Case information processing method and system
Through communication connection with the auto insurance management platform and building a big risk model, the insurance industry relies on manual experience in vehicle accident claims is solved, efficiency and accuracy are improved, and data security is ensured.
Patent Information
- Application Number
- CN202411059343.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-03
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-03
AI Technical Summary
The insurance industry relies on manual experience and rule logic in the process of vehicle accident claims, resulting in problems of subjectivity, inefficiency and low accuracy.
By establishing a communication connection with the auto insurance management platform, automatically converting vehicle damage case information, building a knowledge training corpus and risk model, and building a permission management module for user identity verification to assist in auto insurance claims risk management.
Improve the efficiency and accuracy of auto insurance management while ensuring data security.
Smart Images

Figure CN118628261B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of vehicle damage claims analysis, and in particular to a case information processing method and system. Background Art
[0002] With the development of the insurance industry, most vehicles are insured. In daily life, vehicle accidents always happen. In order to solve the problems related to vehicle accidents, insurance companies have launched a set of insurance claims procedures, which are mainly divided into six links: reporting, investigation, damage assessment, damage verification, claims verification, and settlement. The task of the damage verification link is to inquire about the price of accessories through their own experience or vehicle factory data, vehicle accessories cooperative sales channels, etc. after the damage determination link, and judge whether the price given by the repair shop is risky. The damage verification is mainly conducted by the damage verifier through accident scenes, accident pictures, pictures of damaged parts, and logical relationships of parts loss, based on their own experience, to judge whether the repair shop deliberately expands the loss, whether the repair must replace the parts, or whether the damaged parts are not caused by the current accident, etc., to judge whether there is a risk.
[0003] However, due to the particularity and complexity of the insurance industry, insurance companies face a series of challenges in risk assessment and risk management. Traditional risk assessment methods in the insurance industry mainly rely on manual experience and rule logic, which are subjective, inefficient and inaccurate. Summary of the invention
[0004] The purpose of this application is to provide a case information processing method and system to solve the technical problems in the prior art that the method is subjective, inefficient and inaccurate due to reliance on manual experience and rule logic.
[0005] In view of the above problems, the present application provides a case information processing method and system.
[0006] In a first aspect, the present application provides a case information processing method, which is implemented through a case information processing system, including: establishing a communication connection with a motor vehicle insurance management platform of a first enterprise, receiving the vehicle damage case information of the first enterprise, and converting it according to a predetermined data structure to generate first vehicle damage case data; performing information combing and data conversion on the first vehicle damage case data, and constructing a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question-and-answer scripts; performing model training based on the knowledge training corpus and the vehicle damage case information according to preset rule logic to generate a first risk model; constructing a permission management module for the first risk model, wherein the permission management module includes multiple permission management centers, and the multiple permission management centers are used to verify user access rights according to a preset permission attribute set; when a first user activates the first risk model, the permission management module collects identity attributes of the first user, and verifies the access rights of the first user according to the collection results; when the access right verification passes, the first risk model analyzes and responds to the case analysis request of the first user, and outputs risk analysis information.
[0007] In the second aspect, the present application also provides a case information processing system for executing a case information processing method as described in the first aspect, including: a vehicle damage case receiving module, which is used to establish a communication connection with the vehicle insurance management platform of the first enterprise, receive the vehicle damage case information of the first enterprise, and convert it according to a predetermined data structure to generate first vehicle damage case data; a corpus construction module, which is used to sort out the information and convert the data of the first vehicle damage case data to construct a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question and answer scripts; a model training module, which is used to perform model training based on the knowledge training corpus and the vehicle damage case information according to preset rule logic. Training, generating a first risk model; a permission management module, used to construct the permission management module of the first risk model, wherein the permission management module includes multiple permission management centers, and the multiple permission management centers are used to verify user access rights according to a preset permission attribute set; a permission verification module, used when the first user activates the first risk model, the permission management module collects identity attributes of the first user, and verifies the access rights of the first user according to the collection results; a risk analysis module, used when the access right verification is passed, the first risk model analyzes and responds to the case analysis request of the first user, and outputs risk analysis information.
[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:
[0009] Establish a communication connection with the auto insurance management platform of the first enterprise, receive the vehicle damage case information of the first enterprise, and convert it according to a predetermined data structure to generate the first vehicle damage case data; perform information sorting and data conversion on the first vehicle damage case data to construct a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question and answer scripts; perform model training based on the knowledge training corpus and the vehicle damage case information according to preset rule logic to generate a first risk model; construct a permission management module for the first risk model, wherein the permission management module includes multiple permission management centers, and the multiple permission management centers are used to verify user access rights according to a preset permission attribute set; when the first user activates the first risk model, the permission management module collects identity attributes of the first user, and verifies the access rights of the first user according to the collection results; when the access right verification passes, the first risk model analyzes and responds to the case analysis request of the first user and outputs risk analysis information. By establishing a communication connection with the auto insurance management platform and automatically converting vehicle damage case information, building a knowledge training corpus and training risk big model, and then building a permission management module to perform user identity authentication to call the big model, assisting in auto insurance claims risk management, achieving the technical effect of improving the efficiency and accuracy of auto insurance management while ensuring data security.
