk control for automobile insurance claims, and in particular to a method for filtering and accelerating the execution of risk control rules in the insurance industry. Background Art
[0002] With the rapid development of the vehicle insurance industry, the auto insurance claims system has become one of the core tools for insurance companies to handle auto insurance cases. The current auto insurance claims process usually includes multiple links such as reporting, investigation, damage assessment, loss verification, and compensation. Insurance companies use these processes to evaluate and handle each claim case. In this process, insurance companies usually apply a set of complex risk control rules to detect and prevent possible risks.
[0003] The existing risk control rule system mainly uses a rule engine to execute a series of predefined rules, which cover all aspects of vehicle repair, including parts replacement, labor time management, accident liability, etc. However, with the increase in the number of rules and the complexity of claims cases, the speed of rule execution of existing technologies has been significantly reduced under high concurrency conditions, resulting in reduced processing efficiency. Summary of the invention
[0004] The purpose of this application is to provide a method for filtering and accelerating the execution of risk control rules in the insurance industry, so as to solve the technical problem in the prior art that due to the increase in the number of rules and the complexity of claims cases, the speed of rule execution is significantly reduced in high concurrency situations, resulting in a decrease in processing efficiency.
[0005] In view of the above problems, the present application provides a method for filtering and accelerating the execution of risk control rules in the insurance industry, comprising: obtaining a rule set, wherein the rule set has a risk project label and a rule factor label; clustering and grouping the rule set according to the risk project label to obtain multiple cluster rules and multiple risk project type identifiers; aggregating the rule set according to the rule factor label to obtain a rule factor tree, wherein the rule factor tree includes an intermediate leaf node and a bottom leaf node, the intermediate leaf node is a node with child nodes, the bottom leaf node is a node without child nodes, and the intermediate leaf node has a rule factor type identifier; in response to the pending case uploaded by the user-side interface, obtaining case damage assessment information, wherein the case damage assessment information includes a list of accessory types, a list of working hours, and a subject matter type; performing risk project prediction according to the list of accessory types, the list of working hours, and the subject matter type to obtain a risk project prediction result; based on the risk project prediction result, performing rule statistics according to the multiple risk project type identifiers and the rule factor type identifier to obtain a selected rule set; and performing case detection of the pending case according to the selected rule set.
[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0007] A rule set is obtained, wherein the rule set has a risk project label and a rule factor label; according to the risk project label, the rule set is clustered and grouped to obtain multiple cluster rules and multiple risk project type identifiers; according to the rule factor label, the rule set is aggregated to obtain a rule factor tree, wherein the rule factor tree includes an intermediate leaf node and a bottom leaf node, the intermediate leaf node is a node with child nodes, the bottom leaf node is a node without child nodes, and the intermediate leaf node has a rule factor type identifier; in response to the pending case uploaded by the user-side interface, case damage assessment information is obtained, wherein the case damage assessment information includes a list of accessory types, a list of working hours, and a subject matter type; risk project prediction is performed according to the list of accessory types, the list of working hours, and the subject matter type to obtain a risk project prediction result; based on the risk project prediction result, rule statistics are performed according to the multiple risk project type identifiers and the rule factor type identifier to obtain a selected rule set; case detection of the pending case is performed according to the selected rule set. By obtaining a rule set containing risk item labels and rule factor labels, building a rule factor tree, and combining the case loss assessment information to predict risk items, accurate rule statistics and matching can be performed, and finally the selected rule set can be obtained and case detection can be performed, thereby achieving the technical effect of improving case handling efficiency, enhancing rule execution accuracy, and effectively identifying and preventing potential risks.
[0008] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0009] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.
[0010] Figure 1 A flowchart of a method for filtering and accelerating the execution of risk control rules in the insurance industry for this application;
[0011] Figure 2 A flowchart of obtaining risk project prediction results in a method for filtering insurance industry risk control rules to accelerate execution is provided in this application. DETAILED DESCRIPTION
[0012] This application provides a method for filtering and accelerating the execution of risk control rules in the insurance industry, which solves the technical problem in the prior art that due to the increase in the number of rules and the complexity of claims cases, the speed of rule execution is significantly reduced under high concurrency conditions, resulting in a decrease in processing efficiency. By obtaining a rule set containing risk project labels and rule factor labels, constructing a rule factor tree, and combining the case loss assessment information to predict risk projects, accurate rule statistics and matching are performed, and finally the selected rule set is obtained and case detection is performed, so as to achieve the technical effect of improving case processing efficiency, improving rule execution accuracy, and effectively identifying and preventing potential risks.
[0013] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.
[0014] Please see attached Figure 1 The present application provides a method for filtering insurance industry risk control rules to accelerate execution, which specifically includes the following steps:
[0015] Step 1: Obtain a rule set, wherein the rule set has a risk item label and a rule factor label.
[0016] Specifically, the rule set can be seen as the complete set of all rules that may be applied to a case, and each rule is a set of predefined logical conditions used to determine whether there are certain risks in the case. Risk item labels refer to the detection of specific items in the case by the rule, such as replacement of parts and labor costs in vehicle repairs. Each risk item label corresponds to one or more specific risk items, reflecting the conditions under which these items may be considered risky. For example, if the risk item label of a rule is replacement of parts, then the rule will focus on determining whether there is any abnormality or potential fraud in the replacement of vehicle parts in a case.
