Risk control rule effect evaluation method and system based on artificial intelligence

Abstract

The invention discloses a risk control rule effect evaluation method and system based on artificial intelligence, and relates to the technical field of artificial intelligence, and the method comprises the steps: collecting structured data and unstructured data in vehicle insurance claim settlement, and carrying out the preprocessing; analyzing the processed unstructured data by using a deep learning technology, and extracting risk control feature vectors; based on the processed structured data and unstructured data, constructing a claim settlement relation graph; constructing a graph neural network model by using the risk control feature vector and the claim relation graph; setting a new risk control rule based on the output of the graph neural network model; evaluating the effect of the new risk control rule through the risk control rule test, and generating a new risk control rule effect evaluation report; according to the method, deep learning and graph neural network technologies are combined, multi-source data of car insurance claim settlement are comprehensively analyzed, the identification accuracy and efficiency of fraudulent behaviors are improved, the economic loss of insurance companies is reduced, and the user experience and service quality are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based risk control rule effect evaluation method and system. Background Art

[0002] With the rapid development of big data and artificial intelligence technology, the insurance industry is facing increasingly complex fraud risk challenges. Traditional auto insurance claims risk control methods mainly rely on manual review and rule-based systems. Although these methods are intuitive and effective, they are powerless when faced with large amounts of unstructured data. In recent years, deep learning technology has shown great potential in processing unstructured data, especially in image recognition, speech analysis, and natural language processing. At present, most risk control systems lack effective feature extraction methods when processing unstructured data, and often ignore some key information, resulting in inaccurate risk assessment. In addition, traditional risk control rules are mostly based on expert experience, lack flexibility and adaptability, and are difficult to cope with the ever-changing risk environment. Summary of the invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides an artificial intelligence-based risk control rule effectiveness evaluation method and system to solve the problem in the prior art of lack of effective integration of structured and unstructured data for risk control rule effectiveness evaluation in the field of motor insurance claims.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a risk control rule effect evaluation method based on artificial intelligence, which includes collecting structured data and unstructured data in auto insurance claims and performing preprocessing;

[0007] Use deep learning technology to analyze the processed unstructured data and extract risk control feature vectors;

[0008] Build a claims relationship map based on the processed structured and unstructured data;

[0009] Use risk control feature vectors and claims relationship graphs to build a graph neural network model;

[0010] Set new risk control rules based on the output of the graph neural network model;

[0011] Apply new risk control rules to new claims cases, evaluate the effectiveness of the new risk control rules through risk control rule testing, and generate a new risk control rule effectiveness evaluation report.

[0012] As a preferred solution of the risk control rule effect evaluation method based on artificial intelligence of the present invention, wherein: the structured data includes claim records and basic user information;

[0013] The unstructured data includes photos of the accident scene, customer phone recordings and text descriptions;

[0014] The preprocessing includes data cleaning and normalization.

[0015] As a preferred solution of the risk control rule effect evaluation method based on artificial intelligence described in the present invention, the following steps are included: using deep learning technology to analyze the processed unstructured data and extracting the risk control feature vector:

[0016] The image feature vector is extracted from the accident scene photos using the ResNet-50 network, expressed as,

[0017] x image =f ResNet-50 (I);

[0018] Among them, x image represents the image feature vector extracted from the penultimate layer of ResNet-50, f ResNet-50 represents the ResNet-50 network, and I represents the data format suitable for network input formed by the processed accident scene photos;

[0019] A recurrent neural network with a gated recurrent unit is used to extract a speech feature vector from a customer phone recording, represented as,

[0020] x voice =f GRU (V);

[0021] Among them, x voice represents the speech feature vector extracted from the last time step of GRU, V represents the data format suitable for network input formed by the processed customer phone recording, and f GRU represents a recurrent neural network with a gated recurrent unit;

[0022] Use the Word2Vec model to convert each word in the text description into a text feature vector representation, and then aggregate the word-level text feature vectors through the attention mechanism to obtain the sentence-level text feature vector, which is expressed as,

[0023] x word,i =f Word2Vec (w i );

[0024]

[0025] Among them, xword,i Represents the text feature vector of the current i-th word, z sentence represents the sentence-level text feature vector obtained by aggregation through the attention mechanism, f Word2Vec represents the Word2Vec model, M represents the number of words in the sentence, and w i represents the i-th word in the text description, α i Represents the attention weight of the current i-th word;

[0026] The extracted image feature vector, speech feature vector and sentence-level text feature vector are integrated to obtain the risk control feature vector.