[0010] 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
[0011] 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.
[0012] Figure 1 A flowchart of a case information processing method for this application;
[0013] Figure 2 This is a structural diagram of a case information processing system for this application.
[0014] Description of reference numerals:
[0015] Vehicle damage case receiving module 11, corpus building module 12, model training module 13, authority management module 14, authority verification module 15, risk analysis module 16. DETAILED DESCRIPTION
[0016] This application provides a case information processing method and system to solve the technical problems in the prior art that the system is subjective, inefficient and inaccurate due to reliance on manual experience and rule logic. By establishing a communication connection with the auto insurance management platform and automatically converting vehicle damage case information, building a knowledge training corpus and training a large risk model, and then building a permission management module to perform user identity authentication to call the large model, assist in auto insurance claim risk management, and achieve the technical effect of improving the efficiency and accuracy of auto insurance management while ensuring data security.
[0017] 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.
[0018] For example, please refer to the attached Figure 1 The present application provides a case information processing method, wherein the case information processing method is applied to a case information processing system, and the case information processing method specifically includes the following steps:
[0019] Step 1: Establish a communication connection with the first enterprise's auto insurance management platform, receive the first enterprise's auto damage case information, and convert it according to a predetermined data structure to generate first auto damage case data.
[0020] Specifically, establishing a communication connection with the first enterprise's auto insurance management platform means using standard communication protocols and technologies, such as HTTP, HTTPS, FTP, SOAP, RESTful API, etc., through the Internet or a dedicated network to achieve data exchange and communication with the first enterprise's auto insurance management platform. Among them, the first enterprise refers to any insurance claims company, and the first enterprise's auto insurance management platform is the system platform used for auto insurance relations in the first enterprise. Receiving the vehicle damage case information of the first enterprise means obtaining case information related to its vehicle loss from the first enterprise's auto insurance management platform, including but not limited to case number, collision scene, collision vehicle, number of vehicles, collision direction and vehicle attributes.
[0021] Converting according to a predetermined data structure means that according to the characteristics and needs of the vehicle damage case data, professional and technical personnel in this field define a data structure by themselves, and organize and convert the received vehicle damage case information according to the data structure, so as to generate the first vehicle damage case data, so as to ensure the standardization and consistency of the data, so as to carry out subsequent data analysis and processing, facilitate analysis for different enterprises, and improve the adaptability of data analysis.
[0022] Step 2: Perform information sorting and data conversion on the first vehicle damage case data to construct a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question and answer scripts.
[0023] Specifically, information sorting is to analyze and organize the data of vehicle damage cases in detail and extract key information, such as accident type, loss location, repair cost, etc. Data conversion is to convert this key information according to certain formats and standards for subsequent processing and use.
[0024] Constructing a knowledge training corpus refers to designing and creating a corpus for training a machine learning model based on the characteristics and requirements of the first vehicle damage case data after data conversion. The knowledge training corpus usually includes a large amount of text data, which is used to train the model to understand and handle similar problems. In the context of vehicle damage case data, the knowledge training corpus may include text information related to accident descriptions, loss assessments, repair suggestions, etc. The knowledge training corpus includes a combination of fragmented question-answering techniques, that is, the text data in the corpus is organized in the form of questions and answers, where the questions may be various queries about vehicle damage cases, such as "What is the cause of the accident?" "Which part is damaged?" "What is the extent of the loss?", etc., and the answers are relevant information extracted from the vehicle damage case data. Fragmentation means that these question-answering techniques can exist independently and do not need to rely on context, which can improve the flexibility and reusability of the corpus. Thus, a rich knowledge training corpus is constructed for training machine learning models to automatically process and answer various questions about vehicle damage cases and improve the efficiency and accuracy of auto insurance management.
[0025] Step 3: According to the preset rule logic, model training is performed based on the knowledge training corpus and the vehicle damage case information to generate a first risk model.