[0017] Rule factor labels are used to identify specific conditions or factors in rules and are the basic elements that constitute rules. For example, rule factors can be parts type and labor costs, etc. These factors appear repeatedly in different rules and constitute the judgment logic of the rules. When a factor in a case meets specific conditions, the rule corresponding to the factor label will be activated to further judge the risk items. For example, in the process of auto insurance claims, if a case involves a high-value imported part type (this is a rule factor) and the labor cost is abnormally high (another rule factor), then these factors will trigger the rules related to the replacement of high-value imported parts and labor costs to determine whether there is a risk of false reporting of expenses.
[0018] By aggregating all relevant rules and identifying and classifying them through risk item labels and rule factor labels, in the subsequent case processing, specific rules related to the case can be quickly identified and called, improving the efficiency and accuracy of rule execution.
[0019] Step 2: Cluster and group the rule set according to the risk project label to obtain multiple cluster rules and multiple risk project type identifiers.
[0020] Specifically, clustering refers to grouping rules with similar characteristics or the same goals together to form multiple rule clusters, so that relevant rules can be applied more quickly and accurately when handling cases. First, extract the risk item label of each rule from the rule set, identify which rules are for the same or similar risk items, and cluster these rules together to form a rule cluster. For example, if multiple rules all involve the risk item of vehicle parts replacement, they are grouped into the same cluster to obtain multiple clusters of rules, and the rules in each rule cluster are associated with one or more similar risk items. This not only makes the management and application of rules more orderly, but also reduces the number of rules that need to be traversed in case handling. For example, when handling a case involving vehicle repair costs, it is only necessary to quickly call the rule cluster related to this risk item without having to check all the rules in the entire rule set.
[0021] In addition, each rule cluster corresponds to a risk item type identifier, which reflects the specific risk item type covered by the rule cluster. For example, if a rule cluster includes all rules related to the replacement of high-value vehicle parts, then the risk item type identifier of this rule cluster is the risk of high-value parts replacement. In this way, the rule set is clustered and grouped, making the rule execution process more efficient, while ensuring that in actual applications, the risk assessment of the case can be carried out quickly and accurately.
[0022] Step three: According to the rule factor label, aggregate the rule set to obtain a rule factor tree, wherein the rule factor tree includes an intermediate leaf node and a bottom leaf node, the intermediate leaf node is a node with child nodes, the bottom leaf node is a node without child nodes, and the intermediate leaf node has a rule factor type identifier.
[0023] Specifically, the rule factor label is used to identify the specific conditions or factors in the rule, which are the basic components of rule judgment. By aggregating the rule factor labels, all rule factors are classified and sorted. The aggregation result is a rule factor tree, which is a hierarchical structure used to represent the relationship and execution order between rule factors. The rule factor tree consists of multiple nodes, including intermediate leaf nodes and bottom leaf nodes.
[0024] Intermediate leaf nodes refer to nodes with child nodes, which usually represent higher-level rule factors. For example, if the root node of a rule factor tree is the accessory type, then its intermediate leaf nodes may be domestic parts and imported parts, which are further connected to more specific factors, such as accessory prices. The bottom leaf node is the last node in the rule factor tree, which has no child nodes and usually represents the most specific rule conditions or judgment results. For example, in the rule factor tree of accessory types, the bottom leaf node may be that the price of imported parts is higher than a certain threshold, and this node directly corresponds to a specific rule judgment.
[0025] By constructing a rule factor tree, rule factors can be retrieved from top to bottom according to the tree structure, starting from the root node and going deeper layer by layer until the bottom leaf node that meets the conditions is found. The advantage of this structure is that it avoids repeated rule factor judgments and improves the efficiency of rule execution. For example, if a case involves the factor of imported parts, it only needs to quickly locate the rules related to imported parts along the tree structure without traversing all rule sets.
[0026] Step 4: In response to the pending cases uploaded through the user interface, the case damage assessment information is obtained, wherein the case damage assessment information includes a list of accessory types, a list of working hours, and the type of subject matter.
[0027] Specifically, after uploading the case to be executed on the user interface, the case damage assessment sheet information is obtained. The case damage assessment sheet information refers to the detailed records generated during the case handling process, which contains various key data related to the case, and these data are the basis for subsequent rule execution.
[0028] The damage assessment information of a case usually includes a list of accessory types, a list of labor hours, and the type of subject matter. Among them, the list of accessory types lists all the types of accessories involved in the vehicle repair process. Each accessory has a specific code, name, and category. For example, in a vehicle accident case, the list of accessory types may include front bumpers, engine hoods, door glass, etc. These types of accessories will be used as input conditions for subsequent rule judgments. The labor hour list records the time required to repair each accessory. These times are usually in hours or minutes, and will involve different types of work, such as sheet metal workers, painters, etc. The labor hour list is an important basis for evaluating repair costs. The subject matter type refers to the category of the main object or vehicle involved in the case. Different subject matter types may affect the selection and execution of rules.
