AI model-based safety liability insurance full-process digital management platform

Abstract

The invention discloses an AI model-based safety responsibility insurance full-process digital management platform, relates to the technical field of information management, and solves the problems that the trust degree between beneficiaries and an application subject is too low and the supervision difficulty of a supervision department on the application subject is too high. The full-process digital management platform comprises an acquisition and inspection module, a processing and accounting module, an evaluation and claim settlement module and a management and early warning module. Performing matching judgment according to the keyword and the attachment condition through an information matching algorithm so as to judge the responsibility undertaking proportion of the insurance company; dividing a claim proportion for the judged responsibility undertaking proportion and the extracted keywords through a cancel-after-verification monitoring model; the credibility between the applicant and the application subject is improved, and the supervision difficulty of the supervision department on the application subject is reduced.

Description

Technical Field

[0001] The present invention relates to the field of information management technology, and in particular to a full-process digital management platform for occupational liability insurance based on an AI model. Background Art

[0002] With the continuous development and application of intelligent technology and big data, the insurance industry is also actively embracing digital transformation to improve the efficiency and quality of insurance services and meet the personalized needs of customers. In this context, the full-process digital management platform of occupational liability insurance based on AI models came into being. Through the application of AI models and big data technology, the efficiency of occupational liability insurance claims is improved and risks are controlled as early as possible.

[0003] Production safety liability insurance (abbreviated as production safety liability insurance) refers to commercial insurance in which an insurance institution compensates the workers as the beneficiary when a production safety accident occurs to the insured applicant (i.e. an enterprise or organization) causing casualties to workers, and provides accident prevention services to the insured applicant.

[0004] Production safety liability insurance is an insurance that the applicant is responsible for compensating for the death and disability of a production safety accident, and it plays an important role in maintaining social stability and harmony. For areas where high-risk industries are widely distributed and casualties occur from time to time, developing production safety liability insurance and using economic means such as liability insurance to strengthen and improve production safety management are important measures to strengthen the risk control of safety accidents. It is conducive to enhancing production safety awareness, preventing accidents, and promoting the stabilization and improvement of the regional production safety situation; it is conducive to preventing and resolving social contradictions and reducing the relief burden of governments at all levels after accidents; it is conducive to safeguarding the fundamental interests of the people, promoting the healthy operation of the economy, and maintaining social stability.

[0005] Patent No. CN202110504241X A method and system for processing occupational liability insurance based on trusted storage, the method comprising: converting the policy information into a smart contract and publishing it to the blockchain network; after receiving the insurance information, encrypting the insurance information, constructing a ring confidential transaction through the encrypted information, and recording the ring confidential transaction on the corresponding smart contract; the bank node decrypts the account information through its own private key, and performs premium payment operations and / or transfers claims; the medical institution node decrypts the identity information through its own private key, and enters the medical information matching the identity information into the contract. The above-mentioned application scheme uses ring confidential transactions for encryption, which can well protect the privacy information of both parties to the transaction. The bank can only obtain the user's account information, and the medical institution can only obtain the user's identity information. No node can obtain information that is not related to its own business.

[0006] Patent No. CN2024100215581 discloses an information management system and method for production safety liability insurance based on blockchain, including an enterprise side, an insurance company side, a safety liability insurance information management system and a data sharing module, and the present invention relates to the field of data management technology. The above-mentioned information management system and method for production safety liability insurance based on blockchain, by providing a safety liability insurance information management system, is used for data interaction between the enterprise side and the insurance company side, and uses a data acquisition unit to collect data and then transmit it to the compensation settlement model to complete the screening of data, and then check and supplement the retained data, and match the information after the authenticity is confirmed, so as to complete the subsequent calculation of the compensation result, thereby effectively reducing the compensation process for accident handling by personnel after the accident occurs, reducing the error rate of matching selection caused by multiple data, effectively improving the accuracy of information and reducing the time of compensation.

[0007] Patent No. CN2021101530433 discloses a system and method for underwriting occupational liability insurance based on blockchain technology. Among them, the underwriting occupational liability insurance system based on blockchain technology includes: data perception layer, blockchain network layer, data monitoring system, risk analysis system, accident prevention service system and policy management system. In the above invention, the data perception layer screens and uploads the basic data of the insured enterprises to the chain, providing effective and reliable data support for other system modules; the blockchain network layer is used to effectively reduce various friction costs in the process of information transmission and enhance the degree of trust between the participants; it meets the actual needs of the insured enterprises for providing accident prevention services, and improves the safety production level of the enterprises; and meets the needs of the regulatory authorities for providing risk analysis services for enterprises under their jurisdiction.

[0008] Although the above patents all manage the occupational liability insurance of the staff within the applicant, they are unable to conduct comprehensive management of the claims, selection, extension and suspension of insurance contracts based on the status of each employee, and are unable to ensure that the compensation ratio of each insurance contract is fair and just, resulting in frequent insurance compensation disputes and excessively high failure rates of insurance contracts for beneficiaries within the applicant. Summary of the invention

[0009] The purpose of the present invention is to provide a digital management platform for the entire process of occupational liability insurance based on an AI model, which can parse the accident description text in the insurance application form through a keyword extraction algorithm and extract keywords; match and determine the insurance company's liability ratio based on the keywords and attachment conditions through an information matching algorithm; divide the compensation ratio based on the determined liability ratio and the extracted keywords through a write-off monitoring model; so as to reduce the failure rate of enterprise employee insurance contracts and reduce the frequency of insurance compensation disputes.