[0027] As a preferred solution of the risk control rule effect evaluation method based on artificial intelligence described in the present invention, wherein: based on the processed structured data and unstructured data, a claims relationship map is constructed, including the following steps:

[0028] Use the BiLSTM-CRF-based named entity recognition model to identify entities from the processed text description and connect the identified entities with the entities in the structured data;

[0029] Define the relationship types between entities, use deep learning-based relationship extraction technology to extract the relationships between entities from structured and unstructured data, and verify the extracted relationships;

[0030] Using the Neo4j graph database, nodes in the graph are created based on the results of entity recognition. Each node represents an entity. Edges in the graph are created based on the results of relationship extraction. Each edge represents the relationship between entities. Attributes are added to the nodes and edges. The graph data is stored in a format supported by the graph database. Finally, a claims relationship graph is obtained.

[0031] As a preferred solution of the risk control rule effect evaluation method based on artificial intelligence described in the present invention, the risk control feature vector and the claim relationship map are used to construct a graph neural network model, including the following steps:

[0032] Integrate the risk control feature vector into the corresponding claim relationship graph node, and perform missing value processing, outlier detection and node attribute standardization on the integrated graph;

[0033] GCN is selected as the architecture of the graph neural network model. The connection relationship between nodes in the graph is set as the adjacency matrix. The graph data containing node features and edge information is input as the input feature matrix. The fraud probability of each node is predicted, and the graph neural network model is trained. The node features are updated through graph convolution operations, which are expressed as,

[0034]

[0035] Among them, H (l+1) represents the node feature matrix obtained after the l+1th layer of graph convolution operation, H (l) Represents the node feature matrix of the lth layer. The input feature matrix is ​​in the node feature matrix of the first layer. W (l) represents the weight matrix of the lth layer, represents the adjacency matrix with self-loops added, A represents the adjacency matrix, express The degree matrix of , σ represents the ReLU function;

[0036] Define loss function and optimizer;

[0037] The loss function and optimizer are used to minimize the loss and optimize the model parameters, and finally the constructed graph neural network model is obtained.

[0038] As a preferred solution of the risk control rule effect evaluation method based on artificial intelligence described in the present invention, wherein: based on the output of the graph neural network model, setting a new risk control rule includes the following steps:

[0039] Use the K-means clustering algorithm to cluster the fraud probabilities output by the graph neural network model to identify patterns with similar fraudulent behaviors;

[0040] According to the clustering results, set the clustering threshold;

[0041] When the clustering result is greater than or equal to the clustering threshold, it indicates that fraud exists;

[0042] When the clustering result is less than the clustering threshold, it indicates that there is no fraud.

[0043] As a preferred solution of the risk control rule effect evaluation method based on artificial intelligence described in the present invention, the new risk control rule is applied to a new claim case, the effect of the new risk control rule is evaluated through the risk control rule test, and a new risk control rule effect evaluation report is generated, including the following steps:

[0044] For new claims cases, use new risk control rules to make judgments;

[0045] Risk control rule testing refers to calculating the accuracy, recall, precision, and F1 score of the graph neural network model to evaluate the effectiveness of the risk control rules;

[0046] By summarizing the analysis of accuracy, recall, precision and F1 score, the comprehensive performance of the new risk control rules is evaluated and a new risk control rule effectiveness evaluation report is generated.

[0047] In a second aspect, the present invention provides an artificial intelligence-based risk control rule effect evaluation system, including a data acquisition module, which is responsible for collecting structured data and unstructured data in auto insurance claims and performing preprocessing;

[0048] The unstructured data deep learning analysis module is responsible for analyzing the processed unstructured data using deep learning technology and extracting risk control feature vectors;

[0049] The claims relationship graph construction module is responsible for constructing the claims relationship graph based on the processed structured data and unstructured data;

[0050] The graph neural network model construction module is responsible for building a graph neural network model using risk control feature vectors and claims relationship graphs;

[0051] The risk control rule setting and optimization module is responsible for setting new risk control rules based on the output of the graph neural network model;

[0052] The risk control rule effectiveness evaluation module is responsible for applying new risk control rules to new claims cases, evaluating the effectiveness of new risk control rules through risk control rule testing, and generating a new risk control rule effectiveness evaluation report.