[0026] Specifically, to train the model according to the preset rule logic, it is first necessary to define the training objectives and rules of the model. In the field of auto insurance management, these rules may include how to determine the responsible party of the accident, assess the extent of loss, calculate the amount of compensation, etc. based on the information of the vehicle damage case. The preset rule logic is formulated based on industry knowledge and experience, and is intended to guide the model on how to learn and extract useful information from the vehicle damage case data. Model training is performed based on the knowledge training corpus, that is, the data in the constructed corpus is used to train the machine learning model. The question-answering combination in the knowledge training corpus provides a large number of examples, and the model understands and simulates the way humans answer by learning these examples. During the training process, the model will try to find the correlation between questions and answers, and how to extract key information from the vehicle damage case information to generate accurate answers. Through the above training process, a machine learning model that can process vehicle damage case information and generate risk assessment results is created. This model can be a complex neural network, such as a deep learning model, which can learn and extract useful features from a large amount of data. The goal of the First Risk Big Model is to assess risks based on vehicle damage case information, which may include the severity of the accident, the possibility of compensation, the risk of compensation, etc., to help the First Enterprise better manage its auto insurance business and improve the accuracy and efficiency of decision-making.
[0027] Step 4: Construct a permission management module of the first risk model, wherein the permission management module includes multiple permission management centers, and the multiple permission management centers are used to verify user access rights according to a preset permission attribute set.
[0028] Specifically, a functional module that can control user access rights, namely the permission management module, is constructed. The permission management module usually includes multiple permission management centers, each of which is responsible for managing different permission attributes. The preset permission attribute set is a set of rules and attributes that define user access rights. These attributes may include the user's role, department, position, access level, etc. For example, in the auto insurance management platform, there may be different user roles, such as administrator, customer service staff, auditor, etc., and each role may have different access rights. The role of multiple permission management centers is to verify user access rights according to these preset permission attribute sets. When a user tries to access the first risk model, the permission management module checks the user's permission attributes, such as the user's role and access level, and then determines whether the user has permission to access the model based on these attributes. This permission management mechanism can ensure that only users with appropriate permissions can access and use the first risk model, thereby protecting the security of the model and data and preventing unauthorized access and potential abuse.
[0029] Step 5: When the first user activates the first risk model, the authority management module collects identity attributes of the first user and verifies the access rights of the first user according to the collection results.
[0030] Specifically, when the first user activates the first risk model, the rights management module starts the identity attribute collection process. This process involves collecting relevant information of the first user, which generally includes but is not limited to the user's user name, role, department, position, access level, etc. This information can be collected through the user login interface, or obtained through the user's session information or credentials in the access request. Once the identity attributes of the first user are collected, the rights management module verifies the information according to the preset permission attribute set. The verification process includes checking whether the user's identity attributes match the permission attribute set that allows access to the first risk model. For example, if the model only allows users with the "administrator" role to access, the rights management module checks whether the first user has this role. If the identity attributes of the first user meet the access rights requirements, the rights management module will allow the user to access the first risk model, and may provide different access levels according to the user's role and permission level. For example, an administrator may have full access rights, while an ordinary user may only be able to access a portion of the model's functions. If the identity attributes of the first user do not meet the access rights requirements, the rights management module will reject the user's access request and may provide a corresponding error message or prompt information to inform the user of the reason for the inability to access. Through such identity attribute collection and access rights verification process, it can be ensured that only users with appropriate permissions can access the first risk model, thereby protecting the security of the model and data and preventing unauthorized access and potential abuse.
[0031] Step 6: When the access permission verification is passed, the first risk model analyzes and responds to the case analysis request of the first user and outputs risk analysis information.
[0032] Specifically, when the access permission verification is passed, it means that the first user is authorized to access the first risk model. At this time, if the first user submits a case analysis request, that is, inputs a question to the first risk model, the first risk model will start to process the request. Outputting risk analysis information means that the first risk model generates an analysis result on the case risk based on the results of the case analysis, which may include risks, recommended handling measures, etc. This information can help the first user better understand the risk status of the case and make corresponding decisions, improve the efficiency and accuracy of case handling, and provide valuable decision-making support for the first user.
[0033] Further, step 2 of this application includes:
[0034] A preset knowledge decomposition unit is set to perform knowledge decomposition on the first vehicle damage case data to generate a knowledge unit set; based on the knowledge unit set, an associated database is called to expand vehicle knowledge and generate an expanded knowledge unit set; derivative knowledge points are configured, and question-answer format conversion is performed in combination with the expanded knowledge unit set to generate a fragmented question-answering speech combination and establish the knowledge training corpus.
[0035] Furthermore, the present application also includes the following steps:
[0036] The derived knowledge points are combined with the expanded knowledge unit set to perform linkage multi-layer question-answering logical relationship identification to generate a multi-layer question-answering logical relationship; and the fragmented question-answering speech combination is established based on the multi-layer question-answering logical relationship.
[0037] Specifically, setting a preset knowledge decomposition unit means creating a set of rules or algorithms for decomposing the first vehicle damage case data into smaller, manageable knowledge units. These knowledge units may include specific information such as accident type, loss location, collision scene, collision vehicle, number of vehicles, collision direction, etc. The process of knowledge decomposition involves extracting key information and organizing it into a structured data format for subsequent processing and analysis. The decomposed knowledge units are collected to form a set containing all relevant information to generate a knowledge unit set.
