Method, system and device for tracing artificial intelligence model results based on expert rules

Abstract

The present invention discloses a method and system for tracing the results of an artificial intelligence model based on expert rules, which is used to solve the problem of lack of explainability of the prediction results of the artificial intelligence model. The present invention first collects high-quality historical data of auto insurance cases from insurance companies, and converts the historical data of auto insurance cases into graph structure data according to expert rules; then constructs a graph network through the graph structure data, and obtains expert factor vectors and expert rule vectors based on network representation learning model training; then collects real-time data of auto insurance that are determined to be risk cases by the artificial intelligence model, calculates the set of triggered expert factors, and obtains the quasi-triggering rule vector by default filling; then calculates the similarity between the expert rule vector and the quasi-triggering rule vector; finally, by returning the quasi-triggering rules with high similarity, provides traceability for auto insurance experts. The method of the present invention ensures a low traceability time while also ensuring the quality of traceability.

Description

Technical Field

[0001] The present invention belongs to the field of auto insurance fraud identification, and in particular relates to a method, system, and device for tracing the root cause of prediction results of an artificial intelligence model based on expert rules. Background Art

[0002] With the continued increase in auto insurance fraud cases in recent years and the rising costs of claims, more and more insurance companies have become aware of the dangers of auto insurance fraud and have implemented a series of measures to prevent and combat false auto insurance claims. Currently, anti-fraud efforts in auto insurance have made considerable progress, with expert systems and intelligent claims processing playing a crucial role.

[0003] Expert systems, as an important intelligent system, can leverage the experience and knowledge of auto insurance experts to make decisions, thereby determining the nature of cases and providing early warnings. The excellent interpretability of expert systems has led to their widespread application in auto insurance fraud prevention. However, because they are based on rules and reasoning, they struggle to produce useful results if insufficient conditions are met. Therefore, if the fraudster has some understanding of the rules, they can circumvent the expert system's detection by satisfying subtle conditions. On the other hand, big data-based artificial intelligence technology can overcome the limitations of traditional expert systems. By collecting and analyzing data on various aspects, such as the insured, the insured object, and the circumstances of the claim, it provides a clearer path for investigative work in auto insurance claims. While intelligent claims processing can play a significant role in identifying auto insurance fraud, it unfortunately lacks the interpretability of expert systems, making it difficult to provide a direct investigative entry point. Consequently, auto insurance investigators often spend a considerable amount of time and effort verifying such cases.

[0004] Therefore, how to provide reliable prediction results and trace the root causes for intelligent claims has become a major difficulty in anti-fraud work in auto insurance. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention discloses a method, system and device for tracing the root cause of prediction results of an artificial intelligence model based on expert rules.

[0006] In order to solve the above problems, the technical solution of the present invention is:

[0007] A first aspect of an embodiment of the present invention provides a method for tracing the root cause of an artificial intelligence model prediction result based on expert rules, specifically comprising the following steps:

[0008] Step S1: Collect historical data of auto insurance cases and build a database of historical auto insurance cases; extract risk factors and expert factors, and convert the historical auto insurance case data into graph structure data according to expert rules;

[0009] Step S2: Build a graph network using the graph structure data obtained in step S1, obtain expert factor vectors through network representation learning model training based on the graph network, form expert rule vectors by concatenating or averaging the expert factor vectors, and store the expert factor vectors and expert rule vectors in the rule vector database;

[0010] Step S3: Collect real-time auto insurance data that is determined by the AI ​​model to be a risky case. Extract risk factors and expert factors to obtain a set of triggered expert factors. Combine this with the expert rule vector obtained in step S2 and use default filling to obtain a proposed triggering rule vector.

[0011] Step S4: Calculate the similarity between the expert rule vector and the proposed trigger rule vector;

[0012] Step S5: Provide tracing for real-time car insurance cases based on the similarity results.