[0010] The present invention utilizes the following technical solutions:

[0011] A digital management platform for the entire process of occupational liability insurance based on an AI model, including a collection and inspection module, a processing and accounting module, an evaluation and claims module, and a management and early warning module;

[0012] The collection and inspection module is used to use the information collection unit to obtain the insurance application form and supporting documents of the applicant, and use the information inspection unit to perform information judgment on the insurance application form and supporting documents to generate an information judgment result;

[0013] The processing and accounting module is used to pre-process all the information data that have been inspected to obtain the judgment information, and extract the pre-processed judgment information for risk assessment to calculate the insurance premium rate and further generate the insurance contract;

[0014] Evaluation of the claims module, which is used to extract keywords from the insurance application form using a keyword extraction algorithm and assign different weights, determine the insurance company's liability ratio using an information matching algorithm, and predict the compensation ratio of the beneficiary's insurance application form based on the insurance contract using a write-off monitoring model;

[0015] The management warning module is used to supervise the occupational liability insurance of each beneficiary within the applicant entity and to issue warnings to the administrator based on changes in the applicant entity.

[0016] Preferably, the collection and verification module includes an information collection unit and an information verification unit; the information collection unit is used to use a scanning device to collect the completed insurance application form and the supporting documents submitted by the applicant offline, and / or use an online web page to collect the completed insurance application form and the supporting documents submitted by the applicant through a program client; the information verification unit is used to perform information judgment on the collected insurance application form and supporting documents to generate an information judgment result; the applicant includes an enterprise or organization; the supporting documents include a copy of the identity card, a business license, an asset list, a qualification license, an organizational code, a professional qualification certificate and a tax registration certificate.

[0017] Preferably, the workflow of the information verification unit for information judgment is:

[0018] A: Extract the basic credit information of the applicant within N years from the enterprise credit information query tool;

[0019] B: Arrange and associate the extracted basic credit information according to time to construct a subject information change table;

[0020] C: Compare and judge the collected insurance application form and supporting documents with the basic credit information at the latest time node in the subject information change table to generate information judgment results;

[0021] D: If the information is consistent, the insurance application form and supporting documents will be transmitted to the processing and accounting module to generate the insurance contract; if the information is inconsistent, the inconsistent information will be marked and returned on the insurance application form and supporting documents, and the applicant will be notified.

[0022] Preferably, the steps of calculating the insurance premium rate and generating the insurance contract by the processing and accounting module using the collected information data are as follows:

[0023] S1: Divide all collected information data into several data blocks using data block algorithm according to data attributes;

[0024] S2: classify the information data of each data block according to the data type, and assign different sensitivity levels to the classified information data according to the data sensitivity;

[0025] S3: Use data cleaning algorithms to clean and correct abnormal information data after the classification; data cleaning algorithms include data deduplication algorithms, data filling algorithms, data conversion algorithms, and data simplification algorithms;

[0026] S4: According to the data collection time, the internal information data of all data blocks that have completed data cleaning are associated to form a complete application form; and the complete application forms of different applicants are classified and normalized according to the applicants;

[0027] S5: Extract the determination information in the complete application form for risk assessment to generate risk determination results; the determination information includes industry site type, liability ratio, liability limit, historical compensation, remaining insurance period and site size;

[0028] S6: Calculate the insurance premium rate based on the judgment information and risk judgment results: The claim rate is determined by historical claims, the liability limit is the maximum amount of liability borne by the insurance company, and the risk factor is determined by the industry site type, site size and liability ratio;

[0029] S7: Generate an insurance contract based on the basic information of the applicant, the basic information of the insurance company and the attached conditions; the attached conditions include the insurance subject, insurance liability, liability exemption, insurance period, insurance premium rate, insurance amount, compensation processing, dispute resolution and signing and stamping.

[0030] Preferably, the evaluation and claims settlement module includes an accident analysis unit, a responsibility judgment unit and a claims settlement write-off unit; the accident analysis unit uses a keyword extraction algorithm to parse the accident description text in the insurance application form and extract keywords; the keywords include the time of the accident, the responsible party and the casualties; the responsibility judgment unit uses an information matching algorithm to match and judge the extracted keywords with the attachment conditions in the insurance contract to determine the insurance company's liability ratio; the claims settlement write-off unit uses a preset write-off monitoring model to divide the compensation ratio based on the determined liability ratio and the extracted keywords.

