Intelligent insurance matching method and system combining user portrait and clause knowledge base

Abstract

The application relates to the technical field of artificial intelligence, and provides an intelligent insurance matching method and system combining a user portrait and a clause knowledge base. The method comprises the following steps: constructing a first portrait of a user based on a user data set and optimizing a second portrait; constructing a product portrait by multi-dimensionally extracting features from an insurance clause knowledge base; performing demand coverage risk evaluation on the product portrait based on the second portrait of the user, generating a demand coverage risk matrix; performing guarantee conflict risk evaluation, generating a guarantee conflict risk matrix; introducing a risk analysis channel, combining the second portrait of the user to generate a risk matrix; and fusing the three types of risk matrices to perform multi-level optimization matching, generating an insurance optimization matching graph, so as to solve the technical problem that there is a significant deviation between a recommended result and actual risk preference and guarantee demand of a user, realize dynamic and accurate matching of an insurance product and user demand, reduce guarantee conflict risk and improve the relevance of a recommended result, and achieve the technical effect of meeting the demand of personalized insurance configuration.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to an intelligent insurance matching method and system that combines user profiles with a policy knowledge base. Background Technology

[0002] In traditional insurance service scenarios, the core dilemma faced by users lies in the complexity of insurance terms and the inefficiency of matching needs. Insurance products typically contain a large number of technical terms, exclusions, and underwriting rules, making it difficult for ordinary users to fully understand their coverage and limitations. At the same time, users' own protection needs, risk preferences, and health conditions are highly personalized. Existing recommendation models mainly rely on the experience of human advisors or simple rule engines for filtering, which has several problems: First, human services struggle to process massive amounts of terms and dynamically changing user data at scale, resulting in narrow recommendation coverage; second, rule engines can only match explicit tags (such as age and insurance type), failing to deeply analyze the correlation between the terms' coverage and the user's implicit needs; third, they easily overlook the risks of conflicting coverage and underwriting limitations between products, leading to insufficient coverage or duplicate insurance. Summary of the Invention

[0003] This application provides an intelligent insurance matching method and system that combines user profiles and a policy knowledge base. It aims to solve the technical problems of existing insurance recommendation methods that fail to accurately match user needs, ignore conflicts between policy terms, and result in significant deviations between recommendation results and users' actual risk preferences and protection needs. The system achieves the technical effect of dynamically and accurately matching insurance products with user needs through multi-level optimization matching of a multi-dimensional risk matrix, reducing the risk of protection conflicts, improving the relevance of recommendation results, and meeting the needs of personalized insurance configuration.

[0004] The first aspect disclosed in this application provides an intelligent insurance matching method combining user profiles and a policy knowledge base. The method includes: constructing a first user profile based on a user dataset of a target user, and optimizing coverage needs based on the first user profile to obtain a second user profile; extracting and organizing multi-dimensional knowledge from an insurance policy knowledge base to construct profiles for various insurance products, wherein the insurance policy knowledge base includes multiple policy knowledge sets for multiple insurance products; evaluating the coverage risk of each insurance product profile based on the second user profile to obtain a coverage risk matrix; evaluating the coverage conflict risk of each insurance product profile based on the second user profile to obtain a coverage conflict risk matrix; introducing an underwriting risk analysis channel, and evaluating the underwriting risk of each insurance product profile based on the second user profile to obtain an underwriting risk matrix; and performing multi-level optimization matching of the multiple insurance products based on the coverage risk matrix, the coverage conflict risk matrix, and the underwriting risk matrix to obtain an insurance optimization matching graph.