[0053] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the risk control rule effectiveness evaluation method based on artificial intelligence as described in the first aspect of the present invention is implemented.

[0054] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the risk control rule effect evaluation method based on artificial intelligence as described in the first aspect of the present invention.

[0055] The beneficial effects of the present invention are as follows: through data collection and preprocessing, the effective integration and standardization of structured data and unstructured data in auto insurance claims are achieved, ensuring the accuracy and reliability of the subsequent analysis process. Through unstructured data feature extraction, key features are extracted from different types of unstructured data, providing a rich source of information for the subsequent establishment of a graph neural network model. By constructing a claims relationship graph, a graph that comprehensively reflects the various entities and their relationships in the claims process is constructed, providing an intuitive visualization tool for risk assessment and improving the basis for risk control decision-making. By constructing a graph neural network model, an automated risk prediction function is realized, the accuracy of fraud detection is improved, and the risk control process is optimized. By setting new risk control rules, specific risk control strategies are formulated according to model outputs, improving the efficiency of fraud detection. Through risk control rule testing and evaluation, a basis for continuous improvement is provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0057] Figure 1 This is a flow chart of the risk control rule effectiveness evaluation method based on artificial intelligence in Example 1.

[0058] Figure 2 The risk control feature vector in Example 1 is formed into a graph. DETAILED DESCRIPTION

[0059] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0061] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0062] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a risk control rule effect evaluation method based on artificial intelligence, comprising the following steps:

[0063] S1. Collect structured and unstructured data from auto insurance claims and perform preprocessing.

[0064] Structured data includes claims records and basic user information. Claim records include case number, reporting time, accident time, accident location, accident type and accident description. Basic user information includes basic information, identity information, address information, vehicle information and insurance information.

[0065] Unstructured data includes accident scene photos, customer phone recordings and text descriptions;

[0066] Preprocessing includes data cleaning and normalization;

[0067] Specifically, structured data such as claims records and basic user information are cleaned, missing values ​​are filled, and standardization is performed. Accident scene photos are preprocessed by size standardization and normalization. Customer phone recordings are preprocessed by sampling rate standardization and noise reduction. Text descriptions are preprocessed by word segmentation and stop word removal.

[0068] It should be noted that collecting structured and unstructured data in auto insurance claims and preprocessing them achieves effective data integration and standardization, then ensures the accuracy and reliability of the subsequent analysis process, and finally achieves the beneficial effect of improving data quality.

[0069] S2. Use deep learning technology to analyze the processed unstructured data and extract risk control feature vectors.

[0070] The image feature vector is extracted from the accident scene photos using the ResNet-50 network, expressed as,

[0071] x image =f ResNet-50 (I);

[0072] Among them, x image represents the image feature vector extracted from the penultimate layer of ResNet-50 (usually the layer before the global average pooling layer), f ResNet-50 ResNet-50 network is used to extract image features, and I represents the data format suitable for network input formed by the processed accident scene photos;

[0073] A recurrent neural network with a gated recurrent unit is used to extract a speech feature vector from a customer phone recording, represented as,

[0074] xvoice =f GRU (V);

[0075] Among them, x voice represents the speech feature vector extracted from the last time step of GRU, V represents the data format suitable for network input formed by the processed customer phone recording, and f GRU represents a recurrent neural network with a gated recurrent unit (GRU) for extracting speech features;

[0076] Use the Word2Vec model to convert each word in the text description into a text feature vector representation, and then aggregate the word-level text feature vectors through the attention mechanism to obtain the sentence-level text feature vector, which is expressed as,

[0077] x word,i =f Word2Vec (w i );

[0078]