[0031] Preferably, the accident analysis unit uses a keyword extraction algorithm to extract keywords from the accident description text, and assigns different responsibility levels to the responsible parties in the following steps:

[0032] a: Segment the accident description text D, perform part-of-speech tagging, and remove the stop word Others;

[0033] b: Retain nouns, verbs, idioms, adjectives and adverbs according to their parts of speech to obtain n candidate keywords t i , D ′ =[t 1 ,t 2 ,…,t i ,…,t n ]; D ′ Represents the accident description text without stop words;

[0034] c: Calculate each candidate keyword t i The word frequency TF in the accident description text D i : Among them, Others represents stop words;

[0035] d: Calculate each candidate keyword t i Inverse document frequency IDF in the insurance database: IDF = log(D m / (D t +1));D m represents the total number of insurance application forms in the insurance database, D t Represents the candidate keyword t in the insurance database i The number of insurance application forms that appeared;

[0036] e: Based on word frequency TF i And inverse document frequency IDF is calculated for each candidate keyword t i The word vector TF i _IDF:TF i _IDF=TF i *IDF;

[0037] f: Reverse the word vectors of each candidate keyword t using a clustering algorithm; i ;

[0038] g: Assign weights to the word vectors arranged in reverse order for clustering according to the approximate relationship between candidate keywords t i , and consider the candidate keywords t with weights between 0.8 and 1.0 as keyword T i ; i ;

[0039] h: Assign different responsibility levels R to the responsible parties according to the active - passive relationship for keyword T i ; The responsibility level R: R = C×(A - P); The responsibility level R includes full active responsibility (A >> P), main active responsibility (A > P), minor passive responsibility (A < P), and no responsibility (A << P); A is the active degree coefficient (0 ≤ A ≤ 1), P is the passive degree coefficient (0 ≤ P ≤ 1), and C is the responsibility intensity coefficient (0 ≤ C ≤ 1).

[0040] Preferably, the steps for the responsibility judgment unit to determine the liability ratio of the insurance company according to the keyword and attachment conditions using the information matching algorithm are as follows:

[0041] First, convert the attachment conditions of the insurance contract into structured data and analyze it; then, extract the responsible parties and compensation ratios in the attachment conditions to construct a compensation liability standard library; subsequently, according to keyword T

[0042] in the accident description text and the responsible parties, use the responsibility level in the compensation liability standard library to judge the liability scope of the responsible parties to obtain the liability scope of each responsible party; finally, classify and count the liability scopes of the insurance company and the beneficiary among the responsible parties respectively to determine the liability ratio of each responsible party, and tabulate the liability ratio of the insurance company. i ;

[0043] Preferably, the steps for the claim cancellation unit to predict the compensation ratio using a preset cancellation monitoring model are as follows:

[0044] The cancellation monitoring model first uses a feature separation layer to split the historical compensation files and legal dispute judgment files to extract historical claim data features; then uses a model construction layer to generalize the historical claim data features in combination with the liability ratio of each responsible party, the compensation liability standard library, and keyword t i to form a generalized compensation data matrix; then uses a learning and training layer to perform several iterative trainings on the generalized compensation data matrix to construct a compensation weight feature matrix; finally uses a determination and detection layer to optimize the compensation weight feature matrix comprehensively to obtain a more excellent prediction of the compensation ratio.

[0045] Preferably, the management warning module utilizes the set information verification interval to monitor the insurance contract according to the beneficiary's employment time, historical claims and remaining policy period, and generates a single form for the beneficiary; based on the resignation time, dispute resolution and claims process progress of certain beneficiaries, the beneficiary's insurance premium rate and insurance amount are recalculated or insurance payment is stopped; at the same time, the single form of the normal employed beneficiaries is regularly sent to the administrator for payment reminders and warnings according to the remaining policy period.

[0046] The present invention parses the accident description text in the insurance application form through a keyword extraction algorithm and extracts keywords; matches and determines the insurance company's liability ratio based on the keywords and attachment conditions through an information matching algorithm; predicts the compensation ratio based on the determined liability ratio and the extracted keywords through a write-off monitoring model; reduces the failure rate of the insurance contract of the applicants within the application entity and reduces the frequency of insurance events. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0048] Figure 1 The principle block diagram of the full-process digital management platform;

[0049] Figure 2 This is the schematic diagram for the keyword algorithm implementation. DETAILED DESCRIPTION

[0050] The present invention is described in detail below with reference to the accompanying drawings and embodiments:

[0051] like Figure 1-Figure 2 As shown, the full-process digital management platform of occupational liability insurance based on AI model described in the present invention includes a collection and inspection module, a processing and accounting module, an evaluation and claims module, and a management and early warning module; wherein,

[0052] The collection and inspection module is used to use the information collection unit to obtain the insurance application form and supporting documents of the applicant, and use the information inspection unit to perform information judgment on the insurance application form and supporting documents to generate an information judgment result;

[0053] The acquisition and inspection module includes an information collection unit and an information inspection unit;

[0054] An information collection unit, used to collect the completed insurance application form and the supporting documents submitted by the applicant offline using a scanning device, and / or collect the completed insurance application form and the supporting documents submitted by the applicant through a program client using an online web page;

[0055] The applicants include enterprises or organizations; the supporting documents include the beneficiary’s ID card, business license, asset list, production safety license, operating license, use license, mining license, beneficiary list, safety evaluation report, organization code, professional qualification certificate and tax registration certificate;