[0005] Another aspect disclosed in this application provides an intelligent insurance matching system that combines user profiles and a policy knowledge base. The system includes: a user profile construction module: constructing a first user profile based on a user dataset of the target user, and optimizing protection needs based on the first user profile to obtain a second user profile; a knowledge organization module: extracting and organizing multi-dimensional knowledge based on an insurance policy knowledge base to construct profiles for each insurance product, wherein the insurance policy knowledge base includes multiple sets of insurance policy knowledge for multiple insurance products; a coverage risk assessment module: performing a coverage risk assessment on each insurance product profile based on the second user profile to obtain a coverage risk matrix; a conflict risk assessment module: performing a protection conflict risk assessment on each insurance product profile based on the second user profile to obtain a protection conflict risk matrix; an underwriting risk assessment module: introducing an underwriting risk analysis channel, and combining the second user profile with the underwriting risk assessment of each insurance product profile to obtain an underwriting risk matrix; and a multi-level optimization matching module: performing multi-level optimization matching on the multiple insurance products based on the coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix to obtain an insurance optimization matching graph.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0007] The aforementioned intelligent insurance matching method, combining user profiling and a policy knowledge base, first constructs a first user profile based on the target user's personal data. Then, by analyzing the user's needs and preferences, it optimizes the user's protection requirements to obtain a second user profile, reflecting a more precise understanding of the user's needs and risk tolerance. Subsequently, it extracts relevant information about insurance products from the insurance policy knowledge base, performing multi-dimensional knowledge extraction and organization to form a product profile describing the various terms and characteristics of different insurance products. Next, based on the second user profile, it conducts a multi-dimensional risk assessment of the insurance product profile, including assessments of needs coverage risk, protection conflict risk, and underwriting risk, thus forming multiple risk matrices. Finally, it performs multi-level optimization based on these risk matrices, ultimately obtaining the most suitable insurance product recommendation graph for the user through optimal matching. This intelligently recommends insurance products that best meet the user's needs, optimizes protection content, and reduces potential risks.

[0008] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating an intelligent insurance matching method that combines user profiles with a terms knowledge base in one embodiment.

[0011] Figure 2 This is an architecture diagram of an intelligent insurance matching system that combines user profiles and a terms knowledge base in one embodiment.

[0012] Figure labeling: User profile building module 11, knowledge organization module 12, coverage risk assessment module 13, conflict risk assessment module 14, underwriting risk assessment module 15, multi-layer optimization matching module 16. Detailed Implementation

[0013] This application provides an intelligent insurance matching method and system that combines user profiles and a policy knowledge base. This addresses the technical problem that existing insurance recommendation methods cannot accurately match user needs, ignore conflicts between policy terms, and result in significant deviations between recommendation results and users' actual risk preferences and protection needs. The system achieves the technical effect of dynamically and accurately matching insurance products with user needs through multi-level optimization matching of a multi-dimensional risk matrix, reducing the risk of protection conflicts, improving the relevance of recommendation results, and meeting the needs of personalized insurance configuration.

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0015] It should be noted that the terms “comprising” and “having”, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such process, method, product, or device.

[0016] Example 1, as Figure 1 As shown, this application provides an intelligent insurance matching method that combines user profiles with a terms knowledge base, the method comprising:

[0017] Based on the target user's user dataset, construct a first user profile, and optimize the guarantee requirements based on the first user profile to obtain a second user profile.

[0018] In this embodiment, a user dataset of the target users is first collected. This dataset includes the user's basic personal information, historical behavioral data, insurance purchase records, family structure, occupation, income level, etc. By cleaning and performing correlation analysis on the collected user dataset, a first user profile of the target users is constructed. This first user profile provides preliminary information about the user. After constructing the first user profile, a second user profile is formed by combining the user's protection needs prediction results with the constructed first user profile through in-depth analysis. This second user profile accurately reflects the specific needs in different protection areas (such as life insurance, health insurance, accident insurance, etc.) and the user's tolerance for different risk levels, thus providing more accurate basic data for subsequent intelligent matching and product recommendations.

[0019] Furthermore, this application provides a method for constructing a first user profile based on a user dataset of the target users, including:

[0020] The user dataset is cleaned to obtain a user feature set, which includes a set of basic user information and a set of historical insurance behaviors. Insurance-related sensitivity is evaluated based on the user feature set to obtain a feature sensitivity evaluation sequence. The user feature set is then fitted with a profile based on the feature sensitivity evaluation sequence to generate the first user profile.