[0013] Furthermore, the step S1 generates graph structure data specifically as follows: first, historical data of motor vehicle insurance cases are collected to construct a motor vehicle insurance historical case database; then, fields associated with expert rules in the motor vehicle insurance historical case database are extracted to form a risk factor data set; then, expert factors are extracted, and each field in the risk factor data set is judged according to the factor composition in the expert rules, and the generated new fields constitute an expert factor data set; finally, the expert factor data set is converted into graph structure data.

[0014] Furthermore, the graph structure data is a triple, an edge table or an adjacency matrix.

[0015] Furthermore, the expert rule vector in step S2 may be calculated by using a method such as an autoencoder or nesting method that can convert discrete variables into continuous vector representations.

[0016] Furthermore, the default filling in step S3 specifically includes filling the untriggered expert factors in the intended triggering rule with default values, and the default value is usually a zero vector.

[0017] Furthermore, the step S4 uses the COS cosine theorem to calculate the similarity.

[0018] A second aspect of an embodiment of the present invention provides an artificial intelligence model result traceability system based on expert rules, the system comprising the following units:

[0019] The motor vehicle insurance historical case database is used to store motor vehicle insurance case historical data;

[0020] The risk factor extraction unit forms a risk factor dataset by finding the internal / external connections between expert rules and historical data in the motor vehicle insurance historical case database;

[0021] The expert factor extraction unit judges and processes the risk factor dataset according to the expert factors in the expert rules, obtains new fields guided by the expert factors, and constitutes the expert factor dataset;

[0022] The graph structure data generation unit is used to convert the expert factor dataset into a data format suitable for network representation learning input;

[0023] The network representation learning training unit uses unsupervised or self-supervised graph representation learning methods to train data and obtain expert factor vectors;

[0024] The rule vector database is used to store expert factor vectors, which are concatenated to obtain expert rule vectors, and the default filling method is used to obtain the quasi-trigger rule vectors;

[0025] The rule vector similarity calculation unit is used to calculate the similarity between the expert rule vector and the intended trigger rule vector;

[0026] The decision-making unit provides a traceability for the case based on the similarity between the expert rule vector and the proposed triggering rule vector;

[0027] A third aspect of an embodiment of the present invention provides an artificial intelligence model result tracing device based on expert rules, comprising one or more processors for implementing the above-mentioned artificial intelligence model result tracing method based on expert rules.

[0028] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, it is used to implement the above-mentioned expert rule-based artificial intelligence model result tracing method.

[0029] The beneficial effects of this invention are: this method can assist AI models by providing additional information on possible fraud causes, providing auto insurance experts and investigators with case verification strategies, enabling them to conduct targeted case investigations and significantly improving investigative efficiency. This invention automatically and efficiently traces AI risk cases, and its traceability principle, based on expert rules, minimizes time consumption while ensuring high-quality tracing. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of the method for tracing the results of an artificial intelligence model based on expert rules of the present invention;

[0031] Figure 2 Schematic diagram of the artificial intelligence model result tracing system based on expert rules of the present invention.

[0032] Figure 3 An example diagram of the graph network structure construction of the implementation case of the artificial intelligence model result traceability system based on expert rules of the present invention;

[0033] Figure 4 Schematic diagram of the artificial intelligence model result tracing device based on expert rules of the present invention. DETAILED DESCRIPTION

[0034] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0035] The terms used in this invention are for the purpose of describing specific embodiments only and are not intended to limit the invention. The singular forms "a," "the," and "the" used in this invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0036] It should be understood that although the terms "first," "second," "third," etc. may be used in the present invention to describe various information, such information should not be limited to these terms. These terms are merely used to distinguish information of the same type from one another. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the present invention. Depending on the context, the term "if" as used herein may be interpreted as "when," "when," or "in response to determining."

[0037] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the embodiments of the present invention and the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0038] The present invention is a method for tracing the results of artificial intelligence models based on expert rules (such as Figure 1 ), comprising the following steps:

[0039] Step S1: Collect historical data of auto insurance cases from insurance companies and build a database of auto insurance cases. Convert the historical data of auto insurance cases into a graph structure data format according to expert rules. The specific steps are as follows:

[0040] Step S101: Construct a motor vehicle insurance case history database based on the motor vehicle insurance case history data; wherein the motor vehicle insurance case history data includes but is not limited to underwriting, claim, reporting, investigation, damage assessment, price and damage assessment data, etc. In order to ensure the reliability of backtracking, the data should mostly be data that triggers expert rules; the motor vehicle insurance case history database must meet the requirements of incremental updates, that is, as time goes by, new data will continue to enter the motor vehicle insurance case history database.