[0056] An information verification unit, used to perform information judgment on the collected insurance application form and certification documents to generate an information judgment result;

[0057] The processing and accounting module is used to pre-process the information data of the completed inspection to obtain the judgment information, and to conduct risk assessment based on the judgment information, calculate the insurance premium rate and further generate it;

[0058] In this embodiment, the information data includes an insurance application form and certification documents;

[0059] Evaluation of the claims module, which is used to extract keywords from the insurance application form using a keyword extraction algorithm and assign different weights, determine the insurance company's liability ratio using an information matching algorithm, and predict the compensation ratio of the beneficiary's insurance application form based on the insurance contract using a write-off monitoring model;

[0060] The evaluation claims module includes the accident analysis unit, the responsibility judgment unit and the claims settlement unit;

[0061] The accident analysis unit uses a keyword extraction algorithm to parse the accident description text in the insurance application form, extract keywords and assign different weights to assign different levels of responsibility to the responsible party; keywords include the time of the accident, the responsible party, and the casualties;

[0062] The liability judgment unit uses an information matching algorithm to match the extracted keywords with the attachment conditions in the insurance contract to determine the liability ratio of the insurance company;

[0063] The claim settlement unit uses a preset settlement monitoring model to divide the compensation ratio according to the determined liability ratio and the extracted keywords;

[0064] In the present invention, the steps of the liability judgment unit using the information matching algorithm to determine the liability ratio of the insurance company according to the keywords and the attachment conditions are as follows:

[0065] First, the additional conditions of the insurance contract are converted into structured data and analyzed. Then, the responsible parties and compensation ratios in the additional conditions are extracted to build a compensation liability standard library. Subsequently, the accident description text is used to identify the responsible parties and compensation ratios. i and the responsible parties, and use the responsibility level in the compensation liability standard library to judge the scope of responsibility of the responsible parties, so as to obtain the scope of responsibility of each responsible party; finally, classify and count the scope of responsibility of the insurance company and the beneficiary among the responsible parties, so as to determine the proportion of responsibility of each responsible party, and tabulate the proportion of responsibility of the insurance company;

[0066] In this embodiment, the structured data is a data table formed by taking the responsible party, compensation ratio and number of clauses as columns and the corresponding data as rows;

[0067] In the present invention, the steps of predicting the compensation ratio by the claim settlement and write-off unit using the preset write-off monitoring model are as follows:

[0068] The write-off monitoring model first uses the feature separation layer to split the historical compensation documents and legal dispute judgment documents to extract the historical claims data features; then uses the model building layer to combine the responsibility ratio of each responsible party, the compensation liability standard library and the keyword T i Generalize the historical claims data features to form a generalized claims data matrix; then use the learning training layer to iteratively train the generalized claims data matrix several times to construct a claims weight feature matrix; finally, use the judgment detection layer to comprehensively optimize the claims weight feature matrix to obtain a better prediction of the claims ratio;

[0069] In this embodiment, the feature separation layer includes three extraction branches. The feature separation layer uses the first two extraction branches to split and extract the compensation ratio and the compensation judgment result in the historical compensation file and the legal dispute judgment file respectively, and uses the third extraction branch to fuse the extracted and split compensation ratio data and the compensation judgment result data; the first extraction branch is used to extract the compensation ratio of the historical compensation file to form a compensation ratio data matrix; the first extraction branch includes 3 3X3 convolutional layers, 2 3X3 maximum pooling layers and 3 normalization layers; the second extraction branch is used to extract the compensation judgment result of the legal dispute judgment file, To form a compensation judgment result data matrix; the second extraction branch includes 2 5X5 convolution layers, 2 1X1 convolution layers, 3 3X3 average pooling layers, 3 1X1 average pooling layers and 3 normalization layers; the third extraction branch fuses and extracts the compensation ratio data matrix and the compensation judgment result data matrix to form a deep compensation data matrix; the third extraction branch includes 4 3X3 convolution layers, 3 5X5 convolution layers, 3 3X3 maximum pooling layers and 2 normalization layers; the 4 3X3 convolution layers are connected in parallel with the 3 5X5 convolution layers and 3 3X3 maximum pooling layers in series, and are connected in series with the 2 normalization layers;

[0070] The model construction layer includes two construction branches. The model construction layer uses the first construction branch to combine the liability ratio of each responsible party and the compensation liability standard library to fill the deep compensation data matrix with information; the first construction branch includes 3 3X3 edge convolution layers, 2 5X5 edge convolution layers, 2 1X1 edge convolution layers, 2 3X3 edge pooling layers and 2 3X3 inverse maximum edge pooling layers; 2 3X3 edge pooling layers and 2 3X3 inverse maximum edge pooling layers are trained using the grid simplification algorithm to fill the deep compensation data matrix; and then the second construction branch is used to perform generalization training on the keywords. Then, a generalized compensation data matrix is ​​obtained; the second construction branch includes 4 5X5 edge convolution layers, 2 1X1 edge convolution layers, 4 3X3 edge pooling layers, 3 3X3 inverse maximum edge pooling layers, a neural fusion mechanism, a wrapping deformation branch and a compensating deformation branch to complete the construction of the generalized compensation data matrix; the second branch uses the neural fusion shape mechanism to splice the wrapping deformation branch and the compensating deformation branch to each other, the wrapping deformation branch is connected in series with 2 5X5 edge convolution layers and 1 1X1 edge convolution layer, and the compensating deformation branch is connected in parallel with the 2 5X5 edge convolution layers and 1 1X1 edge convolution layer connected in series;