[0021] Preferably, after collecting the user dataset, to ensure data quality and accuracy, the user dataset is cleaned through steps such as removing redundant data, handling missing values, filtering outliers, and standardizing data format. Specifically, when handling missing values, for continuous variables, the mean or median can be used for padding; for categorical variables, the mode can be used, or "0" or "none" can be used to indicate that the item is not provided or is not applicable. When filtering outliers, the Z-Score of each data point is calculated and compared with a preset threshold to remove data points exceeding that threshold. When standardizing data format, for continuous variables, Z-score normalization or maximum-minimum normalization can be used; for categorical variables, One-Hot encoding can be used. After data cleaning, data can be extracted according to preset parameter names to construct a user feature set. This user feature set includes a basic user information set and a historical insurance behavior set. The basic user information set includes the target user's personal information, such as name, age, gender, marital status, income level, and occupation. The historical insurance behavior set includes the user's insurance purchase records, policy type, insured amount, coverage, policy purchase time, and claims records. Subsequently, an insurance correlation sensitivity evaluation is performed based on the obtained user feature set. The purpose of this step is to analyze which user characteristics have a high correlation with insurance needs and identify the main factors influencing users' insurance choices. Specifically, for continuous characteristics and continuous insurance needs, such as using... and Let n represent the income and premium of the i-th user, respectively, and let n represent the total number of users. and These are the mean values ​​of income and premium, respectively, calculated using the Pearson correlation coefficient formula: The Pearson correlation coefficient r between income and premium is obtained to measure the sensitivity of their insurance association, with a value ranging from -1 to 1. For ease of subsequent use, a standardized formula will be used: The insurance-related sensitivity is standardized and mapped to the interval [0,1], where S represents the standardized insurance-related sensitivity. -1, The threshold is 1. For categorical features and categorical insurance needs, such as occupation and insurance type preference, the insurance correlation sensitivity between the two can be quantified by statistically analyzing the proportion of users in different occupations purchasing each type of insurance. For continuous features and categorical insurance needs, such as age and underwriting results, users can be segmented by age, the underwriting pass rate for each group can be calculated, and then verified using a t-test or analysis of variance (ANOVA) to calculate the p-value. The p-value is then quantified as the insurance correlation sensitivity between the two through a non-linear mapping. Next, the largest insurance correlation sensitivity is extracted from multiple insurance correlation sensitivities for each user feature, and the user features are sorted in descending order according to these extracted insurance correlation sensitivities to obtain a feature sensitivity evaluation sequence. This feature sensitivity evaluation sequence represents the sensitivity of a user feature to its insurance needs. Then, based on this feature sensitivity evaluation sequence, a user profile is fitted to the user feature set. Specifically, features with low insurance relevance (below a preset sensitivity threshold) are removed from the feature sensitivity evaluation sequence. The remaining features are then mapped to a low-dimensional vector space, and core user information is calculated through density clustering, generating tagged descriptions, such as a 30-year-old engineer with a monthly income of 20,000 yuan who is interested in critical illness insurance coverage. Finally, the tagged descriptions are stored in JSON format, forming the final first user profile. This first user profile includes, but is not limited to, the user's basic information, insurance needs and preferences, and risk tolerance, reflecting the comprehensive characteristics and needs of the target user and providing accurate data support for subsequent insurance product recommendations.

[0022] Furthermore, this application provides a method for optimizing security requirements based on the first user profile to obtain a second user profile, including:

[0023] Based on the first user profile, multi-dimensional protection needs are predicted to obtain a protection needs profile; the first user profile and the protection needs profile are merged to generate the second user profile.