[0041] Step S102: By finding the internal / external connection between the expert rules and the original fields in the motor vehicle insurance case history database, the field data associated with the expert rules are extracted to form a risk factor data set; wherein the expert rules are derived from the experience summary of motor vehicle insurance experts on motor vehicle insurance fraud cases, and an expert rule is often composed of one or more expert factors. The expert rule will only be triggered when all expert factors are triggered; wherein the internal / external connection between the expert rules and the original fields in the motor vehicle insurance case history database needs to be constructed based on the conditional factors of the expert rules. If the calculation of the conditional factors requires complex judgments involving one or more original fields (for example, the calculation of the distance between the accident location and the repair shop), it is said that there is an internal connection; if the calculation of the conditional factors requires ordinary judgments involving one or more original fields (for example, whether it is a single-vehicle accident), it is said that there is an external connection; wherein the risk factor data set should be a set of field data associated with the expert rules in the motor vehicle insurance case history database, usually a data subset from the motor vehicle insurance case history database, and also includes a new data set that undergoes preliminary data processing on this basis.

[0042] Step S103: Perform corresponding judgment operations on each field in the risk factor dataset according to the expert factor composition in the expert rule, and obtain new fields guided by the expert factors, thereby forming an expert factor dataset; wherein the expert factor is a component of the expert rule, and one expert factor corresponds to one discrimination condition. The expert rule will be triggered only when all expert factors are triggered; wherein the expert factor dataset refers to processing each field in the risk factor dataset according to the discrimination condition of the expert factor, thereby obtaining a new dataset with the expert factor as the field.

[0043] Step S104: Convert the expert factor dataset into graph structure data suitable for network representation learning. The graph structure data must be in a data format that complies with network representation learning, including but not limited to triples, edge tables, and adjacency matrices.

[0044] Step S2: Use the graph structure data obtained in step S1 to construct a graph network, obtain expert factor vectors through network representation learning model training based on the graph network, obtain expert rule vectors by splicing (or averaging) the expert factor vectors, and store the expert factor vectors and expert rule vectors in the rule vector database; wherein splicing is a broad description, and any method of outputting one or more vectors into a fixed-dimensional vector through some combination or calculation is applicable, and specific methods include but are not limited to vector splicing, vector averaging, etc.; wherein the rule vector database is used to store expert factor vectors and expert rule vectors. In order to ensure the validity of the rule vector database, the expert factor vectors and expert rule vectors will be regularly trained and updated as the motor vehicle insurance historical case database is updated.

[0045] Step S3: Collect real-time auto insurance data from insurance companies that are identified as risky cases by the AI ​​model to obtain a set of triggered expert factors. Combined with the pre-built rule vector database in step 2, the proposed triggering rule vector is obtained by comparing it with the expert rules and using a default filling method. The specific steps are as follows:

[0046] Step S301: Analyze the real-time auto insurance data that is determined to be a risk case by the artificial intelligence model, and calculate the triggered expert factor set based on steps S102 and S103; wherein the analysis refers to converting the data format of the real-time auto insurance data into a format that conforms to the format of the historical data of auto insurance cases. If the data formats of the two are the same, this process is not required.

[0047] Step S302: Based on the expert factor vector obtained by pre-training the network representation learning model in step S2, a default filling method is used to obtain a vector of quasi-triggering rules; wherein, the default filling refers to filling the default value of the untriggered expert factors in the quasi-triggering rules, and the default value is usually a zero vector; wherein the quasi-triggering rules refer to all untriggered expert rules.

[0048] Step S4: Calculate the similarity between the expert rule vector and the proposed trigger rule vector obtained in step S3; wherein the similarity calculation criteria include but are not limited to various vector similarity calculation methods such as COS cosine theorem.