[0071] The learning and training layer uses the multi-scale attention mechanism and the position attention mechanism to perform several iterative training on the generalized compensation data matrix to obtain the compensation weight feature matrix; the learning and training layer includes 2 series-connected RES-ECA blocks, 3 3X3 convolutional layers, 3 1X1 convolutional layers, and 2 Relu activation functions; the RES-ECA block includes 4 3X3 convolutional layers, 6 residual blocks of parallel Inception V3 blocks, 3 global pooling layers, and 3 Sigmoid activation functions;

[0072] The judgment detection layer performs all-round optimization on the payout weight feature matrix; the judgment detection layer includes 3 InceptionV1 blocks, 2 Inception V2 blocks, 2 5X5 upsampling layers, 1 3X3 average pooling layer and 2 MISH activation functions;

[0073] In this embodiment, the working principle of the write-off monitoring model is as follows:

[0074] First, historical claims documents and legal dispute judgment documents are classified and separated according to certain rules to facilitate subsequent feature extraction. Relevant features of historical claims data are extracted from these documents, which may include the amount of compensation, type of compensation, time of compensation, information on the responsible party, etc.

[0075] Then, according to the liability ratio of each responsible party, the extracted features are matched with the specific responsibilities of the responsible party; the compensation liability standard library is used to standardize the features to ensure that all features are processed according to the unified compensation standards; all keywords are combined with features to create a more comprehensive data representation, which helps the model understand and generalize the compensation data; through the above steps, the model forms a generalized compensation data matrix, which contains generalized features and provides a data basis for subsequent training.

[0076] Then, the generalized claims data matrix is ​​trained multiple times using a machine learning algorithm to learn the relationship and weights between features. A claims weight feature matrix is ​​constructed through the training model, which contains the importance weight of each feature for claims prediction.

[0077] Finally, a variety of optimization techniques may be used, such as adjusting weights, using regularization methods, improving algorithm parameters, etc., to improve prediction performance; through the optimized claim weight feature matrix, the model can more accurately predict future claim ratios, thereby providing decision support for write-off monitoring.

[0078] The management warning module is used to supervise the safety liability insurance of each beneficiary within the applicant entity and issue warnings to the administrator based on changes in the applicant entity;

[0079] In the present invention, the management warning module uses the set information verification interval to monitor the insurance contract according to the beneficiary's employment time, historical claims and remaining insurance period, and generates a single form for the beneficiary; according to the resignation time, dispute handling and claim process progress of certain beneficiaries, the insurance premium rate and insurance amount of the beneficiary are recalculated or insurance payment is stopped; at the same time, the single form of the normal working beneficiaries is regularly sent to the administrator for payment reminders and warnings according to the remaining insurance period;

[0080] In the present invention, the claim settlement process includes reporting, investigation and loss assessment, document settlement and verification, approval and settlement of claims;

[0081] In the present invention, the workflow of the information verification unit for information judgment is as follows:

[0082] A: Extract the basic credit information of the applicant within N years from the enterprise credit information query tool;

[0083] Among them, basic credit information includes the profile information of the enterprise or organization, operating conditions, financial conditions, contract performance, social responsibility performance and records of violations of laws and regulations;

[0084] B: Arrange and associate the extracted basic credit information according to time to construct a subject information change table;

[0085] C: Compare and judge the collected insurance application form and supporting documents with the basic credit information at the latest time node in the subject information change table to generate information judgment results;

[0086] D: If the information is consistent, the insurance application form and the supporting documents will be transmitted to the processing and accounting module to generate the insurance contract; if the information is inconsistent, the inconsistent information will be marked on the insurance application form and the supporting documents, returned, and the applicant will be notified;

[0087] In the present invention, the steps of calculating the insurance premium rate and generating the insurance contract by the processing and accounting module using the collected information data are as follows:

[0088] S1: Divide all collected information data into several data blocks using data block algorithm according to data attributes;

[0089] Among them, data attributes include beneficiaries and applicants;

[0090] S2: classify the information data of each data block according to the data type, and assign different sensitivity levels to the classified information data according to the data sensitivity;

[0091] In the present invention, data types include character type and numeric type; data sensitivity refers to the confidentiality level of data, including confidentiality level, privacy level and public level;

[0092] S3: Use data cleaning algorithms to clean and correct abnormal information data after the classification; data cleaning algorithms include data deduplication algorithms, data filling algorithms, data conversion algorithms, and data simplification algorithms;

[0093] S4: According to the data collection time, the internal information data of all data blocks that have completed data cleaning are associated to form a complete application form; and the complete application forms of different applicants are classified and normalized according to the applicants;

[0094] S5: Extract the determination information in the complete application form for risk assessment to generate risk determination results; the determination information includes industry site type, liability ratio, liability limit, historical compensation, remaining insurance period and site size;

[0095] S6: Calculate the insurance premium rate based on the judgment information and risk judgment results: The claim rate is determined by historical claims, the liability limit is the maximum amount of liability borne by the insurance company, and the risk factor is determined by the industry site type, site size and liability ratio;

[0096] In this embodiment, the above insurance premium rate calculation method is a conventional technology.