[0029] When a user uploads a case to be executed through the interface, the corresponding damage assessment form information is automatically extracted and converted into a standardized data format to generate the case damage assessment form information. Through this process, all necessary information related to the case can be quickly obtained, laying the foundation for subsequent rule execution and risk assessment.
[0030] Step 5: Predict risk items based on the accessory type list, the labor hour list, and the subject matter type to obtain risk item prediction results.
[0031] Specifically, risk item prediction refers to identifying possible risk items by analyzing information such as accessories, work hours, and subject matter in the case. First, select the appropriate risk prediction model based on the subject matter type. The subject matter type (such as family cars, light trucks, etc.) determines the risk assessment criteria and focus of different vehicle categories. Next, analyze it in combination with the list of accessory types and the list of work hours. These lists provide the specific repair items involved in the case and the time required. This information is input into the prediction model, and the model calculates the risk value of each item based on factors such as the type of accessories and the length of repair time. In this way, the generated risk item prediction results include not only which items may be risky, but also the specific manifestations of these risks, such as high repair costs, non-standard operations, etc. These prediction results will be directly used to guide subsequent rule execution to ensure that high-risk items can be handled first, thereby improving the accuracy and efficiency of case handling.
[0032] Step 6: Based on the risk project prediction result, rule statistics are performed according to the multiple risk project type identifiers and the rule factor type identifiers to obtain a selected rule set.
[0033] Specifically, the risk project prediction results are generated after analyzing the case information in the previous step, reflecting which projects in the case may be risky. Next, rule statistics are performed based on the risk project type identification and the rule factor type identification. When performing rule statistics, all rules related to the risk project type identification are first extracted from the rule factor tree. For example, if the risk project prediction results show that there is an abnormality in the replacement of accessories, all rules related to the replacement of accessories will be found to generate a selected rule set. In this way, the scope of rules that need to be executed can be quickly narrowed, and the rule set most relevant to the current case can be filtered out from the original large rule set, effectively avoiding unnecessary rule execution and focusing on projects that are most likely to be at risk, thereby significantly improving the efficiency and accuracy of rule execution.
[0034] Step 7: Perform case detection on the case to be executed according to the selected rule set.
[0035] Specifically, case detection refers to using a selected set of rules to check each aspect of a case one by one to confirm whether there are any violations or anomalies. This process relies on previously selected rule sets, which have been accurately matched according to the specific circumstances of the case, so that potential problems in the case can be detected in a targeted manner. For example, if the selected rule set includes price monitoring rules for high-priced imported parts and management rules for excessive working hours, the price of accessories and working hours data in the case will be checked separately. First, it will be checked whether the price of accessories exceeds the reasonable range, such as whether the price of a certain accessory is more than 50% higher than the market reference price. Then, it will be checked whether the working hours exceed the standard, such as whether the time to repair a certain accessory obviously exceeds the standard working hours. In this way, anomalies and potential risks in the case can be quickly discovered in case detection, and corresponding feedback can be given. This not only improves the efficiency of case handling, but also effectively prevents unreasonable claims or fraud.
[0036] Further, step 2 of this application includes:
[0037] Obtain a first rule of the rule set, wherein the first rule stores a first risk project label in the risk project label; obtain a second rule of the rule set, wherein the second rule stores a second risk project label in the risk project label; when the first risk project label and the second risk project label are the same, add the first rule and the second rule into the same cluster; when the first risk project label and the second risk project label are different, add the first rule and the second rule into different clusters; until all rule sets are traversed and analyzed, output the multi-cluster rules and the multiple risk project type identifiers.
[0038] Specifically, first, any rule in the rule set is recorded as the first rule, and the first rule has a corresponding first risk project label stored in the risk project label. Then, any rule other than the first rule is extracted from the rule set and recorded as the second rule. Similarly, the second rule has a corresponding second risk project label. Next, the first risk project label and the second risk project label are compared to determine whether they are the same. If the first risk project label and the second risk project label are the same, it means that the two rules are both for the same risk project type, for example, they may both be rules about accessory replacement. In this case, the first rule and the second rule will be added to the same rule cluster. If the first risk project label and the second risk project label are different, the two rules will be placed in different rule clusters respectively. For example, the first rule may be about accessory replacement, while the second rule is about working time management. The two involve different risk projects, so they will be classified into different clusters. This process will continue until all the rules in the rule set are traversed and analyzed, and the rule set is divided into multiple rule clusters, each of which corresponds to a risk project type identifier. In this way, a large number of rules can be effectively organized and managed, laying the foundation for subsequent rule execution.
[0039] Further, step three of this application includes:
[0040] According to the rule factor label, the rule set is clustered and grouped to obtain multiple groups of rule sets and multiple rule factor type identifiers; according to the number of multiple rules in the multiple groups of rule sets, the multiple rule factor type identifiers are sorted from large to small to obtain the rule factor type identifier sorting result; the rule factor type identifier sorting result is traversed, starting from the root node, to build a rule factor tree trunk; the multiple groups of rule sets are distributed on the rule factor tree trunk to construct branch nodes to obtain the rule factor tree; wherein, the rule factor tree includes intermediate leaf nodes and bottom leaf nodes, the intermediate leaf nodes are nodes with child nodes, the bottom leaf nodes are nodes without child nodes, and the intermediate leaf nodes have rule factor type identifiers.