[0079] Among them, x word,i Represents the text feature vector of the current i-th word, z sentence represents the sentence-level text feature vector obtained by aggregation through the attention mechanism, f Word2Vec represents the Word2Vec model, which is used to convert words into text feature vector representations. M represents the number of words in a sentence, and w i represents the i-th word in the text description, α i Represents the attention weight of the current i-th word;

[0080] The attention weight is expressed as,

[0081]

[0082] Among them, x i represents the feature vector of the current i-th word, h i Represents the hidden state vector of the current i-th word, W a Represents the weight matrix, which is used to adjust the linear combination of the feature vector and the hidden state vector, v a represents the weight vector used to calculate the linear combination of attention scores, x j represents the feature vector of the jth word, h j represents the hidden state vector of the jth word, where j represents the word index;

[0083] It should be noted that the importance of attention weight in this step lies in that it can highlight more important words in the sentence, thereby improving the model's sensitivity to key information. The beneficial effect is that it can more accurately capture the core meaning of the sentence and enhance the model's ability to understand the text content.

[0084] The extracted image feature vector, speech feature vector and sentence-level text feature vector are integrated to obtain the risk control feature vector.

[0085] It should be noted that deep learning technology is used to analyze the processed unstructured data and extract risk control feature vectors, thereby realizing the extraction of key features from different types of unstructured data, and then providing a rich source of information for the subsequent establishment of a graph neural network model, ultimately achieving the beneficial effect of enhancing the model's recognition capabilities.

[0086] S3. Build a claims relationship map based on the processed structured data and unstructured data.

[0087] Use the BiLSTM-CRF-based named entity recognition model to identify entities (such as user ID, license plate number, accident location, etc.) from the processed text description, and connect the identified entities with the entities in the structured data to ensure the consistency of the entities;

[0088] Define the relationship types between entities (e.g., the binding relationship between users and vehicles, the accident location and weather conditions in the claim record, etc.), use deep learning-based relationship extraction technology to extract the relationships between entities from structured and unstructured data, and verify the extracted relationships to ensure the accuracy and consistency of the relationships;

[0089] Using the Neo4j graph database, nodes in the graph are created based on the results of entity recognition. Each node represents an entity (such as a user, a vehicle, a claim record, etc.). Edges in the graph are created based on the results of relationship extraction. Each edge represents a relationship between entities (such as the binding relationship between a user and a vehicle, etc.). Attributes are added to nodes and edges (such as user age, vehicle model, claim amount, etc.). The graph data is stored in a format supported by the graph database, and finally a claim relationship graph is obtained.

[0090] It should be noted that based on the processed structured data and unstructured data, a claims relationship map is constructed, which realizes the construction of a map that comprehensively reflects the various entities and their relationships in the claims process, and then provides an intuitive visualization tool for risk assessment, and finally achieves the beneficial effect of improving the basis for risk control decision-making.

[0091] S4. Use risk control feature vectors and claims relationship graph to build a graph neural network model.

[0092] Integrate the risk control feature vector into the corresponding claim relationship graph node, and perform missing value processing, outlier detection and node attribute standardization on the integrated graph;

[0093] GCN is selected as the architecture of the graph neural network model. The connection relationship between nodes in the graph is set as the adjacency matrix. The graph data containing node features and edge information is input as the input feature matrix. The fraud probability of each node is predicted, and the graph neural network model is trained. The node features are updated through graph convolution operations, which are expressed as,

[0094]

[0095] Among them, H (l+1) represents the node feature matrix obtained after the l+1th layer of graph convolution operation, H (l) Represents the node feature matrix of the lth layer. The input feature matrix is ​​in the node feature matrix of the first layer. W (l) represents the weight matrix of the lth layer, represents the adjacency matrix with self-loops added, A represents the adjacency matrix, express The degree matrix of , σ represents the ReLU function;

[0096] It should be noted that GCN was chosen as the architecture of the graph neural network model because it can effectively process graph structured data and capture local structural information between nodes through graph convolution operations, thereby enhancing the model's fraud detection capabilities. Its beneficial effect is that it can make full use of node features and edge information in the graph to improve the accuracy of fraud prediction.

[0097] The ReLU function expression is f(x)=max(0,x), where if x is a positive number, the output is x, and if x is a negative number, the output is 0.