[0097] S7: Generate an insurance contract based on the basic information of the applicant, the basic information of the insurance company and the attached conditions;

[0098] Among them, the attached conditions include the subject matter of insurance, insurance liability, liability exemption, insurance period, insurance premium rate, insurance amount, compensation processing, dispute resolution and signing and stamping;

[0099] In the present invention, the accident analysis unit uses a keyword extraction algorithm to extract keywords from the accident description text, and assigns different responsibility levels to the responsible parties in the following steps:

[0100] a: Segment the accident description text D, perform part-of-speech tagging, and remove the stop word Others;

[0101] b: Retain nouns, verbs, idioms, adjectives and adverbs according to their parts of speech to obtain n candidate keywords t i , D ′ =[t 1 ,t 2 ,…,t i ,…,t n ]; D ′ Represents the accident description text without stop words;

[0102] c: Calculate each candidate keyword t i The word frequency TF in the accident description text D i : Among them, Others represents stop words;

[0103] d: Calculate each candidate keyword t i Inverse document frequency IDF in the insurance database: IDF = log(D m / (D t +1));D m represents the total number of insurance application forms in the insurance database, D t Represents the candidate keyword t in the insurance database i The number of insurance application forms that appeared;

[0104] In the present invention, the insurance database is a data center that centrally manages and stores insurance products and related information, and is intended to provide convenient information query and data analysis services for insurance companies, regulatory agencies and consumers;

[0105] e: Based on word frequency TF i And inverse document frequency IDF is calculated for each candidate keyword t i The word vector TF i _IDF:TF i _IDF=TF i *IDF;

[0106] f: Reverse the word vectors of each candidate keyword t using a clustering algorithm; i ;

[0107] g: Assign weights to the reverse-ordered word vectors according to the approximate degree relationship between candidate keywords t i , and consider the candidate keywords t with weights between 0.8 and 1.0 as keyword T i ; i ;

[0108] In this embodiment, the steps for determining weights using the approximate degree relationship are as follows:

[0109] g1: Standardize the approximate degrees between candidate keywords to obtain a standard score:

[0110]

[0111] where Weight(k) represents the standard score of the k-th approximate degree, Similarity(k) represents the k-th approximate degree, and Similarity(all) represents the approximate degrees between all candidate keywords;

[0112] g2: Reverse-order the standardized approximate degree scores to ensure that the most similar word vectors have the highest weights;

[0113] g3: Assign corresponding weights to each word vector according to the reverse-ordered result;

[0114] For example, the most similar word vector is assigned the highest weight, and as the similarity decreases, the weight also decreases;

[0115] g4: Perform clustering analysis on the weighted word vectors using a hierarchical clustering algorithm, and correct the weights of each word vector using a PCA linear algorithm;

[0116] h: Assign different responsibility levels R to the responsible parties according to the active-passive relationship of keyword T i ;

[0117] Responsibility level R: R = A - P; The responsibility level R includes full active responsibility (A >> P), major active responsibility (A > P), minor passive responsibility (A < P), and no responsibility (A << P); A is the active degree coefficient (0 ≤ A ≤ 1), and P is the passive degree coefficient (0 ≤ P ≤ 1);

[0118] In this embodiment, full active liability: The liable party caused the accident due to willful act or gross negligence and bears the highest level of liability; major active liability: The liable party caused the accident due to ordinary negligence or oversight and bears a relatively high level of liability; minor passive liability: The liable party had minor negligence in the accident but was not the main cause of the accident; no liability: The liable party had no negligence in the accident or was completely a passive victim.

[0119] Full active liability: A >> P, that is, A = 1 and P = 0; major active liability: A > P, that is, A takes a relatively high value (such as 0.7 - 0.9) and P takes a relatively low value (such as 0.1 - 0.3); minor passive liability: A < P, that is, A takes a relatively low value (such as 0 - 0.3) and P takes a relatively high value (such as 0.7 - 1); no liability: A << P, that is, A = 0 and P = 1.

[0120] Among them, positive numbers represent full active liability or major active liability; negative numbers represent minor passive liability or no liability.

[0121] Embodiment:

[0122] First, the information collection unit of the inspection module collects, through the scanning device, the completed insurance application form and provided supporting documents submitted offline by the applicant entity, and / or collects, through the online web page, the completed insurance application form and provided supporting documents submitted by the applicant entity through the program client; the applicant entity includes enterprises or organizations; the supporting documents include copies of identity cards, business licenses, asset lists, qualification licenses, organization codes, professional qualification certificates, and tax registration certificates.