[0024] Preferably, after obtaining the first user profile, the labeled information in the first user profile is extracted, and the extracted feature information is then input into a protection demand prediction network pre-trained based on a Long Short-Term Memory (LSTM) network. The protection demand prediction network analyzes the received feature information according to the mapping relationship learned from the sample data to obtain protection demand feature information, and organizes this protection demand feature information into a protection demand profile. The sample data usually comes from historical insurance policy data, which typically includes multi-dimensional feature information on the user side (such as age, gender, occupation, income level, family structure, health status, etc.), product-side label information (such as insurance product category, type of coverage, coverage range, premium level, etc.), and information on the user's actual purchase behavior. The protection demand feature information includes possible future insurance types, coverage ranges, premium budgets, etc. Before training the protection demand prediction network, sample user profiles (the user's first profile at a certain moment) and sample protection demand profiles (subsequent actual insurance purchases by the user) are collected, and these sample data are input into the LSTM. Iterative training is performed through forward propagation, loss calculation, backpropagation, parameter optimization, and other steps until the maximum number of iterations is met. After training, validation is performed using data not used for training. If the validation results show that the accuracy is greater than the preset accuracy, the current LSTM is stored as the final insurance demand prediction network; otherwise, hyperparameters such as the learning rate and the number of training batches are adjusted for optimization. After obtaining the insurance demand profile, the various feature information in the first user profile is merged and concatenated with the predicted demand feature information in the insurance demand profile to construct the final second user profile. This second user profile not only contains the target user's personal information but also clearly reflects the insurance content the user may need in the future and their risk tolerance. This will serve as the basis for subsequent intelligent insurance recommendations and risk assessments, helping to accurately match insurance products that meet the user's needs and provide customized insurance services based on the user's requirements.

[0025] Based on the insurance terms knowledge base, multi-dimensional knowledge is extracted and sorted to construct profiles for each insurance product. The insurance terms knowledge base includes multiple sets of insurance terms knowledge for multiple insurance products.

[0026] In one embodiment, to avoid excessively long retrieval times due to retrieving more than the limit of insurance clauses at once in real-world scenarios, after accessing the insurance clause knowledge base through a preset interface, a rapid tag-based matching retrieval is performed in the insurance clause knowledge base based on key tag information from the user's second profile, such as age, occupation, health status, income level, and type of protection needs. Boolean filtering is used to filter out insurance products that broadly match the user's characteristics in terms of tags. For example, when the user profile includes tags such as "family-oriented," "critical illness protection needs," and "middle income," a set of insurance products with similar product attribute tags is prioritized. These insurance products are then aggregated to form a preliminary set of matched insurance clauses. This insurance clause knowledge set includes the insurance product's attribute information, coverage information, and exclusion information. Subsequently, multi-dimensional knowledge extraction is performed on the matched insurance clause knowledge sets according to insurance knowledge elements to obtain specific data corresponding to each insurance knowledge element. The extracted data is then enhanced to create a detailed profile for each insurance product. Each profile includes basic information, coverage, and exclusions for the corresponding insurance product, providing data support for subsequent intelligent insurance matching and ensuring that users select the insurance product that best suits their needs.

[0027] Furthermore, this application provides a method for extracting and organizing multi-dimensional knowledge based on an insurance terms knowledge base to construct profiles for various insurance products, including:

[0028] Activate the insurance knowledge element set, which includes product attributes, coverage, exclusions, and claims conditions; perform multi-dimensional knowledge extraction on the multiple insurance clause knowledge sets based on the insurance knowledge element set to obtain the extraction results of each product element; configure attention based on the insurance knowledge element set to construct a multi-level element attention channel; enhance the extraction results of each product element based on the multi-level element attention channel to obtain the feature matrix of each product element; and generate the profile of each insurance product based on the feature matrix of each product element.