[0049] Step S5: Based on the similarity between the expert rule vector and the proposed trigger rule vector, provide tracing for the auto insurance expert; wherein the tracing method includes but is not limited to returning the proposed trigger rule with the highest similarity, returning the proposed trigger rule based on threshold screening, etc., which can provide a basis for real-time auto insurance data.

[0050] The present invention also proposes an artificial intelligence model result traceability system based on expert rules (such as Figure 2As shown), it includes: a motor vehicle insurance historical case database, a risk factor extraction unit, an expert factor extraction unit, a graph structure data generation unit, a network representation learning and training unit, a rule vector database, a rule vector similarity calculation unit and a decision unit.

[0051] The motor vehicle insurance case history database is used to store motor vehicle insurance case history data and supports incremental updates of data, that is, as time goes by, new data will continue to enter the motor vehicle insurance case history database.

[0052] The risk factor extraction unit finds the intrinsic / extrinsic connection between the expert rules and the original fields corresponding to the historical data in the motor vehicle insurance historical case database, extracts the field data associated with the expert rules, and forms a risk factor data set.

[0053] The expert factor extraction unit performs corresponding judgment processing on the risk factor dataset extracted by the risk factor extraction unit according to the expert factors in the expert rules, obtains a new field guided by the expert factors, and forms an expert factor dataset.

[0054] The graph structure data generating unit is used to convert the expert factor data set into a triple, an edge table, an adjacency matrix and other data formats suitable for the input required by the network representation learning and training unit.

[0055] The network representation learning training unit adopts an unsupervised or self-supervised graph representation learning method to train data and obtain an expert factor vector.

[0056] The rule vector database stores the training results of the network representation learning unit, including expert factor vectors. Expert rule vectors are formed by concatenating the expert factor vectors, and the proposed trigger rule vector is obtained using a default filling method. To ensure data validity, the rule vector database is periodically updated based on incremental updates to the historical auto insurance case database. Furthermore, the rule vector database supports incremental data updates to accommodate changes in expert rules.

[0057] The rule vector similarity calculation unit is used to calculate the similarity between the expert rule vector output by the rule vector database and the intended trigger rule vector.

[0058] The decision-making unit provides causation for the case based on the similarity between the expert rule vector and the proposed triggering rule vector.

[0059] Example 1

[0060] The present invention is an artificial intelligence model result tracing method based on expert rules, which includes the following five steps:

[0061] Step (1): Collect historical data of auto insurance cases and convert them into structured graph data according to expert rules. The specific steps are as follows:

[0062] (1.1) Risk factor dataset extraction.

[0063] Historical auto insurance case data encompasses all aspects of the process, from underwriting to claims settlement. However, expert rules, as a summary of professional experience, often focus only on certain anomalies in the data—specific field values. To facilitate subsequent work, it is crucial to extract fields associated with expert factors from historical auto insurance data based on established expert rules.

[0064] This example uses 10,000 pieces of historical auto insurance case data as a data sample. To illustrate the implementation of the solution in detail, the following uses one expert rule, two historical auto insurance cases, and one real-time auto insurance case as specific examples for description:

[0065] Rule 1: Single vehicle accident + subject + rollover + estimated damage amount [30000,] + subject vehicle age _ years [7,] + ! Personal injury → false rollover

[0066] First data:

[0067] {

[0068] Case number: First case number

[0069] Is it a target: Yes

[0070] Accident Type: Bicycle Accident

[0071] Cause of accident: Capsize

[0072] Estimated loss amount: 83,800

[0073] Age: 8

[0074] Cost name: Three-yuan accessories

[0075] Cost name: Exterior taillights

[0076] …

[0077] }

[0078] Second data:

[0079] {

[0080] Case No.: Second Case No.