[0038] Calling the associated database based on the knowledge unit set means using the information in the knowledge unit set to query the associated database to obtain more relevant information. For example, in the case information, the collision scene, collision vehicle, number of vehicles, and collision direction are usually expressed through fields such as accident type (single party / double party / multiple parties), number of third-party vehicles (0 / 1 / 2 / 3 / 4), and accident process. In the vehicle information, although there are only simple VIN and vehicle name, more vehicle knowledge such as vehicle structure, price, configuration, engine model, and number of doors can be obtained through database association. The associated database contains data such as vehicle structure, price, configuration, engine model, and number of doors corresponding to the VIN and vehicle name, which can help expand the knowledge unit set and provide more comprehensive information support. The relevant information queried in the associated database is added to the original knowledge unit set to form a more comprehensive and richer information set. For example, if the knowledge unit set contains an accident type, the associated database may provide data such as engine model and number of doors related to the accident type.
[0039] Configuring derived knowledge points refers to defining a set of additional knowledge points related to the original knowledge units, which are specifically set by professional and technical personnel in this field based on actual needs. For example, in the damage assessment information, the repair degree of all items of the accident is listed, and additional fields need to be derived from it, such as the number of accessories, the number of paints, the number of sheet metals, the attributes of accessories (such as the location, fragility, material), whether the accessories belong to the vehicle, whether the price is too high, etc., which are used as derived knowledge points, and combined with the expanded knowledge unit set for question-and-answer conversion, which means converting the expanded knowledge unit set into a question-and-answer format, that is, converting each knowledge unit into one or more questions and their answers. This conversion can help build a knowledge training corpus for training machine learning models, organize the converted question-and-answer dialogues into a set of reusable, independent words, and generate a fragmented question-and-answer combination for training machine learning models, so that they can understand and answer various questions about vehicle damage cases, and improve the efficiency and accuracy of auto insurance management.
[0040] Specifically, the specific steps of establishing the knowledge training corpus are as follows: Linked multi-layer question-answer logical relationship identification refers to analyzing and identifying the logical relationship between derived knowledge points and expanded knowledge unit sets, which may include causal relationships, subordinate relationships, parallel relationships, etc. For example, a derived knowledge point may explain the cause of a certain type of accident, and the expanded knowledge unit set may contain statistical data of the accident type, and there is a causal relationship between the two. Generating multi-layer question-answer logical relationships refers to organizing the identified logical relationships into a multi-level question-answer structure. In this structure, each question-answer dialogue is logically connected to other question-answer dialogues, forming a hierarchical and interconnected question-answer network. Such a structure can help machine learning models better understand and handle complex question-answer scenarios. Exemplarily, the question-answer format for vehicle attributes is converted into detailed answers including attributes such as brand, car series, vehicle type, drive form, body structure and energy type. In addition, the question-answer format for information such as body structure and damaged items in single-party collision accident cases is converted, including items of loss-related risks and recommended verification content, so that the model forms linked multi-layer questions and answers based on logical relationships and finally obtains the required answers.
[0041] Establishing the fragmented question-answering speech combination based on the multi-layer question-answering logical relationship refers to using the identified multi-layer question-answering logical relationship to construct a set of question-answering speech. These speech are fragmented and can exist independently without relying on context. Each speech corresponds to a node in the question-answering logical relationship, which contains questions, answers, and logical links with other speech. Through such a process, a structured and hierarchical question-answering speech combination can be created for training machine learning models. Such a speech combination not only provides rich information, but also can simulate the question-answering process in real scenarios, and improve the accuracy and efficiency of the model in practical applications.
[0042] Further, step three of this application includes:
[0043] The preset rule logic includes formula form logic and natural language conversion logic; the knowledge training corpus is converted into formula discourse through the formula form logic, and the formula conversion result is converted into natural language through the natural language conversion logic to generate a knowledge base conversion result; the vehicle damage case information and the corresponding case results are converted into natural language to generate a case information conversion result; the first risk big model is trained based on the knowledge base conversion result and the case information conversion result.
[0044] Furthermore, the present application also includes the following steps:
[0045] The knowledge base conversion result is used as training data, and the underlying model is generated based on the pre-training of the Transformer architecture; the risk control rule knowledge base is obtained, and the underlying model is fine-tuned for a second time in combination with the case information and the knowledge base conversion result to generate the first risk model.
[0046] Specifically, the preset rule logic includes two types of logic: formula form logic and natural language conversion logic. Simply put, in view of the large amount of logical knowledge formed by traditional rule-based risk judgment, it is necessary to ensure that the model can clearly understand the logic of the rule itself. To achieve this goal, it is necessary to establish a set of formula-based discourse so that the rule itself can clearly express judgment logic, loop logic, comparison logic, and inclusion logic. Exemplarily, the formula-based logic is set to the form of "when [field] [judgment condition] [value] = [result]", so that the rule content can be converted into a knowledge base corpus, making it easier for the model to understand and apply. In this way, the model can better utilize the rule knowledge base for risk judgment and decision-making, and improve the intelligence and efficiency of the system.