[0041] Specifically, based on the rule factor label, the rules with the same rule factor label in the rule set are grouped together to form multiple rule sets. Each rule set contains all the rules related to a specific rule factor. For example, all the rules related to imported accessories are grouped in one rule set. Next, the rules are sorted according to the number of rules contained in each rule set, and the multiple rule sets are arranged from most to least according to the number of rules to generate the rule factor type identification sorting result.
[0042] Then, we traverse the sorted results of the rule factor type identifiers, starting from the root node, and gradually build the trunk of the rule factor tree. The trunk of the rule factor tree is composed of these sorted rule factor type identifiers, and the root node is usually the most important or common rule factor. For example, the root node may be the type of accessory, followed by the working hours, and then the price of the accessory, etc. This top-down construction method ensures the logic and hierarchy of the rule factor tree.
[0043] After building the trunk, the multiple sets of rules obtained by clustering and grouping are distributed in the trunk of the rule factor tree, and branch nodes are gradually constructed. In this process, the rule sets are decomposed layer by layer according to the conditions of different rule factors, and finally a complete rule factor tree is formed. The rule factor tree contains intermediate leaf nodes and bottom leaf nodes. Intermediate leaf nodes are nodes with child nodes, which usually correspond to relatively broad rule factors. Bottom leaf nodes are the last level nodes in the rule factor tree, without child nodes, and usually represent the most specific rule judgment conditions or the final executed rules. In this way, a structured rule factor tree is finally obtained. When handling actual cases, the rules are gradually screened and applied in an efficient and accurate manner, thereby improving the speed and accuracy of rule execution. For example, when a case involves imported accessories, the relevant specific rules can be quickly located along the rule factor tree without traversing the entire rule base.
[0044] Furthermore, the present application also includes the following steps:
[0045] According to the sorting result of the rule factor type identifier, a first sequence number rule factor type identifier, a second sequence number rule factor type identifier, and so on are obtained until the Nth sequence number rule factor type identifier; the first sequence number rule factor type identifier is set as a root node; the second sequence number rule factor type identifier is linked to the root node and is set as a secondary trunk node; until the Nth sequence number rule factor type identifier is linked to the N-1th level trunk node and is set as an Nth level trunk node, thereby generating the rule factor tree trunk.
[0046] Specifically, according to the obtained sorting result of the rule factor type identifier, the first-order rule factor type identifier is obtained first, which is the rule factor type with the largest number of rules in the sorting. Then, the second-order rule factor type identifier is obtained in sequence, until the N-th order rule factor type identifier, and these identifiers constitute the order of the number of rules from high to low. When constructing the trunk of the rule factor tree, the first-order rule factor type identifier is first set as the root node. The root node is the starting point of the rule factor tree. Next, the second-order rule factor type identifier is linked to the root node and set as the secondary trunk node. The secondary trunk node extends directly from the root node and represents the next-level rule factor. Continuing according to this logic, the third, fourth, and N-th order rule factor type identifiers are linked to their upper-level nodes step by step, forming a trunk chain from the root node to multiple child nodes, that is, the trunk of the rule factor tree. Each node on the trunk of the rule factor tree represents a rule factor type identifier. When executing rules, relevant rules can be filtered layer by layer from the root node in an orderly manner, thus avoiding disordered traversal and repeated judgments, and ensuring the efficiency and accuracy of the rule execution process.
[0047] Furthermore, the present application also includes the following steps:
[0048] Step a: obtain a set of k-level trunk node rules; step b: divide the rules triggered when the k-level trunk node rule factor is met into a group, and store them in the first leaf node of the k-level trunk node; step c: divide the rules triggered when the k-level trunk node rule factor is not met into a group, and store them in the second leaf node of the k-level trunk node; based on the execution process of the above steps a, b and c, traverse the trunk of the rule factor tree to construct branch nodes and obtain the rule factor tree.
[0049] Specifically, in the process of constructing the rule factor tree, the trunk nodes will be further refined and the corresponding branch nodes will be generated to ensure the accurate matching and execution of the rules. This process includes classifying the rules of each trunk node and generating corresponding leaf nodes based on the classification results, and finally forming a complete rule factor tree. Specifically, first obtain the set of k-level trunk node rules. In the structure of the rule factor tree, the k-level trunk node refers to the trunk node located at the kth layer in the trunk of the rule factor tree, and this node usually represents a rule factor type identifier. Next, the rules triggered when the k-level trunk node rule factor is met are grouped together and stored in the first leaf node of the k-level trunk node. The first leaf node represents those rules that should be activated when specific conditions are met. For example, if the k-level trunk node is an imported accessory, then the first leaf node may contain all the rules related to imported accessories and triggered when the price exceeds the standard.
[0050] At the same time, the rules triggered when the k-level trunk node rule factor is not met are also classified into another group and stored in the second leaf node of the k-level trunk node. The second leaf node represents the rules that need to be activated when a specific condition is not met. For example, if the condition is imported accessories, and the accessories in a case are domestically produced, then the relevant rules will be classified into the second leaf node.