[0081] Is it a target: Yes

[0082] Accident Type: Two-vehicle accident

[0083] Cause of accident: Capsize

[0084] Estimated loss amount: 45,000

[0085] Age: 10

[0086] Cost Name: Headlights

[0087] Cost name: Sheet metal painting

[0088] …

[0089] }

[0090] Taking the first rule as an example, by comparing the first data with the second data, it can be found that it contains 5 fields involving common judgments of expert factors (accident cause: associated with the expert factor single-vehicle accident, whether it is a target: associated with the expert factor target, accident type: associated with the expert factor overturning, estimated loss amount: associated with the expert factor exceeding estimated loss amount_30000, vehicle age: associated with the expert factor exceeding vehicle age_7), and 1 field involving complex judgments of expert factors (expense name: corresponding to the expert factor non-personal injury, whether the expense name contains personal medical expenses is judged to contain personal injury). The risk factor dataset consists of these fields.

[0091] (1.2) Expert factor dataset extraction.

[0092] After obtaining the risk factor dataset, each field in the risk factor dataset needs to be processed according to the discrimination conditions of the expert factors, thereby obtaining a new dataset with the expert factors as fields. Taking the second data as an example, the new data formed is as follows:

[0093] {

[0094] Case No.: Second Case No.

[0095] Target: Yes

[0096] Bicycle accident: No

[0097] Capsize: Yes

[0098] Exceeding estimated loss amount_30000: Yes

[0099] Over vehicle age_7: Yes

[0100] Non-human injuries: Yes

[0101] }

[0102] (1.3) Graph structure data conversion.

[0103] In this example, the case number and expert factor are used as nodes, and the graph structure data is stored in the form of an edgelist. Taking the second data as an example, the items with the true value of the expert factor field will be stored in the edgelist data. In this case, the edgelist data should contain the following items:

[0104] (Second Case No., Subject Matter)

[0105] (Case No. 2, overturning)

[0106] (Second case number, exceeding estimated loss amount_30000)

[0107] (Second case number, vehicle age exceeded 7)

[0108] (Case No. 2, non-human injury)

[0109] Among them, the value of the single-vehicle accident field is no, that is, the second data does not trigger this factor, and therefore is not included in the edge table item.

[0110] Step (2): Build a network through graph structure data, obtain expert factor vectors based on network representation learning model training, and then obtain expert rule vectors. In this embodiment, 10,000 graph data are used to build the network. Taking the first data, the second data and the first rule as an example, Figure 3 The construction form of the graph network structure is shown in .

[0111] As attached Figure 3 As shown, the upper portion of the dotted line represents the case number node, and the lower portion represents the expert factor node. In this embodiment, deep walk training of graph structure data is used to obtain expert factor vectors, and vector averaging is used as the calculation method for expert rule vectors. Taking the first rule as an example, the following is its expert factor vector and the expert rule vector obtained by vector averaging:

[0112] Expert Factor Vector value Bicycle Accident (0.12,0.05,…,0.77) Target (0.24,0.35,…,0.99) Capsize (0.35,0.45,…,0.16) Exceeding estimated loss amount_30000 (0.72,0.32,…,0.27) Over 7 years old (0.93,0.48,…,0.47) Non-human injuries (0.64,0.35,…,0.67)

[0113]

[0114] The expert factor vector and expert rule vector will be stored in the rule vector database for subsequent calculations.

[0115] Step (3): Collect real-time data of auto insurance cases that are identified as risky cases by the artificial intelligence model and calculate the vector of the rules to be triggered. The specific steps are as follows:

[0116] (3.1) Calculate the set of expert factors triggered by real-time auto insurance data.

[0117] The third data of the existing real-time car insurance case:

[0118] {

[0119] Case No.: Case No. 3

[0120] Is it a target: Yes

[0121] Accident Type: Bicycle Accident

[0122] Cause of the accident: Collision

[0123] Estimated loss amount: 83,800

[0124] Age: 12

[0125] Fee name: Back

[0126] Cost Name: Trunk Cover

[0127] …

[0128] }

[0129] By comparing with the first rule, the activated expert factor set (single vehicle accident, target, exceeding a specific vehicle age_1, exceeding estimated loss amount_1, non-personnel injury) is obtained through step (1).

[0130] (3.2) Query the rule vector database and obtain the proposed trigger rule vector by default filling.