[0098] Define the loss function, expressed as,

[0099]

[0100] in, represents the loss function, y q represents the true label of sample q, represents the predicted label of sample q, and n represents the number of samples;

[0101] Define the optimizer, expressed as,

[0102]

[0103] Among them, W (l+1) represents the weight matrix of the l+1th layer, η represents the learning rate, Represents the loss function with respect to the weight matrix W(l) The gradient of

[0104] The loss function and optimizer are used to minimize the loss and optimize the model parameters, and finally the constructed graph neural network model is obtained.

[0105] It should be noted that by using risk control feature vectors and claims relationship graphs to construct a graph neural network model, the automated risk prediction function was realized, and then the accuracy of fraud detection was improved, ultimately achieving the beneficial effect of optimizing the risk control process.

[0106] S5. Set new risk control rules based on the output of the graph neural network model.

[0107] Use the K-means clustering algorithm to cluster the fraud probabilities output by the graph neural network model to identify patterns with similar fraudulent behaviors;

[0108] According to the clustering results, set the clustering threshold;

[0109] When the clustering result is greater than or equal to the clustering threshold, it indicates that fraud exists;

[0110] When the clustering result is less than the clustering threshold, it indicates that there is no fraud.

[0111] It should be noted that by setting new risk control rules based on the output of the graph neural network model, it is possible to formulate specific risk control strategies based on the model output, and then enhance the pertinence and effectiveness of the risk control rules, ultimately achieving the beneficial effect of improving the efficiency of fraud detection.

[0112] S6. Apply the new risk control rules to new claims cases, evaluate the effectiveness of the new risk control rules through risk control rule testing, and generate a new risk control rule effectiveness evaluation report.

[0113] For new claims cases, use new risk control rules to make judgments;

[0114] Risk control rule testing refers to evaluating the effectiveness of risk control rules by calculating the accuracy, recall, precision, and F1 score of the graph neural network model, which are expressed as:

[0115]

[0116] Among them, A1 represents accuracy, A2 represents recall, P represents precision, F1 represents F1 score, TP represents the number of correctly identified risk cases, TN represents the number of correctly identified non-risk cases, FN represents the number of cases incorrectly identified as non-risk, and FP represents the number of cases incorrectly identified as risk;

[0117] It should be noted that accuracy reflects the proportion of all cases correctly classified by the model. It is an intuitive and easy-to-understand indicator that can quickly give the overall performance of the model. However, in the case of an unbalanced data set, accuracy alone may mislead judgment because the model may achieve high accuracy simply by predicting the majority category. Recall focuses on the model's ability to identify all actual risk cases, which is particularly important for risk management because it is directly related to whether potential fraud can be effectively discovered and losses can be reduced. In risk control scenarios, missing any real risk case may result in significant losses. Precision measures how many of the cases marked as risks by the model are actually risk cases, which helps avoid unnecessary investigation costs and ensures that resources are used efficiently. In the field of auto insurance claims, reducing the false alarm rate can significantly reduce ineffective manual review work. The F1 score is the harmonic mean of precision and recall, which balances the contradiction between the two. In practical applications, both underreporting and false alarm risks should be avoided. The F1 score can provide a single metric that enables a trade-off between the accuracy and coverage of the model. Through comprehensive consideration of these indicators, the performance of the model can be comprehensively evaluated and the risk control rules can be adjusted accordingly to achieve the optimal risk management effect.

[0118] By summarizing the analysis of accuracy, recall, precision and F1 score, the comprehensive performance of the new risk control rules is evaluated and a new risk control rule effectiveness evaluation report is generated.

[0119] It should be noted that by applying the new risk control rules to new claims cases and evaluating their effectiveness through risk control rule testing, a quantitative evaluation of the performance of the new risk control rules was achieved, which then provided a basis for continuous improvement and finally achieved the beneficial effect of continuously optimizing the risk control system.