[0051] Through this classification process, the rules that need to be triggered in different situations can be effectively grouped to ensure that the system can select an appropriate set of rules according to the actual situation when handling a case. Based on the execution process of the above steps a, b and c, the trunk of the rule factor tree is traversed, and the branch nodes of each trunk node are gradually constructed. This process is achieved by analyzing the rule set of each trunk node layer by layer and further refining it into leaf nodes under different conditions. By traversing and constructing branch nodes, a complete rule factor tree is finally obtained. The rule factor tree not only contains the hierarchical relationship from the root node to the trunk nodes at all levels, but also refines it to the leaf nodes, accurately mapping the specific rules that should be executed in various situations. Through this fine rule division and node construction, it can be ensured that in the case handling process, all rules can be accurately matched and efficiently executed according to the actual situation, thereby improving the accuracy and efficiency of rule execution.
[0052] Further, as attached Figure 2 As shown, step five of this application includes:
[0053] According to the type of the target object, a target risk project prediction model is selected from a risk project prediction model library; the target risk project prediction model includes a sample retrieval layer, and the sample retrieval layer is used to collect a sample distribution tree with the accessory type list and the working hour list as constraints; the target risk project prediction model includes a sample aggregation layer, and the sample aggregation layer is used to aggregate the sample distribution tree from the bottom layer to the top layer, and output the risk project prediction result.
[0054] Specifically, when predicting risk projects, the most suitable target risk project prediction model is selected from the risk project prediction model library according to the type of the subject matter, ensuring that the most suitable model can be used for risk assessment according to the specific characteristics of the case, thereby improving the accuracy and reliability of the prediction. Specifically, different types of subject matter may have different focuses in risk assessment, so it is crucial to select a suitable risk project prediction model. According to the type of the subject matter of the case, the target risk project prediction model that matches it is screened out from the model library. Among them, the risk project prediction model library is a collection for storing and managing various risk prediction models. These models are designed to evaluate different types of risk projects and accurately predict possible risks based on different types of subject matter, types of accessories and working hours. The core function of the risk project prediction model library is to provide a variety of optional prediction models to cope with different case types and complex scenarios. For example, each model is optimized for a specific risk project or case type. For example, the risk project prediction model library may include risk prediction models specifically for different vehicle types such as family cars and light trucks, which are specifically constructed by professional and technical personnel in this field by collecting a large amount of sample data for different types of subject matter.
[0055] Once the target risk project prediction model is determined, enter the sample retrieval layer of the target risk project prediction model. The role of the sample retrieval layer is to build a sample distribution tree by retrieving and collecting relevant sample data. In the sample retrieval process, the list of parts types and the list of working hours are used as constraints to collect samples that match the parts type and working hours information in the case. For example, if the case involves the replacement of imported parts and the working hours exceed the standard time, all similar case data will be searched in the sample library to form a sample distribution tree based on these conditions.
[0056] The sample distribution tree is a structured data set that organizes and arranges related sample data in a certain hierarchical relationship, similar to the nodes of a tree structure. Each node represents a sample data point, and the tree structure shows the distribution of these samples under different conditions. This tree structure can help the system quickly identify the historical data that is most similar to the current case, providing a basis for subsequent risk prediction.
[0057] Next, we enter the sample aggregation layer. The function of the sample aggregation layer is to aggregate the data in the sample distribution tree from the bottom layer to the top layer, so as to generate the final risk project prediction results. This aggregation process is completed by statistically analyzing the commonalities and differences of the sample data. Starting from the bottom node of the sample distribution tree, the data is gradually merged upward to form a higher-level comprehensive result. For example, all samples related to imported accessories are aggregated to calculate the average risk level of this type of accessories in similar cases. Finally, after processing by the sample aggregation layer, a comprehensive risk project prediction result will be output, including the risk projects in the current case. The risk project prediction results will be directly used to guide subsequent rule execution and decision-making, so that high-risk projects can be dealt with in a targeted manner to avoid potential losses.
[0058] Furthermore, the present application also includes the following steps:
[0059] Constructing a constraint condition, wherein the constraint condition has a first constraint function and a first constraint function threshold, a second constraint function and a second constraint function threshold; wherein the first constraint function is the ratio of the intersection of the sample and the accessory type list to the union of the accessory type, and the first constraint function threshold is equal to 1; wherein the second constraint function is the sample working hour list of the sample accessory type list, and the mean deviation of the working hours of the same type of accessories of the benchmark working hour list of the benchmark accessory type list, and the second constraint function threshold is a user-defined working hour deviation threshold; taking the accessory type list as the first-level benchmark accessory type list, and taking the working hour list as the first-level benchmark working hour list, retrieve the first number of second-level samples that meet the constraint condition; constructing a benchmark condition with the first number of second-level samples, and retrieving the first number of third-level samples that meet the constraint condition; until constructing a benchmark condition with the first number of M-1-level samples, and retrieving the first number of M-level samples that meet the constraint condition, wherein 20≥M≥5; taking the first-level benchmark accessory type list and the first-level benchmark working hour list as root nodes, traverse the first number of second-level samples, the first number of third-level samples until the first number of M-level samples, and construct the sample distribution tree.