[0131] After obtaining the set of triggered expert factors, the corresponding vector values ​​are retrieved by querying the rule vector database. The proposed triggering rule vector is then calculated by vector concatenation or averaging. In this example, vector averaging is used as the calculation method for the proposed triggering rule vector. Since the expert factor overturning is not triggered, the default filling rule requires a default value, that is, a zero vector.

[0132]

[0133] Step (4): Calculate the expert rule vector (v1, v2, ..., v n ) and the quasi-triggering rule vector (v1′,v2′,...,v n ′).

[0134] In this embodiment, the cosine theorem is used to calculate the similarity of vectors. The calculation formula is as follows:

[0135]

[0136] Calculate the similarity between the first expert rule and the first proposed trigger rule, where v1, v2, ..., v n Represent the values ​​of the first, second to nth dimensions of the expert rule vector, v1′, v2′, ..., v n ′ represents the values ​​of the first, second to nth dimensions of the proposed trigger rule vector:

[0137] The existing expert rule 1 = (0.50, 0.33, ..., 0.56) and the proposed trigger rule 1 = (0.44, 0.26, ..., 0.53), then according to the above formula, the similarity between the two is:

[0138]

[0139] Step (5): Provide tracing back to the auto insurance expert by finding the most similar proposed triggering rule or returning the highly similar proposed triggering vector based on threshold screening. In this example, the threshold screening method is adopted, and the threshold is selected as 0.8. At this time, the first rule will be returned as the result of root cause tracing.

[0140] Corresponding to the aforementioned embodiment of the method for tracing the root cause of the prediction results of the artificial intelligence model based on expert rules, the present invention also provides an embodiment of the device for tracing the root cause of the prediction results of the artificial intelligence model based on expert rules.

[0141] See also Figure 4 An embodiment of the present invention provides a root cause tracing device for an artificial intelligence model prediction result based on expert rules, comprising one or more processors for implementing the root cause tracing method for an artificial intelligence model prediction result based on expert rules in the above embodiment.

[0142] The embodiment of the root cause tracing device of the prediction result of the artificial intelligence model based on expert rules of the present invention can be applied to any device with data processing capabilities, and the any device with data processing capabilities can be a device or apparatus such as a computer. The device embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a device in a logical sense, it is formed by the processor of any device with data processing capabilities in which it is located to read the corresponding computer program instructions in the non-volatile memory into the memory for execution. From the hardware level, if Figure 4 As shown in the figure, it is a hardware structure diagram of any device with data processing capability where the artificial intelligence model prediction result root cause tracing device based on expert rules of the present invention is located. Figure 4 In addition to the processor, memory, network interface, and non-volatile memory shown, any device with data processing capabilities in which the apparatus in the embodiment is located may also include other hardware, generally based on the actual functions of the device with data processing capabilities, which will not be described in detail.

[0143] The implementation process of the functions and effects of each unit in the above-mentioned device is specifically described in the implementation process of the corresponding steps in the above-mentioned method, and will not be repeated here.

[0144] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present invention. A person of ordinary skill in the art can understand and implement the present invention without inventive work.

[0145] An embodiment of the present invention also provides a computer-readable storage medium having a program stored thereon. When the program is executed by a processor, the root cause tracing method of the prediction result of the artificial intelligence model based on expert rules in the above embodiment is implemented.

[0146] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the aforementioned embodiments, such as a hard disk or memory. The computer-readable storage medium may also be any device with data processing capabilities, such as a plug-in hard disk, a smart media card (SMC), an SD card, a flash card, etc. equipped on the device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of any device with data processing capabilities and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store data that has been output or is to be output.

[0147] The above examples only use graph network representation learning as an implementation method. Other methods suitable for converting expert rules and expert factors into vector representations, such as autoencoders and embedding, also fall within the scope of the present invention. Therefore, any equivalent changes or modifications based on the principles and design concepts disclosed in this invention are within the scope of protection of this invention.