[0029] Preferably, before extracting knowledge from the insurance policy knowledge base, the insurance knowledge element set is activated. This insurance knowledge element set is a pre-constructed dataset that includes keywords such as product attributes, coverage, exclusions, and claim conditions of insurance products. Among them, product attributes refer to the basic information of insurance products, such as the type of insurance (health insurance, life insurance, accident insurance, etc.), product name, insurance company, insurance premium, and insurance period; coverage refers to the various protections provided by the insurance company to customers, such as critical illness protection, hospitalization allowance protection, and accidental injury protection; exclusions refer to the conditions that exempt the insurance company from liability for compensation, including some situations not covered by the coverage, such as certain diseases, natural disasters, and the insured's concealment of medical history; claim conditions refer to the conditions and requirements that users must meet when applying for a claim, including the claim process, required documents, and claim time limit. Subsequently, based on the key names corresponding to each keyword in the insurance knowledge element set, multi-dimensional knowledge extraction is performed from multiple insurance clause knowledge sets in the insurance clause knowledge base. Specifically, by comparing key names, product attributes, coverage responsibilities, exclusions, and claim conditions of each insurance product are extracted from multiple insurance clause knowledge sets. These extracted data are then aggregated to obtain the extraction results for each product element. Furthermore, a multi-level element attention channel is constructed based on the insurance knowledge element set. Specifically, initial weights are assigned to each insurance knowledge element based on expert experience (e.g., "coverage responsibility" weight 0.4, "exclusions" weight 0.3), and the weights are adjusted based on the frequency of user access to each insurance knowledge element. That is, the access frequency (normalized) is multiplied by the element weight, and the adjusted element weight is used as the attention for the corresponding insurance knowledge element, constructing a coarse-grained attention channel. Next, similar attention allocation is performed on specific items of each insurance knowledge element to construct a fine-grained attention channel. By merging the coarse-grained and fine-grained attention channels, a multi-level element attention channel is constructed. Then, the data in the product element extraction results are numerically processed. Specifically, one-hot encoding (e.g., "Insurance Type" → [1,0,0] represents critical illness insurance) is used to quantify categorical data, normalization is used to quantize numerical data to [0,1], and a pre-trained model (e.g., BERT) is used to quantize textual data into embedding vectors (e.g., "Exclusion Scope: Drunk Driving" → vector dimension 512). The quantized results are then input into the multi-level element attention channel and multiplied by the corresponding attention weights (usually the product of the weights in the coarse-grained and fine-grained levels) to complete the enhancement processing of the product element extraction results.Finally, the results of the enhancement processing are synchronized to an empty matrix to construct feature matrices for each product element. These feature matrices are then tagged and stored in JSON format to construct profiles for each insurance product. These insurance product profiles will serve as the core data for matching, helping the system provide users with personalized and accurate insurance recommendations.

[0030] Based on the second user profile, a demand coverage risk assessment is performed on each insurance product profile to obtain a demand coverage risk matrix.

[0031] In one embodiment, after obtaining the user's second profile and the profiles of each insurance product, a demand coverage risk assessment is performed based on the matching between the user's second profile and the insurance product profiles. Specifically, the user's specific protection needs are first extracted from the second profile, and these needs are compared with the protection liabilities in each insurance product profile to identify protection needs not covered by any insurance product. By assessing the coverage risk of these missing protection needs against the target user's total protection needs, a demand coverage risk coefficient for each insurance product can be obtained. These coefficients are then organized to form a demand coverage risk matrix, which displays the degree of matching between different insurance products in meeting user needs. Using this matrix, the system can assess which products have a high risk of missing protection, thereby excluding these products in the subsequent recommendation process. This ensures that the insurance products chosen by the user can meet their protection needs to the greatest extent possible, reducing the risk of insufficient protection.

[0032] Furthermore, this application provides a method for assessing the demand coverage risk of each insurance product profile based on the second user profile, thereby obtaining a demand coverage risk matrix, including:

[0033] Based on the second user profile, an insurance requirement itemset is extracted; based on the profiles of each insurance product, a first insurance product profile is extracted, and a first product insurance liability itemset is constructed based on the first insurance product profile; the first product insurance liability itemset is mapped and compared based on the insurance requirement itemset to determine a first insurance missing itemset; the insurance requirement itemset is evaluated for coverage risk based on the first insurance missing itemset to obtain a first demand coverage risk coefficient, and the first demand coverage risk coefficient is added to the demand coverage risk matrix.