[0120] This embodiment also provides a risk control rule effect evaluation system based on artificial intelligence, including: a data acquisition module, which is responsible for collecting structured data and unstructured data in auto insurance claims and performing pre-processing;

[0121] The unstructured data deep learning analysis module is responsible for analyzing the processed unstructured data using deep learning technology and extracting risk control feature vectors;

[0122] The claims relationship graph construction module is responsible for constructing the claims relationship graph based on the processed structured data and unstructured data;

[0123] The graph neural network model construction module is responsible for building a graph neural network model using risk control feature vectors and claims relationship graphs;

[0124] The risk control rule setting and optimization module is responsible for setting new risk control rules based on the output of the graph neural network model;

[0125] The risk control rule effectiveness evaluation module is responsible for applying new risk control rules to new claims cases, evaluating the effectiveness of new risk control rules through risk control rule testing, and generating a new risk control rule effectiveness evaluation report.

[0126] This embodiment also provides a computer device, which is suitable for the case of an artificial intelligence-based risk control rule effect evaluation method, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the artificial intelligence-based risk control rule effect evaluation method proposed in the above embodiment.

[0127] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0128] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for evaluating the effect of risk control rules based on artificial intelligence as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, disk or optical disk.

[0129] In summary, the present invention realizes the effective integration and standardization of structured data and unstructured data in auto insurance claims through data collection and preprocessing, ensuring the accuracy and reliability of the subsequent analysis process. Through unstructured data feature extraction, it is possible to extract key features from different types of unstructured data, providing a rich source of information for the subsequent establishment of a graph neural network model. By constructing a claims relationship graph, a graph that comprehensively reflects the various entities and their relationships in the claims process is constructed, providing an intuitive visualization tool for risk assessment and improving the basis for risk control decision-making. By constructing a graph neural network model, an automated risk prediction function is realized, the accuracy of fraud detection is improved, and the risk control process is optimized. By setting new risk control rules, it is possible to formulate specific risk control strategies based on model output, thereby improving the efficiency of fraud detection. Through risk control rule testing and evaluation, a basis for continuous improvement is provided.

[0130] Example 2

[0131] Referring to Table 1, which is the second embodiment of the present invention, in order to further verify the advancement of the present invention, experimental simulation data of the risk control rule effect evaluation method based on artificial intelligence is provided.

[0132] Verify the improvement of the proposed auto insurance claims risk control system over traditional methods in handling claims cases, especially in identifying potential fraudulent behaviors. The experimental equipment and software include computer: Intel Core i7 CPU@3.60GHz, 16GB RAM, operating system: Windows 11, programming environment: Python 3.9, database: Neo4j 4.4.6, deep learning framework: TensorFlow 2.8.0.

[0133] 1. Collected 1,000 auto insurance claims cases, including 500 structured data and 500 unstructured data.

[0134] Structured data includes claims records and basic user information.

[0135] Unstructured data includes accident scene photos, customer phone recordings and text descriptions.

[0136] The structured data were preprocessed by cleaning, filling missing values ​​and standardizing.

[0137] The unstructured data was preprocessed by size standardization, normalization, sampling rate standardization, noise reduction, word segmentation, and stop word removal.

[0138] 2. Use the ResNet-50 network to extract image feature vectors from accident scene photos.

[0139] A recurrent neural network with a gated recurrent unit (GRU) was used to extract speech feature vectors from customer phone recordings.

[0140] The Word2Vec model is used to convert each word in the text description into a text feature vector representation, and the word-level text feature vectors are aggregated through the attention mechanism to obtain the sentence-level text feature vector.

[0141] The above feature vectors are integrated to obtain the risk control feature vector.

[0142] 3. Use the BiLSTM-CRF based named entity recognition model to identify entities from the processed text description.

[0143] Define the relationship types between entities and use deep learning-based relationship extraction technology to extract the relationships between entities from structured and unstructured data.

[0144] Use the Neo4j graph database to create nodes and edges in the graph and add properties to the nodes and edges.

[0145] 4. Integrate the risk control feature vector into the claim relationship graph node.

[0146] GCN is selected as the architecture of the graph neural network model, and node features are updated through graph convolution operations.

[0147] Define the loss function and optimizer to train the graph neural network model.

[0148] 5. Use the K-means clustering algorithm to cluster the fraud probability output by the graph neural network model.

[0149] Set the clustering threshold based on the clustering results.

[0150] 6. Use new risk control rules to judge new claims cases.