Claims

1. A method for tracing the results of an artificial intelligence model based on expert rules, characterized in that: The following steps are involved: Step S1: Collect historical data of auto insurance cases and build a database of historical auto insurance cases; Extract risk factors and expert factors, and convert historical data of auto insurance cases into graph-structured data based on expert rules; Step S1 includes: first collecting historical data of auto insurance cases to build a database of auto insurance cases; then extracting fields associated with expert rules from the database of auto insurance cases to form a risk factor dataset; then extracting expert factors, judging each field in the risk factor dataset according to the factor composition in the expert rules, and generating new fields to form an expert factor dataset; finally, converting the expert factor dataset into graph structure data; The process of obtaining the expert factor dataset includes: Performing corresponding judgment operations on each field in the risk factor dataset based on the expert factor composition in the expert rule to obtain new fields guided by the expert factors, thereby forming an expert factor dataset; wherein the expert factor is a component of the expert rule, and each expert factor corresponds to a discrimination condition, and the expert rule is triggered only when all expert factors are triggered; wherein the expert factor dataset refers to a new dataset with expert factors as fields obtained by processing each field in the risk factor dataset based on the discrimination condition of the expert factor; Step S2: Build a graph network using the graph structure data obtained in step S1, obtain expert factor vectors through network representation learning model training based on the graph network, form expert rule vectors by concatenating or averaging the expert factor vectors, and store the expert factor vectors and expert rule vectors in the rule vector database; Step S3: Collect real-time auto insurance data that is determined by the AI ​​model to be a risky case. Extract risk factors and expert factors to obtain a set of triggered expert factors. Combine this with the expert rule vector obtained in step S2 and use default filling to obtain a proposed triggering rule vector. Step S4: Calculate the similarity between the expert rule vector and the proposed trigger rule vector; Step S5: Provide tracing for real-time car insurance cases based on the similarity results.

2. The method for tracing the results of an artificial intelligence model based on expert rules according to claim 1, characterized in that: The graph structure data is a triple, an edge table or an adjacency matrix.

3. The method for tracing the results of an artificial intelligence model based on expert rules according to claim 1, characterized in that: The calculation method of the expert rule vector in step S2 is replaced by: using a method including autoencoder or nesting that can convert discrete variables into continuous vector representation.

4. The method for tracing the results of an artificial intelligence model based on expert rules according to claim 1, characterized in that: The default filling in step S3 specifically includes filling the untriggered expert factors in the intended triggering rule with a default value, which is usually a zero vector.

5. The method for tracing the results of an artificial intelligence model based on expert rules according to claim 1, characterized in that: The step S4 uses the COS cosine theorem to calculate the similarity.

6. An artificial intelligence model result traceability system based on expert rules, characterized by: For implementing the method for tracing the results of an artificial intelligence model based on expert rules according to any one of claims 1 to 5, the system comprises the following units: The motor vehicle insurance historical case database is used to store motor vehicle insurance case historical data; The risk factor extraction unit forms a risk factor dataset by finding the internal / external connections between expert rules and historical data in the motor vehicle insurance historical case database; The expert factor extraction unit judges and processes the risk factor dataset according to the expert factors in the expert rules, obtains new fields guided by the expert factors, and constitutes the expert factor dataset; The graph structure data generation unit is used to convert the expert factor dataset into a data format suitable for network representation learning input; The network representation learning training unit uses unsupervised or self-supervised graph representation learning methods to train data and obtain expert factor vectors; The rule vector database is used to store expert factor vectors, which are concatenated to obtain expert rule vectors, and the default filling method is used to obtain the quasi-trigger rule vectors; The rule vector similarity calculation unit is used to calculate the similarity between the expert rule vector and the intended trigger rule vector; The decision-making unit provides causation for the case based on the similarity between the expert rule vector and the proposed triggering rule vector.

7. An artificial intelligence model result tracing device based on expert rules, characterized in that: It includes one or more processors for implementing the artificial intelligence model result tracing method based on expert rules as described in any one of claims 1-5.

8. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, it is used to implement the artificial intelligence model result tracing method based on expert rules as described in any one of claims 1 to 5.

Images