[0034] Preferably, the user's specific protection needs are first extracted from the second user profile, and these extracted protection needs are stored as a protection need itemset. This protection need itemset includes protection type needs (such as health insurance, life insurance, accident insurance, etc.), protection amount needs (the amount of protection for each type of protection that the user expects), protection content needs (such as hospitalization allowance, critical illness protection, specific disease protection, etc.), and supplementary protection needs (such as family liability insurance, pension protection, etc.). Subsequently, an insurance product profile is randomly extracted from each insurance product profile as the first insurance product profile, and all protection liabilities of that insurance product are extracted from the first insurance product profile to form the first product protection liability itemset. This first product protection liability itemset includes protection liability type (such as disease protection, hospitalization protection, accidental injury protection, etc.), protection amount and scope (such as the sum insured, scope of protection, and specific conditions for each type of protection), and protection timeliness (the validity period, deductible period, waiting period, etc. of the protection). Next, the set of coverage needs is compared with the first set of product coverage items to determine whether each need is covered by the insurance product. For example, if a user's need includes "critical illness coverage" and the insurance product also provides this coverage, then that need is considered covered. If a need (such as "high hospitalization allowance") is not included in the insurance product, it is considered a missing item. After the comparison, a first set of missing coverage items is obtained, which represents the inadequacy of the insurance product in covering the user's needs. Then, the identified first set of missing coverage items is used to assess the coverage risk of the user's coverage needs. Specifically, the number of items in the first set of missing coverage items is divided by the number of items in the coverage needs set to calculate the first coverage risk coefficient. This coefficient measures the degree of matching between an insurance product and the user's needs. A higher coefficient indicates a more severe deficiency in the insurance product's coverage of the user's needs, and a greater risk. For example, if a user has 10 coverage needs, and the insurance product is missing 3, the coverage risk coefficient is 0.3, meaning there is a 30% risk in meeting the user's needs. Finally, the calculated first coverage risk coefficient is added to the coverage risk matrix. This matrix records the coverage risk of each insurance product in terms of protecting the user's needs. Each row represents an insurance product, and each column represents the coverage of a single set of coverage needs. This coverage risk matrix ensures that insurance products can meet the user's coverage needs to the greatest extent possible, helping users choose the most suitable product while reducing the risk of insufficient coverage.

[0035] Based on the second user profile, the insurance product profiles are evaluated for protection conflict risk to obtain a protection conflict risk matrix.

[0036] In one embodiment, conflict identification is performed on the profiles of each insurance product based on the user's second profile. This assesses whether conflicts exist between different insurance products. The identification results are then input into a pre-built protection conflict risk detection tree to calculate the protection conflict risk coefficient for each insurance product. These coefficients reflect the conflict risk in the coverage of the insurance products. By mapping these coefficients to an empty matrix, a protection conflict risk matrix is ​​constructed for subsequent matching and optimization, ensuring that users are provided with more accurate and optimized insurance products.

[0037] Furthermore, this application provides a method for assessing the risk of protection conflicts for each insurance product profile based on the second user profile, thereby obtaining a protection conflict risk matrix, including:

[0038] Based on the second user profile, multi-layer conflict identification is performed on the first insurance product profile to obtain a first protection conflict vector; a protection conflict vector sample set and a protection conflict risk sample set are obtained, and the protection conflict risk sample set is clustered based on the protection conflict vector sample set to obtain each conflict risk sample region corresponding to each protection conflict vector sample; support optimization fusion is performed based on each conflict risk sample region to obtain each conflict risk feature value; using each protection conflict vector sample as the detection input node and the each conflict risk feature value as the detection output node, a protection conflict risk detection tree is constructed; the first protection conflict vector is input into the protection conflict risk detection tree to obtain a first protection conflict risk coefficient, and the first protection conflict risk coefficient is added to the protection conflict risk matrix.