[0047] Natural language conversion logic is a logic that converts the result of formula formal logic processing into natural language. This logic usually involves natural language processing (NLP) technology, such as text generation, semantic analysis, etc. The formula conversion result is converted into natural language through natural language conversion logic, and the knowledge base conversion result can be generated. Large models must communicate with users in a form that shows the emotional way of human communication in the dialogue effect. Therefore, the results obtained by formula conversion are too dependent on formulaic words, which are suitable for developers to understand, but may be difficult for front-line operators to understand. In order to make the logic of the rules easier to understand, the rules need to be described by professional experts and converted into easy-to-understand vernacular training corpus. In this way, front-line operators can understand the logic of the rules more clearly and improve their application and use efficiency of the model.
[0048] The vehicle damage case information and the corresponding case results are converted into natural language, that is, the vehicle damage case data is converted into natural language descriptions using natural language processing technology. For example, data such as accident descriptions and degree of loss are converted into easy-to-understand text descriptions. The case information conversion results generated in this way can be used to train machine learning models to help the model understand the specific circumstances of the case. Training the first large risk model based on the knowledge base conversion results and the case information conversion results means using these converted data to train the machine learning model. The knowledge base conversion results provide knowledge that the model can learn, and the case information conversion results provide training data for the model. Through such a training process, the first large risk model can learn how to generate risk assessment results based on vehicle damage case information, thereby improving its accuracy and efficiency in practical applications.
[0049] Specifically, using the knowledge base conversion result as training data, it is first necessary to prepare a data set for training the machine learning model. The knowledge base conversion result contains a large number of question-answering dialogues, which can be used as the input and output of the model. In the field of auto insurance management, these question-answering dialogues may involve aspects such as accident types, loss assessment, and compensation recommendations. Pre-training based on the Transformer architecture to generate the underlying model refers to using the Transformer model to process the question-answering dialogues in the knowledge base conversion result. Transformer is a deep learning model that processes sequence data, such as natural language text, through a self-attention mechanism. Pre-training refers to training the Transformer model on a large amount of text data so that it can understand and generate natural language. The underlying model is generated in this pre-training stage and can understand and process natural language text.
[0050] Acquiring a risk control rule knowledge base refers to collecting and organizing rules and knowledge related to auto insurance risk assessment. These rules may include how to assess risk based on accident type and loss level, how to calculate compensation, etc. The risk control rule knowledge base is a collection of expert knowledge and experience in the field of auto insurance management, which can help the model better understand and process vehicle damage case information and is established by professional technicians in this field.
[0051] The secondary fine-tuning training of the underlying model in combination with the case information and the knowledge base conversion result refers to further training and adjusting the underlying model using the case information and the knowledge base conversion result. This process is called fine-tuning, which is a refinement and optimization based on pre-training. Exemplarily, the existing LoRA method can be used to fine-tune the underlying model. LoRA (Low-Rank Adaptation) is a fine-tuning method based on freezing the weight parameters of the pre-trained model. Its principle is mainly to add additional network layers to the underlying model and only train these newly added network layer parameters while keeping the original model parameters frozen. LoRA freezes the pre-trained model weight parameters and only trains the newly added network layer parameters, thereby reducing the number of parameters that need to be updated during the fine-tuning process and reducing the training cost. LoRA adds a bypass next to the underlying model to simulate the update amount of parameters through low-rank decomposition. By reducing the dimension first and then increasing the dimension, the high-rank matrix in the original model can be effectively approximated, thereby reducing the number of parameters while maintaining the main feature representation ability of the model. The LoRA method uses a pluggable method to switch tasks. It only needs to replace the corresponding reduced and increased dimensionality matrices, making switching between different tasks simple and efficient. During the inference phase, LoRA can multiply the reduced-dimensional matrix and the increased-dimensional matrix and add them to the original parameters without introducing additional inference delays, thus ensuring inference efficiency.
[0052] Through such training, the model can learn how to generate risk assessment results based on specific case information. Generating the first large risk model means creating a machine learning model that can process vehicle damage case information and generate risk assessment results through the above-mentioned pre-training and fine-tuning process. This model can be a complex neural network, such as a deep learning model, which can learn and extract useful features from large amounts of data. The goal of the first large risk model is to assess risks based on vehicle damage case information, which may include the severity of the accident, the risk of compensation, etc., to help the first enterprise better manage its auto insurance business and improve the accuracy and efficiency of decision-making.