[0060] Specifically, with the accessory type list and the work hour list as constraints, the steps of collecting the sample distribution tree include: first, defining constraint bars, including a first constraint function and a second constraint function, each function having a corresponding threshold value for limiting and filtering sample data. The purpose of the first constraint function is to evaluate the relevance of the sample to the accessory type list in the case. Specifically, the first constraint function calculates the ratio of the intersection and union of the sample accessory type and the accessory type list in the current case. This ratio reflects the degree of match between the accessory types in the sample data and those in the current case. If the ratio of the intersection to the union is 1, it means that the accessory type in the sample is exactly the same as the accessory type in the case, which is the standard set by the threshold of the first constraint function. In practical applications, samples that meet this condition will be regarded as samples that are highly relevant to the case.
[0061] The second constraint function focuses on the matching degree of the labor time list, calculates the average labor time deviation of the same type of accessories in the sample and the case, compares the sample labor time list with the benchmark labor time list of the benchmark accessory type list, and calculates the average deviation between the two. The second constraint function threshold is a user-defined labor time deviation standard used to determine samples with labor time deviations within an acceptable range. Only when the labor time deviation of a sample is within the threshold set by the user will the sample be considered qualified.
[0062] After defining the constraints, the accessory type list is used as the first-level benchmark accessory type list, and the work time list is used as the first-level benchmark work time list, and sample retrieval is started based on this. First, all the first-quantity second-level samples that meet the above constraints are retrieved. The first-quantity second-level samples are a set of samples that meet the accessory type and work time constraints, representing the layer of data closest to the current case.
[0063] Based on the first number of second-level samples, the first number of third-level samples that meet the constraints are further retrieved. This process will go deeper layer by layer until the first number of M-level samples that meet the conditions are retrieved, where the range of M is usually set between 5 and 20. Each level of samples is screened and filtered based on the benchmark conditions of the previous layer to ensure accurate matching of sample data layer by layer.
[0064] Finally, the first-level benchmark accessory type list and the first-level benchmark working hours list are used as root nodes, and all retrieved samples are traversed, from the first number of second-level samples to the first number of M-level samples, to build a sample distribution tree layer by layer. The sample distribution tree is a structured tree data structure, each node represents a sample or a group of samples, showing the hierarchical relationship of samples under different constraints. Through the above steps, the historical data most relevant to the current case can be accurately identified and organized, providing a solid data foundation for subsequent risk prediction, and ensuring the high accuracy and relevance of the prediction results.
[0065] Furthermore, the present application also includes the following steps:
[0066] Construct an aggregation function, the aggregation function is used to count the triggering frequency of risk project types; according to the aggregation function, aggregate the first number of M-1 level samples of the j-th M-1 level sample of the first number of M-1 level samples to extract the maximum value of the risk project type triggering frequency, and obtain the newly added risk project type of the j-th M-1 level sample, wherein the newly added risk project type of the j-th M-1 level sample has the triggering frequency label of the newly added risk project type of the j-th M-1 level sample; according to the triggering frequency label of the newly added risk project type of the j-th M-1 level sample and the newly added risk project type of the j-th M-1 level sample, update the j-th M-1 level sample of the first number of M-1 level samples; based on the above principle, aggregate the sample distribution tree from the bottom layer upwards, and output the risk project prediction result.
[0067] Specifically, the role of the aggregation function is to count the triggering frequency of risk project types. The triggering frequency refers to the number of times a certain risk project type is activated in different samples. Through the aggregation function, it is possible to calculate which risk project types are more likely to be triggered in samples at each level, and then evaluate the significance of these risks. According to the aggregation function, data aggregation is performed on samples at different levels. Assume that the system is currently processing the M-th level sample data. Starting from the jth M-1-level sample of the first number of M-1-level samples, aggregate the samples related to it, that is, count which risk project types have the highest triggering frequency in the first number of M-level samples contained in the jth M-1-level sample, extract the maximum value of the triggering frequency, that is, find out the risk project type that is triggered most in the first number of M-level samples. This risk project type with the highest frequency is called the newly added risk project type of the jth M-1-level sample, and a trigger frequency label is assigned to its configuration, that is, the trigger frequency label of the newly added risk project type of the jth M-1-level sample.
[0068] Further, according to the newly added risk project type trigger frequency label of the jth M-1 level sample and the newly added risk project type of the jth M-1 level sample, the jth M-1 level sample is updated to include the newly added risk project type trigger frequency label of the jth M-1 level sample and the newly added risk project type of the jth M-1 level sample, so as to ensure that the M-1 level sample can reflect the latest risk assessment results, thereby improving the accuracy of the next prediction. According to the above steps, the data is aggregated upward from the bottom layer of the sample distribution tree, and data integration and analysis are performed layer by layer. The data aggregation of each layer will be based on the aggregation results of the previous layer, so as to gradually and progressively perform risk assessment. Finally, the risk project prediction result is output at the highest level. This result will integrate the data of samples at all levels to provide a comprehensive risk assessment. Through this layer-by-layer aggregation and frequency analysis method, the most important risk project types can be effectively identified and extracted, and an accurate risk project prediction result can be output based on these data, thereby improving the accuracy and reliability of risk assessment.