[0039] Preferably, when assessing the risk of coverage conflicts, the user's second profile is first used to perform multi-layered conflict identification on the extracted first insurance product profile to ensure that the user's needs do not unnecessarily conflict with the product's coverage responsibilities. Specifically, it checks whether there is overlap in coverage responsibilities between the target user's current insurance products and the first insurance product. For example, if the user has already purchased a certain health insurance, and the recommended insurance product also provides the same critical illness coverage, this overlap will not increase the effectiveness of the coverage but may instead waste premiums. By comparing the user's needs with the product's coverage responsibilities, overlaps in coverage responsibilities are identified and marked as conflicts. It also checks whether the coverage recipients of the first insurance product (such as the types of diseases and accidents covered) match the target user. For example, some insurance products may have specific restrictions on the coverage recipients (such as not being elderly or not having a hereditary disease). When the user's health condition... If the needs do not meet these conditions, conflicts in the coverage are identified and marked as conflict items. The system also checks whether the exclusions of the first insurance product conflict with the target user's coverage needs. For example, some insurance products may not cover specific diseases (such as certain chronic diseases or congenital diseases), while the target user needs this type of coverage. In this case, the system identifies the conflict between the exclusions and the user's needs and marks it as a conflict item. Furthermore, the system checks whether the insurance product's claim conditions match the target user's actual situation. For example, if the user's claim history does not meet certain claim conditions (such as missing required documents or overly complex claim requirements), this conflict is identified and marked as a conflict item. Through the above multi-layered conflict identification, a first protection conflict vector for the first insurance product is generated. This first protection conflict vector contains all conflict items and is used to represent the conflict situation of the product across multiple conflict dimensions. Subsequently, a protection conflict vector sample set and a protection conflict risk sample set are obtained from historical protection logs. Each sample in the protection conflict vector sample set represents the conflict situation of an insurance product, and each sample in the protection conflict risk sample set represents the risk magnitude of the corresponding insurance product conflict. Next, clustering algorithms (such as K-means, DBSCAN, etc.) are used to group insurance products with similar conflict characteristics. Taking K-means as an example, K conflict vector samples are first randomly selected as initial cluster centers. Then, based on the distance between each conflict vector sample and the cluster center, the conflict vector samples are assigned to the nearest cluster center. After assignment, the mean of all conflict vector samples in each cluster is calculated, and this mean is used as the new cluster center. This process is repeated until the cluster centers no longer change significantly or the maximum number of iterations is reached. After clustering, the conflict risk sample set is divided into corresponding conflict risk sample regions based on the clustering results of the conflict vector samples.After obtaining each conflict risk sample region, the system performs support optimization fusion to optimize the conflict risk feature values ​​in each region. During this process, the number of samples in each conflict risk sample region is divided by the total number of samples to obtain the support score for each region. These support scores are compared with a preset support threshold, and conflict risk sample regions with scores below this threshold are removed to ensure that only valid conflict sample regions are used when constructing the detection tree. After removal, for each remaining conflict risk sample region, the mean conflict risk of all samples within that region is calculated as its conflict risk feature value. Then, a decision tree is used, with the conflict vector samples as input nodes and the conflict risk feature values ​​as output nodes. The system recursively splits each conflict vector sample based on its features to minimize errors (e.g., mean squared error). At each split, the contribution of each feature is evaluated until a stopping condition (e.g., maximum tree depth or number of leaf node samples) is met. After training, the final conflict risk detection tree is constructed. Finally, the first protection conflict vector is input into the protection conflict risk detection tree for evaluation to obtain the first protection conflict risk coefficient of the first insurance product. When there is a unique index of the first insurance product in the protection conflict risk matrix, the calculated first protection conflict risk coefficient will be directly written to that position by assignment. When there is no unique index of the first insurance product in the protection conflict risk matrix, such as in the first calculation or dynamic expansion scenario, a new row or column will be added at the end of the protection conflict risk matrix to store the first protection conflict risk coefficient, which will help the system to perform subsequent insurance product screening and optimization.