[0053] Further, step five of this application includes:
[0054] Define preset access attribute rules. When the collection results meet the preset access attribute rules, call the multiple permission management centers to distribute keys and generate multiple permission verification keys. The first user decrypts the model data of the first risk model through the multiple permission verification keys to complete the access permission verification.
[0055] Specifically, the preset access attribute rules refer to the rules formulated by professional and technical personnel in this field based on business needs and security considerations to determine whether a user has the authority to access the first risk model. The preset access attribute rules may include a threshold for the number of user attributes. In layman's terms, the preset permission attribute set is a set of rules and attributes that define user access rights. These attributes may include a variety of attributes such as the user's role, department, position, etc. Exemplarily, the preset permission attribute set contains six identity sequences. As long as the identity attributes of the first user are the same as the four attributes in the preset permission attribute set, the requirements are met, that is, the collection results meet the preset access attribute rules, and the user's identity attributes meet the preset access rules.
[0056] At this time, the multiple permission management centers are called to distribute keys and generate multiple permission verification keys. That is to say, the permission management module will generate a set of keys. Specifically, the number of keys is the same as the user attribute number threshold in the preset access attribute rule, and the number of multiple permission management centers is the same as the number of attributes in the preset permission attribute set. According to the actual user attribute number threshold, a corresponding number of permission management centers are randomly selected from multiple permission management centers to generate a corresponding number of keys, that is, multiple permission verification keys. Among them, multiple permission management centers are functional modules for generating keys, which can be constructed based on existing data encryption technology.
[0057] The first user decrypts the model data of the first risk model through the multiple permission verification keys to complete the access permission verification, that is, the user uses these keys to decrypt the model data, and the user can access the model only when the keys are correct and valid. This method can ensure that only users with appropriate permissions can access and use the first risk model, thereby protecting the security of the model and data and preventing unauthorized access and potential abuse.
[0058] Furthermore, step six of this application includes:
[0059] Receive the request question information of the first user through the user interaction interface; classify the request question information through the first risk model, collect key conversation information based on the classification result, perform information missing analysis, and generate information supplement prompts; receive information supplement results through the user interaction interface, retrieve related information from the corresponding knowledge base according to keywords, perform case risk analysis based on the information retrieval results, and generate a request response answer, which is sent to the first user.
[0060] Specifically, receiving the request question information of the first user through the user interaction interface means providing a user interface to allow the first user to input or select the question he wants to ask. This interface can be a web page, a mobile application or any other form of user interface. Classifying the request question information through the first risk big model means using the first risk big model to identify and understand the user's question and classify it into appropriate categories. For example, the question may be about accident responsibility, compensation amount or case trend. The classification result can help the model determine how to deal with the user's question. Collecting the key information of the conversation based on the classification result, performing information missing analysis, and generating information supplement prompts means that the model identifies the key information required to deal with the problem based on the classification result and analyzes whether the information is complete. If the information is missing, the model will generate a prompt to guide the user to supplement the missing information. For example, the user inputs: What leakage risk exists in case A. The first risk big model: There are vehicles X and Y under case number A. Please ask which vehicle's leakage risk needs to be inquired. Logical processing: Problem classification = case risk. Logical processing: Key field case number = A. Logical processing: There are vehicles X and Y under case number A, and it is a two-party accident. Logical processing: The key field license plate number is missing. At this time, the user is prompted to supplement the license plate number description.
[0061] Receiving the information supplement result through the user interaction interface means allowing the user to supplement the missing information through the interface according to the prompt of the model. The information supplemented by the user will be received by the model and used for subsequent processing. Retrieving related information in the corresponding knowledge base according to keywords means using the keywords in the information supplemented by the user to retrieve relevant information from the knowledge base, using the information retrieved from the knowledge base to perform risk analysis on the user's question and generate an answer. This answer may be about case risk assessment, recommended processing steps or other relevant information. For example, user: leakage risk of vehicle X. The first risk model: The degree of loss of the left front headlight of vehicle X is too light, and the means of split replacement can be adopted. Logical processing: problem classification = case risk. Logical processing: key fields case number = A, license plate number = X. Logical processing: retrieve the core system, the vehicle is a sedan, the vehicle code is L, the accessories are headlights, engine hood, and the working hours are front fender painting and front bumper skin painting. Logical processing: retrieve the knowledge base, the vehicle with vehicle code L, the OEM of the headlight is B, which can be provided in separate parts, the separate parts are the headlight shell and the headlight bulb. Logical processing: retrieve the knowledge base, when the headlight is seriously damaged and needs to be replaced, the following accessories will also be damaged accordingly, the front fender and the front bumper skin. Logical processing: transmit the above information to the big model, and let the big model return the specific risk information in the case, to help users better understand the risk status of the case and make corresponding decisions.