[0069] Furthermore, step six of this application includes:
[0070] According to the risk project prediction result, matching is performed on the multi-cluster rules based on the multiple risk project type identifiers to obtain a first rule set; the intermediate leaf node of the rule factor tree is executed according to the target object type, and the bottom leaf node is matched, wherein the bottom leaf node has a second rule set; the intersection of the first rule set and the second rule set is taken to obtain the selected rule set.
[0071] Specifically, according to the generated risk project prediction results, matching is performed between multiple risk project type identifiers and multiple cluster rules to identify which rules match the risk projects in the risk project prediction results, and all rules related to the risk project prediction results are screened out to form a first rule set.
[0072] Next, further matching is performed in the constructed rule factor tree according to the target type. The nodes in the rule factor tree represent different rule factor types, which are divided into intermediate leaf nodes and bottom leaf nodes. The intermediate leaf nodes are executed first, which usually represent broader conditions. Then, based on the matching results of the intermediate leaf nodes, the bottom leaf nodes are further checked. The bottom leaf nodes represent more specific and detailed rule conditions. In this process, all bottom leaf node rules that meet the current target type are screened out and these rules are combined into the second rule set.
[0073] Finally, the first rule set is intersected with the second rule set, that is, the common parts of the two rule sets are extracted, that is, the rules that meet both the risk project type and the subject matter type. Through this intersection operation, the scope of rules that need to be executed can be effectively narrowed, and only those rules that are highly relevant to the case are retained to form the final selected rule set. Through the dual matching and intersection operations based on the risk project prediction results and the rule factor tree, the most relevant rules can be accurately identified and screened to ensure that only those rules that are truly needed are executed when handling cases. This not only improves the efficiency of rule execution, but also greatly reduces the waste of system resources, making the entire risk management process more accurate and efficient.
[0074] In summary, the method for filtering insurance industry risk control rules and accelerating execution provided by this application has the following technical effects:
[0075] A rule set is obtained, wherein the rule set has a risk project label and a rule factor label; according to the risk project label, the rule set is clustered and grouped to obtain multiple cluster rules and multiple risk project type identifiers; according to the rule factor label, the rule set is aggregated to obtain a rule factor tree, wherein the rule factor tree includes an intermediate leaf node and a bottom leaf node, the intermediate leaf node is a node with child nodes, the bottom leaf node is a node without child nodes, and the intermediate leaf node has a rule factor type identifier; in response to the pending case uploaded by the user-side interface, case damage assessment information is obtained, wherein the case damage assessment information includes a list of accessory types, a list of working hours, and a subject matter type; risk project prediction is performed according to the list of accessory types, the list of working hours, and the subject matter type to obtain a risk project prediction result; based on the risk project prediction result, rule statistics are performed according to the multiple risk project type identifiers and the rule factor type identifier to obtain a selected rule set; case detection of the pending case is performed according to the selected rule set. By obtaining a rule set containing risk item labels and rule factor labels, building a rule factor tree, and combining the case loss assessment information to predict risk items, accurate rule statistics and matching can be performed, and finally the selected rule set can be obtained and case detection can be performed, thereby achieving the technical effect of improving case handling efficiency, enhancing rule execution accuracy, and effectively identifying and preventing potential risks.
[0076] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
[0077] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.
Claims
1. A method for filtering insurance industry risk control rules to accelerate execution, characterized in that: include: Obtaining a rule set, wherein the rule set has a risk item label and a rule factor label; Clustering and grouping the rule set according to the risk project label to obtain multiple cluster rules and multiple risk project type identifiers; According to the rule factor label, the rule set is aggregated to obtain a rule factor tree, wherein the rule factor tree includes an intermediate leaf node and a bottom leaf node, the intermediate leaf node is a node with child nodes, the bottom leaf node is a node without child nodes, and the intermediate leaf node has a rule factor type identifier; In response to the pending cases uploaded on the user interface, obtaining the case damage assessment sheet information, wherein the case damage assessment sheet information includes a list of accessory types, a list of working hours, and a type of subject matter; Predicting risk items according to the accessory type list, the work hour list and the subject matter type to obtain risk item prediction results; Based on the risk project prediction result, rule statistics are performed according to the multiple risk project type identifiers and the rule factor type identifiers to obtain a selected rule set; Perform case detection of the case to be executed according to the selected rule set; According to the rule factor label, the rule set is aggregated to obtain a rule factor tree, including: Clustering and grouping the rule set according to the rule factor label to obtain multiple groups of rule sets and multiple rule factor type identifiers; According to the number of multiple rules in the multiple groups of rule sets, the multiple rule factor type identifiers are sorted from large to small to obtain a rule factor type identifier sorting result; Traversing the sorting results of the rule factor type identifiers, starting from the root node, constructing the rule factor tree trunk; Distributing the multiple groups of rule sets on the trunk of the rule factor tree to construct branch nodes, thereby obtaining the rule factor tree; The rule factor tree includes an intermediate leaf node and a bottom leaf node, the intermediate leaf node is a node with child nodes, the bottom leaf node is a node without child nodes, and the intermediate leaf node has a rule factor type identifier; Traverse the sorting results of the rule factor type identifiers, and build the rule factor tree trunk starting from the root node, including: According to the sorting result of the rule factor type identifier, obtain the first sequence number rule factor type identifier, the second sequence number rule factor type identifier, and so on until the Nth sequence number rule factor type identifier; Setting the first sequence rule factor type identifier as a root node; Linking the second sequence number rule factor type identifier to the root node to set it as a secondary backbone node; Until the Nth order rule factor type identifier is linked to the N-1 level trunk node and set as the N level trunk node, the rule factor tree trunk is generated.