[0151] Calculate metrics such as accuracy, recall, precision, and F1 score.

[0152] The details are shown in Table 1:

[0153] Table 1 Experimental record table

[0154]

[0155] By analyzing the experimental data in Table 1, we can see the following beneficial effects:

[0156] 1. Effectiveness of feature extraction:

[0157] The average image feature score is 0.85, indicating that the ResNet-50 network can effectively extract useful information from accident scene photos.

[0158] The average speech feature score is 0.81, indicating that the GRU recurrent neural network can process customer phone recordings well.

[0159] The average text feature score is 0.88, indicating that the Word2Vec model combined with the attention mechanism can accurately capture the key information in the text description.

[0160] 2. Advantages of node feature integration:

[0161] The average node feature integration score is 0.86, which shows that integrating different types of feature vectors can better capture the complex relationships in claim cases.

[0162] 3. Accuracy of risk prediction:

[0163] The average risk prediction score was 0.82, showing the model’s high accuracy in identifying potential fraudulent behavior.

[0164] The experimental results show that for true fraud cases (true label is 1), the model's prediction score is higher, while for non-fraud cases (true label is 0), the model's prediction score is lower.

[0165] 4. Comparison with existing technologies:

[0166] Compared with traditional methods, the risk control system of the present invention can more accurately identify potential fraudulent behaviors and reduce false positives and missed positives.

[0167] By calculating the accuracy, recall, precision and F1 score of the model, it can be seen that the present invention has significant advantages in identifying fraudulent behavior.

[0168] In summary, by implementing the contents of the present invention, the fraud detection capability in the process of motor insurance claims can be significantly improved, and the probability of false positives and false negatives can be effectively reduced, thereby improving the efficiency and accuracy of the entire claims system.

[0169] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A risk control rule effect evaluation method based on artificial intelligence, characterized by: include, Collect structured and unstructured data from auto insurance claims and perform pre-processing; Use deep learning technology to analyze the processed unstructured data and extract risk control feature vectors; Build a claims relationship map based on the processed structured and unstructured data; Use risk control feature vectors and claims relationship graphs to build a graph neural network model; Set new risk control rules based on the output of the graph neural network model; Apply new risk control rules to new claims cases, evaluate the effectiveness of the new risk control rules through risk control rule testing, and generate a new risk control rule effectiveness evaluation report.

2. The method for evaluating the effect of risk control rules based on artificial intelligence according to claim 1, characterized in that: The structured data includes claim records and basic user information; The unstructured data includes photos of the accident scene, customer phone recordings and text descriptions; The preprocessing includes data cleaning and normalization.

3. The method for evaluating the effect of risk control rules based on artificial intelligence according to claim 1, characterized in that: Using deep learning technology to analyze the processed unstructured data and extract risk control feature vectors includes the following steps: The image feature vector is extracted from the accident scene photos using the ResNet-50 network, expressed as, x image =f ResNet-50 (I); Among them, x image represents the image feature vector extracted from the penultimate layer of ResNet-50, f ResNet-50 represents the ResNet-50 network, and I represents the data format suitable for network input formed by the processed accident scene photos; A recurrent neural network with a gated recurrent unit is used to extract a speech feature vector from a customer phone recording, represented as, x voice =f GRU (V); Among them, x voice represents the speech feature vector extracted from the last time step of GRU, V represents the data format suitable for network input formed by the processed customer phone recording, and f GRU represents a recurrent neural network with a gated recurrent unit; Use the Word2Vec model to convert each word in the text description into a text feature vector representation, and then aggregate the word-level text feature vectors through the attention mechanism to obtain the sentence-level text feature vector, which is expressed as, x word,i =f word2Vec (w i ); Among them, x word,i Represents the text feature vector of the current i-th word, z sentence represents the sentence-level text feature vector obtained by aggregation through the attention mechanism, f Word2Vec represents the Word2Vec model, M represents the number of words in the sentence, and w i represents the i-th word in the text description, α i Represents the attention weight of the current i-th word; The extracted image feature vector, speech feature vector and sentence-level text feature vector are integrated to obtain the risk control feature vector.