[0062] In summary, the case information processing method provided by this application has the following technical effects:
[0063] Establish a communication connection with the auto insurance management platform of the first enterprise, receive the vehicle damage case information of the first enterprise, and convert it according to a predetermined data structure to generate the first vehicle damage case data; perform information sorting and data conversion on the first vehicle damage case data to construct a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question and answer scripts; perform model training based on the knowledge training corpus and the vehicle damage case information according to preset rule logic to generate a first risk model; construct a permission management module for the first risk model, wherein the permission management module includes multiple permission management centers, and the multiple permission management centers are used to verify user access rights according to a preset permission attribute set; when the first user activates the first risk model, the permission management module collects identity attributes of the first user, and verifies the access rights of the first user according to the collection results; when the access right verification passes, the first risk model analyzes and responds to the case analysis request of the first user and outputs risk analysis information. By establishing a communication connection with the auto insurance management platform and automatically converting vehicle damage case information, building a knowledge training corpus and training risk big model, and then building a permission management module to perform user identity authentication to call the big model, assisting in auto insurance claims risk management, achieving the technical effect of improving the efficiency and accuracy of auto insurance management while ensuring data security.
[0064] Embodiment 2: Based on the case information processing method in the above embodiment, and with the same inventive concept, the present application also provides a case information processing system, as shown in the attached Figure 2 , the case information processing system comprises:
[0065] The vehicle damage case receiving module 11 is used to establish a communication connection with the vehicle insurance management platform of the first enterprise, receive the vehicle damage case information of the first enterprise, and convert it according to a predetermined data structure to generate first vehicle damage case data.
[0066] The corpus construction module 12 is used to perform information sorting and data conversion on the first vehicle damage case data to construct a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question and answer scripts.
[0067] The model training module 13 is used to perform model training based on the knowledge training corpus and the vehicle damage case information according to preset rule logic to generate a first risk model.
[0068] The permission management module 14 is used to construct the permission management module of the first risk model, wherein the permission management module includes multiple permission management centers, and the multiple permission management centers are used to verify user access rights according to a preset permission attribute set.
[0069] The authority verification module 15 is used for, when the first user activates the first risk model, the authority management module collects identity attributes of the first user and verifies the access rights of the first user according to the collection results.
[0070] The risk analysis module 16 is used for, when the access authority verification is passed, the first risk model to analyze and respond to the case analysis request of the first user and output risk analysis information.
[0071] Furthermore, the corpus construction module 12 in the case information processing system is also used for:
[0072] A preset knowledge decomposition unit is set to perform knowledge decomposition on the first vehicle damage case data to generate a knowledge unit set; based on the knowledge unit set, an associated database is called to expand vehicle knowledge and generate an expanded knowledge unit set; derivative knowledge points are configured, and question-answer format conversion is performed in combination with the expanded knowledge unit set to generate a fragmented question-answering speech combination and establish the knowledge training corpus.
[0073] Furthermore, the corpus construction module 12 in the case information processing system is also used for:
[0074] The derived knowledge points are combined with the expanded knowledge unit set to perform linkage multi-layer question-answering logical relationship identification to generate a multi-layer question-answering logical relationship; and the fragmented question-answering speech combination is established based on the multi-layer question-answering logical relationship.
[0075] Furthermore, the model training module 13 in the case information processing system is also used for:
[0076] The preset rule logic includes formula form logic and natural language conversion logic; the knowledge training corpus is converted into formula discourse through the formula form logic, and the formula conversion result is converted into natural language through the natural language conversion logic to generate a knowledge base conversion result; the vehicle damage case information and the corresponding case results are converted into natural language to generate a case information conversion result; the first risk big model is trained based on the knowledge base conversion result and the case information conversion result.
[0077] Furthermore, the model training module 13 in the case information processing system is also used for:
[0078] The knowledge base conversion result is used as training data, and the underlying model is generated based on the pre-training of the Transformer architecture; the risk control rule knowledge base is obtained, and the underlying model is fine-tuned for a second time in combination with the case information and the knowledge base conversion result to generate the first risk model.
[0079] Furthermore, the authority verification module 15 in the case information processing system is also used for:
[0080] Define preset access attribute rules. When the collection results meet the preset access attribute rules, call the multiple permission management centers to distribute keys and generate multiple permission verification keys. The first user decrypts the model data of the first risk model through the multiple permission verification keys to complete the access permission verification.
[0081] Furthermore, the risk analysis module 16 in the case information processing system is also used for:
[0082] Receive the request question information of the first user through the user interaction interface; classify the request question information through the first risk model, collect key conversation information based on the classification result, perform information missing analysis, and generate information supplement prompts; receive information supplement results through the user interaction interface, retrieve related information from the corresponding knowledge base according to keywords, perform case risk analysis based on the information retrieval results, and generate a request response answer, which is sent to the first user.