[0123] The information inspection unit uses an enterprise credit information query tool to extract the basic credit information of the applicant entity in the past N years; compares and judges the collected insurance application form and supporting documents with the entity information change form to generate an information judgment result; if the information is consistent, transmits the insurance application form and supporting documents to the processing and accounting module for generating an insurance contract; if the information is inconsistent, marks and returns the inconsistent information on the insurance application form and supporting documents and notifies the applicant entity.

[0124] Then, according to the data attributes, all the collected information data are divided into several data blocks using the data segmentation algorithm; the information data of each data block is classified according to the data type, and the classified information data is assigned different sensitivity levels according to the data sensitivity; the classified information data is cleaned and corrected according to the same content in other types of information data using the data cleaning algorithm; the data cleaning algorithm includes data deduplication algorithm, data filling algorithm, data conversion algorithm and data simplification algorithm; according to the data collection time, the internal information data of all the data blocks that have completed data cleaning are associated to form a complete application form; and the complete application forms of different applicants are classified and normalized according to the applicant; the type of information industry venue, liability ratio, liability limit, historical compensation, remaining insurance period and venue scale in the complete application form are extracted for risk assessment to generate risk determination results;

[0125] Calculate the insurance premium rate based on the judgment information and risk judgment results: The claim rate is determined by historical claims, the liability limit is the maximum amount of liability borne by the insurance company, and the risk factor is determined by the industry site type, site size and liability ratio;

[0126] Generate an insurance contract based on the basic information of the applicant, the insurance company and the attached conditions; the attached conditions include the subject of insurance, insurance liability, liability exemption, insurance period, insurance premium rate, insurance amount, compensation processing, dispute resolution and signing and stamping;

[0127] First, the additional conditions of the insurance contract are converted into structured data and analyzed. Then, the responsible parties and compensation ratios in the additional conditions are extracted to build a compensation liability standard library. Subsequently, the accident description text is used to identify the responsible parties and compensation ratios. i and the responsible parties, and use the responsibility level in the compensation liability standard library to judge the scope of responsibility of the responsible parties, so as to obtain the scope of responsibility of each responsible party; finally, classify and count the scope of responsibility of the insurance company and the beneficiary among the responsible parties, so as to determine the proportion of responsibility of each responsible party, and tabulate the proportion of responsibility of the insurance company;

[0128] Finally, the management and early warning module uses the set information verification interval to monitor the insurance contract according to the beneficiary's employment time, historical claims and remaining policy period, and generates a single form for the beneficiary; based on the resignation time, dispute resolution and claims process progress of certain beneficiaries, the beneficiary's insurance premium rate and insurance amount are recalculated or insurance payment is stopped; at the same time, the single form of normal employed beneficiaries is regularly sent to the administrator for payment reminders and early warnings based on the remaining policy period.

Claims

1. A digital management platform for the entire process of occupational liability insurance based on an AI model, characterized by: It includes collection and inspection module, processing and accounting module, evaluation and claims module and management and early warning module; among them, The collection and inspection module is used to use the information collection unit to obtain the insurance application form and supporting documents of the applicant, and use the information inspection unit to perform information judgment on the insurance application form and supporting documents to generate an information judgment result; The processing and accounting module is used to pre-process all the information data that have been inspected to obtain the judgment information, and extract the pre-processed judgment information for risk assessment to calculate the insurance premium rate and further generate the insurance contract; Evaluation of the claims module, which is used to extract keywords from the insurance application form using a keyword extraction algorithm and assign different weights, determine the insurance company's liability ratio using an information matching algorithm, and predict the compensation ratio of the beneficiary's insurance application form based on the insurance contract using a write-off monitoring model; The management warning module is used to supervise the occupational liability insurance of each beneficiary within the applicant entity and to issue warnings to the administrator based on changes in the applicant entity.

2. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 1 is characterized by: The collection and verification module includes an information collection unit and an information verification unit; the information collection unit is used to collect the completed insurance application form and the supporting documents submitted by the applicant offline by using a scanning device, and / or collect the completed insurance application form and the supporting documents submitted by the applicant through the program client by using an online web page; The information verification unit is used to perform information judgment on the collected insurance application forms and supporting documents to generate information judgment results.

3. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 2 is characterized by: The workflow of the information verification unit for information judgment is as follows: A: Extract the basic credit information of the applicant within N years from the enterprise credit information query tool; B: Arrange and associate the extracted basic credit information according to time to construct a subject information change table; C: Compare and judge the collected insurance application form and supporting documents with the basic credit information at the latest time node in the subject information change table to generate information judgment results; D: If the information is consistent, the insurance application form and supporting documents will be transmitted to the processing and accounting module to generate the insurance contract; if the information is inconsistent, the inconsistent information will be marked and returned on the insurance application form and supporting documents, and the applicant will be notified.

4. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 1 is characterized by: The steps of the processing and accounting module using the collected information data to calculate the insurance premium rate and generate the insurance contract are as follows: S1: Divide all collected information data into several data blocks using a data block algorithm according to data attributes; the information data includes insurance application forms and certification documents; S2: classify the information data of each data block according to the data type, and assign different sensitivity levels to the classified information data according to the data sensitivity; S3: Use data cleaning algorithms to clean and correct abnormal information data after the classification; data cleaning algorithms include data deduplication algorithms, data filling algorithms, data conversion algorithms, and data simplification algorithms; S4: According to the data collection time, the internal information data of all data blocks that have completed data cleaning are associated to form a complete application form; and the complete application forms of different applicants are classified and normalized according to the applicants; S5: Extract the judgment information in the complete application form for risk assessment to generate a risk judgment result; the judgment information includes the type of industry venue, liability assumption ratio, liability limit, historical compensation, remaining insurance period, and venue scale; S6: Calculate the insurance premium rate based on the judgment information and risk judgment results: The claim rate is determined by historical claims, the liability limit is the maximum amount of liability borne by the insurance company, and the risk factor is determined by the industry site type, site size and liability ratio; S7: Generate an insurance contract based on the basic information of the applicant, the basic information of the insurance company, and the attachment conditions.

5. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 1 is characterized by: The evaluation and claims settlement module includes an accident analysis unit, a liability judgment unit, and a claims settlement and cancellation unit; the accident analysis unit uses a keyword extraction algorithm to parse the accident description text in the insurance application form and extract keywords; the keywords include the accident occurrence time, the liable party, and the casualty situation; the liability judgment unit uses an information matching algorithm to match and judge according to the extracted keywords and the attachment conditions in the insurance contract to determine the liability assumption ratio of the insurance company; the claims settlement and cancellation unit uses a preset cancellation monitoring model to divide the compensation ratio based on the determined liability assumption ratio and the extracted keywords.

6. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 5 is characterized by: The steps for the accident analysis unit to use the keyword extraction algorithm to extract keywords in the accident description text and assign different liability levels to the liable party are: a: Segment the accident description text D, perform part-of-speech tagging, and remove stop words Others; b: Retain nouns, verbs, idioms, adjectives and adverbs according to their parts of speech to obtain n candidate keywords t i , D ′ =[t1,t2,…,t i ,…,t n ]; D ′ Represents the accident description text without stop words; c: Calculate each candidate keyword t i The word frequency TF in the accident description text D i : Among them, Others represents stop words; d: Calculate each candidate keyword t i Inverse document frequency IDF in the insurance database: IDF = log(D m / (D t +1));D m represents the total number of insurance application forms in the insurance database, D t Represents the candidate keyword t in the insurance database i The number of insurance application forms that appeared; e: Based on word frequency TF i And inverse document frequency IDF is calculated for each candidate keyword t i The word vector TF i _IDF:TF i _IDF=TF i *IDF; f: Use clustering algorithm to cluster each candidate keyword t i The word vectors are arranged in reverse order; g: According to the candidate keyword t i The similarity relationship between them assigns weights to the word vectors of the clusters in reverse order, and assigns candidate keywords t with weights between 0.8 and 1.0 i Treat as keyword T i ; h: Keyword T i Different levels of responsibility R are assigned to the responsible parties according to the active-passive relationship; Liability level R: R = A - P; the liability level R includes full active liability (A >> P), major active liability (A > P), minor passive liability (A < P), and no liability (A << P); A is the active degree coefficient (0 ≤ A ≤ 1), and P is the passive degree coefficient (0 ≤ P ≤ 1).

7. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 5 is characterized by: The steps for the liability judgment unit to use the information matching algorithm to determine the liability assumption ratio of the insurance company according to the keywords and the attachment conditions are: First, the additional conditions of the insurance contract are converted into structured data and analyzed. Then, the responsible parties and compensation ratios in the additional conditions are extracted to build a compensation liability standard library. Subsequently, the accident description text is used to identify the responsible parties and compensation ratios. i and the responsible parties, and use the liability level in the compensation liability standard library to judge the liability scope of the responsible parties to obtain the liability scope of each responsible party; finally, the liability scope of the insurance company and the beneficiary among the responsible parties are classified and counted separately to determine the liability ratio of each responsible party, and the liability ratio of the insurance company is tabulated.

8. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 5 is characterized by: The steps for the claims settlement and cancellation unit to use the preset cancellation monitoring model to predict the compensation ratio are: The write-off monitoring model first uses the feature separation layer to split the historical compensation documents and legal dispute judgment documents to extract the historical claims data features; then uses the model building layer to combine the responsibility ratio of each responsible party, the compensation liability standard library and the keyword T i Generalize the historical claims data features to form a generalized claims data matrix; Then use the learning and training layer to perform several iterative trainings on the generalization compensation data matrix to construct a compensation weight feature matrix; Finally, use the judgment and detection layer to perform all-round optimization on the compensation weight feature matrix to obtain a more excellent compensation ratio prediction.

9. The AI ​​model-based full-process digital management platform for occupational liability insurance according to claim 1 is characterized by: The management and warning module uses the set information verification interval to monitor the insurance contract according to the beneficiary's employment time, historical compensation, and remaining insurance period, and generate a single form for the beneficiary; according to the departure time, dispute handling, and claims settlement process progress of some beneficiaries, recalculate the insurance rate and insurance amount of the beneficiary or stop the insurance premium payment; at the same time, for the single form of the normally employed beneficiary, regularly send it to the administrator according to the remaining insurance period for payment reminder and warning.

Images