2. A method for filtering insurance industry risk control rules to accelerate execution as claimed in claim 1, characterized in that: According to the risk project label, the rule set is clustered and grouped to obtain multiple cluster rules and multiple risk project type identifiers, including: Obtaining a first rule of the rule set, wherein the first rule stores a first risk project tag in the risk project tag; Obtaining a second rule of the rule set, wherein the second rule stores a second risk project tag in the risk project tag; When the first risk item label and the second risk item label are the same, adding the first rule and the second rule into the same cluster; When the first risk item label and the second risk item label are different, adding the first rule and the second rule into a different cluster; Until all rule sets are traversed and analyzed, the multiple cluster rules and the multiple risk project type identifiers are output.
3. A method for filtering insurance industry risk control rules to accelerate execution as claimed in claim 1, characterized in that: Distributing the multiple groups of rule sets on the trunk of the rule factor tree to construct branch nodes to obtain the rule factor tree includes: Step a: Obtain a set of k-level backbone node rules; Step b: Divide the rules triggered when the k-level backbone node rule factor is met into a group and store them in the first leaf node of the k-level backbone node; Step c: The rules triggered when the rule factor of the k-level trunk node is not satisfied are divided into a group and stored in the second leaf node of the k-level trunk node; Based on the execution process of the above step a, step b and step c, the rule factor tree trunk is traversed to construct branch nodes to obtain the rule factor tree.
4. A method for accelerating the execution of insurance industry risk control rules by filtering according to claim 1, characterized in that: The risk project prediction is performed according to the accessory type list, the working time list and the subject matter type to obtain the risk project prediction result, including: According to the type of the subject matter, a target risk project prediction model is selected from a risk project prediction model library; The target risk project prediction model includes a sample retrieval layer, and the sample retrieval layer is used to collect a sample distribution tree with the accessory type list and the working hour list as constraints; The target risk project prediction model includes a sample aggregation layer, and the sample aggregation layer is used to aggregate the sample distribution tree from the bottom layer upwards to output the risk project prediction result.
5. A method for filtering insurance industry risk control rules to accelerate execution as claimed in claim 4, characterized in that: Taking the accessory type list and the working hour list as constraints, a sample distribution tree is collected, including: Constructing a constraint condition, wherein the constraint condition comprises a first constraint function and a first constraint function threshold, a second constraint function and a second constraint function threshold; The first constraint function is the ratio of the intersection of the sample and the accessory type list to the union of the accessory types, and the threshold of the first constraint function is equal to 1; The second constraint function is a sample working time list of a sample accessory type list and a mean deviation of working time of accessories of the same type in a benchmark working time list of a benchmark accessory type list, and the second constraint function threshold is a user-defined working time deviation threshold; Taking the accessory type list as the first-level reference accessory type list and the working hour list as the first-level reference working hour list, a first number of second-level samples satisfying the constraint condition are retrieved; constructing a benchmark condition with the first number of second-level samples, and retrieving the first number of third-level samples that meet the constraint condition; Until the benchmark condition is established with the first number M-1 level samples, the first number M level samples satisfying the constraint condition are retrieved, where 20≥M≥5; The sample distribution tree is constructed by taking the first-level benchmark accessory type list and the first-level benchmark working hour list as root nodes and traversing the first number of second-level samples, the first number of third-level samples, and finally the first number of M-level samples.
6. A method for filtering insurance industry risk control rules to accelerate execution as claimed in claim 4, characterized in that: The sample aggregation layer is used to aggregate the sample distribution tree from the bottom layer upwards and output the risk project prediction result, including: Constructing an aggregation function, wherein the aggregation function is used to count the triggering frequency of risk project types; According to the aggregation function, the first number of M-level samples of the j-th M-1-level sample of the first number of M-1-level samples are aggregated to extract the maximum value of the risk project type trigger frequency, and the newly added risk project type of the j-th M-1-level sample is obtained, wherein the newly added risk project type of the j-th M-1-level sample has the newly added risk project type trigger frequency label of the j-th M-1-level sample; According to the j-th M-1-level sample newly added risk item type trigger frequency label and the j-th M-1-level sample newly added risk item type, updating the j-th M-1-level sample of the first number of M-1-level samples; Based on the above principle, the sample distribution tree is aggregated upward from the bottom layer to output the risk project prediction result.
7. A method for accelerating the execution of insurance industry risk control rules by filtering according to claim 1, characterized in that: Based on the risk project prediction result, rule statistics are performed according to the multiple risk project type identifiers and the rule factor type identifiers to obtain a selected rule set, including: According to the risk project prediction result, matching is performed on the multiple cluster rules based on the multiple risk project type identifiers to obtain a first rule set; Executing the intermediate leaf node of the rule factor tree according to the subject matter type, matching the bottom leaf node, wherein the bottom leaf node has a second rule set; The selected rule set is obtained by taking the intersection of the first rule set and the second rule set.