[0083] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. Figure 1 The case information processing method and specific examples in the first embodiment are also applicable to the case information processing system in the present embodiment. Through the above detailed description of the case information processing method, those skilled in the art can clearly understand the case information processing system in the present embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.
[0084] 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.
[0085] 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 case information processing method, characterized in that: include: Establishing a communication connection with the first enterprise's auto insurance management platform, receiving the first enterprise's auto damage case information, and converting it according to a predetermined data structure to generate first auto damage case data; Performing information sorting and data conversion on the first vehicle damage case data to construct a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question-and-answer scripts; According to preset rule logic, model training is performed based on the knowledge training corpus and the vehicle damage case information to generate a first risk model; Constructing a rights management module of the first risk model, wherein the rights management module includes a plurality of rights management centers, and the plurality of rights management centers are used to verify user access rights according to a preset set of rights attributes; When the first user activates the first risk model, the authority management module collects identity attributes of the first user and verifies the access rights of the first user according to the collection results; When the access permission verification is passed, the first risk model analyzes and responds to the case analysis request of the first user and outputs risk analysis information; The first vehicle damage case data is sorted and converted to construct a knowledge training corpus, including: Setting a preset knowledge decomposition unit to perform knowledge decomposition on the first vehicle damage case data to generate a knowledge unit set; Based on the knowledge unit set, calling the associated database, expanding the vehicle knowledge, and generating an expanded knowledge unit set; Configuring derived knowledge points, and converting the question-answer format in combination with the expanded knowledge unit set to generate a fragmented question-answering speech combination, and establishing the knowledge training corpus; Configuring derived knowledge points, and converting the question-answer format in combination with the expanded knowledge unit set to generate a fragmented question-answering speech combination, and establishing the knowledge training corpus, including: Combining the derived knowledge points with the expanded knowledge unit set to perform linkage multi-layer question-answer logic relationship identification to generate a multi-layer question-answer logic relationship; Establishing the fragmented question-answering speech combination based on the multi-layer question-answering logical relationship; According to the preset rule logic, model training is performed based on the knowledge training corpus and the vehicle damage case information to generate a first risk model, including: The preset rule logic includes formula form logic and natural language conversion logic; Performing formula discourse conversion on the knowledge training corpus through the formula formal logic, and converting the formula conversion result into natural language through the natural language conversion logic to generate a knowledge base conversion result; Performing natural language conversion on the vehicle damage case information and the corresponding case results to generate a case information conversion result; Training the first risk model based on the knowledge base conversion result and the case information conversion result; Training the first risk model based on the knowledge base conversion result and the case information conversion result includes: Using the knowledge base conversion results as training data, the underlying model is generated based on the pre-training of the Transformer architecture; Acquire a risk control rule knowledge base, and perform secondary fine-tuning training on the underlying model in combination with the case information and the knowledge base conversion result to generate the first risk macro model; When the first user activates the first risk model, the authority management module collects identity attributes of the first user and verifies the access rights of the first user according to the collection results, including: Defining a preset access attribute rule, and when the collection result meets the preset access attribute rule, calling the multiple authority management centers to distribute keys and generate multiple authority verification keys; The first user decrypts the model data of the first risk model using the multiple permission verification keys to complete the access permission verification; The first risk model analyzes and responds to the case analysis request of the first user, and outputs risk analysis information, including: Receiving the request question information of the first user through the user interaction interface; Classifying the request question information through the first risk model, collecting key conversation information based on the classification result, performing information missing analysis, and generating information supplement prompts; The information supplement results are received through the user interaction interface, related information is retrieved from the corresponding knowledge base according to the keywords, case risk analysis is performed based on the information retrieval results, and a request response answer is generated and sent to the first user.
2. A case information processing system, characterized in that: The steps for implementing the case information processing method of claim 1 include: A vehicle damage case receiving module, used to establish a communication connection with the vehicle insurance management platform of the first enterprise, receive the vehicle damage case information of the first enterprise, and convert it according to a predetermined data structure to generate first vehicle damage case data; A corpus construction module, used to perform information sorting and data conversion on the first vehicle damage case data to construct a knowledge training corpus, wherein the knowledge training corpus includes a combination of fragmented question-answering techniques; A model training module, used to perform model training based on the knowledge training corpus and the vehicle damage case information according to preset rule logic to generate a first risk model; A rights management module, used to construct the rights management module of the first risk model, wherein the rights management module includes a plurality of rights management centers, and the plurality of rights management centers are used to verify user access rights according to a preset rights attribute set; An authority verification module, used for, when a first user activates the first risk macro model, the authority management module collects identity attributes of the first user, and verifies the access rights of the first user according to the collection results; The risk analysis module is used to analyze and respond to the case analysis request of the first user by the first risk model when the access permission verification is passed, and output risk analysis information.