4. The method for evaluating the effect of risk control rules based on artificial intelligence according to claim 3, characterized in that: Based on the processed structured data and unstructured data, a claims relationship graph is constructed, including the following steps: Use the BiLSTM-CRF-based named entity recognition model to identify entities from the processed text description and connect the identified entities with the entities in the structured data; Define the relationship types between entities, use deep learning-based relationship extraction technology to extract the relationships between entities from structured and unstructured data, and verify the extracted relationships; Using the Neo4j graph database, nodes in the graph are created based on the results of entity recognition. Each node represents an entity. Edges in the graph are created based on the results of relationship extraction. Each edge represents the relationship between entities. Attributes are added to the nodes and edges. The graph data is stored in a format supported by the graph database. Finally, a claims relationship graph is obtained.

5. The method for evaluating the effect of risk control rules based on artificial intelligence according to claim 4, characterized in that: Using the risk control feature vector and the claims relationship graph, a graph neural network model is constructed, which includes the following steps: Integrate the risk control feature vector into the corresponding claim relationship graph node, and perform missing value processing, outlier detection and node attribute standardization on the integrated graph; GCN is selected as the architecture of the graph neural network model. The connection relationship between nodes in the graph is set as the adjacency matrix. The graph data containing node features and edge information is input as the input feature matrix. The fraud probability of each node is predicted, and the graph neural network model is trained. The node features are updated through graph convolution operations, which are expressed as, Among them, H (l+1) represents the node feature matrix obtained after the l+1th layer of graph convolution operation, H (l) Represents the node feature matrix of the lth layer. The input feature matrix is ​​in the node feature matrix of the first layer. W (l) represents the weight matrix of the lth layer, represents the adjacency matrix with self-loops added, A represents the adjacency matrix, express The degree matrix of , σ represents the ReLU function; Define loss function and optimizer; The loss function and optimizer are used to minimize the loss and optimize the model parameters, and finally the constructed graph neural network model is obtained.

6. The method for evaluating the effect of risk control rules based on artificial intelligence according to claim 5, characterized in that: Based on the output of the graph neural network model, new risk control rules are set, including the following steps: Use the K-means clustering algorithm to cluster the fraud probabilities output by the graph neural network model to identify patterns with similar fraudulent behaviors; According to the clustering results, set the clustering threshold; When the clustering result is greater than or equal to the clustering threshold, it indicates that fraud exists; When the clustering result is less than the clustering threshold, it indicates that there is no fraud.

7. The method for evaluating the effect of risk control rules based on artificial intelligence according to claim 6, characterized in that: Apply the new risk control rules to new claims cases, evaluate the effectiveness of the new risk control rules through risk control rule testing, and generate a new risk control rule effectiveness evaluation report, including the following steps: For new claims cases, use new risk control rules to make judgments; Risk control rule testing refers to calculating the accuracy, recall, precision, and F1 score of the graph neural network model to evaluate the effectiveness of the risk control rules; By summarizing the analysis of accuracy, recall, precision and F1 score, the comprehensive performance of the new risk control rules is evaluated and a new risk control rule effectiveness evaluation report is generated.

8. An artificial intelligence-based risk control rule effect evaluation system, based on the artificial intelligence-based risk control rule effect evaluation method according to any one of claims 1 to 7, characterized in that: include, The data collection module is responsible for collecting structured and unstructured data in auto insurance claims and performing pre-processing; The unstructured data deep learning analysis module is responsible for analyzing the processed unstructured data using deep learning technology and extracting risk control feature vectors; The claims relationship graph construction module is responsible for constructing the claims relationship graph based on the processed structured data and unstructured data; The graph neural network model construction module is responsible for building a graph neural network model using risk control feature vectors and claims relationship graphs; The risk control rule setting and optimization module is responsible for setting new risk control rules based on the output of the graph neural network model; The risk control rule effectiveness evaluation module is responsible for applying new risk control rules to new claims cases, evaluating the effectiveness of new risk control rules through risk control rule testing, and generating a new risk control rule effectiveness evaluation report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the risk control rule effect evaluation method based on artificial intelligence described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the risk control rule effect evaluation method based on artificial intelligence described in any one of claims 1 to 7 are implemented.

Images