[0040] An underwriting risk analysis channel is introduced, and the underwriting risk of each insurance product profile is evaluated in conjunction with the second user profile to obtain an underwriting risk matrix.

[0041] In one embodiment, underwriting limitations for each insurance product are extracted from its profile, such as health requirements, age restrictions, and coverage limits. These limitations are then matched with a second user profile to map the user's sensitive characteristics under these underwriting conditions, such as health status, age, and financial capacity, generating a first underwriting sensitivity map. This first map is then input into an underwriting risk analysis channel for risk assessment, yielding multiple underwriting risk coefficients representing the potential risks each insurance product may face during the underwriting process. Subsequently, based on the prediction accuracy of the underwriting risk analysis channel, these coefficients are adjusted and merged to obtain the underwriting risk coefficient for each insurance product, which is then added to the underwriting risk matrix. In this way, the system can comprehensively determine which insurance products are more suitable for a user based on their profile and the underwriting risks of the insurance products, avoiding potential risks caused by mismatched underwriting conditions.

[0042] Furthermore, this application provides an underwriting risk analysis channel, which combines the second user profile with the underwriting risk assessment of each insurance product profile to obtain an underwriting risk matrix, including:

[0043] Based on the first insurance product profile, underwriting limitation features are extracted to obtain a first underwriting limitation sequence; based on the first underwriting limitation sequence, sensitive feature mapping is performed on the second user profile to obtain a first underwriting sensitivity mapping map; the first underwriting sensitivity mapping map is input into the underwriting risk analysis channel, and multiple underwriting risk coefficients are output based on multiple underwriting risk assessment models within the underwriting risk analysis channel; multiple risk assessment accuracies of the multiple underwriting risk assessment models are obtained, and output incentive configuration is performed on the multiple underwriting risk assessment models based on the multiple risk assessment accuracies to obtain multiple output incentive coefficients; output fusion calculation is performed on the multiple underwriting risk coefficients based on the multiple output incentive coefficients to obtain the first product underwriting risk coefficient, and the first product underwriting risk coefficient is added to the underwriting risk matrix.

[0044] Preferably, when assessing the underwriting risk of each insurance product, the underwriting limitation features are first extracted from the first insurance product profile to form a first underwriting limitation sequence. This first underwriting limitation sequence includes various restrictions that the insurance product may set during the underwriting process, such as age restrictions, health requirements, occupational restrictions, and past claims records. Subsequently, the first underwriting limitation sequence is used to perform sensitive feature mapping on the second user profile. That is, features that do not meet the restriction requirements in the first underwriting limitation sequence are selected from the second user profile, and a first underwriting sensitivity mapping map is generated based on the selected features. This first underwriting sensitivity mapping map reflects the limitations that the user may face during the insurance product underwriting process. Afterwards, the obtained first underwriting sensitivity mapping map is input into a pre-constructed underwriting risk analysis channel. This underwriting risk analysis channel consists of multiple underwriting risk assessment models connected in parallel, specifically used to analyze and identify the risks that the insurance product may encounter during the underwriting process. These underwriting risk assessment models are built based on different algorithms, such as decision trees, regression models, random forests, and neural networks, and the specific training process is similar to that described above. By analyzing the first underwriting sensitivity map using these evaluation models, multiple underwriting risk coefficients can be obtained. Each coefficient reflects the risk assessment result of the product under a specific evaluation model. Then, the accuracy of each underwriting risk assessment model during validation after training is obtained. Based on these accuracies, an output incentive coefficient is assigned to each model, typically the ratio of each accuracy to the total accuracy, to determine the contribution of each model to the final underwriting risk assessment result. Finally, the underwriting risk coefficient output by each underwriting risk assessment model is multiplied by its corresponding output incentive coefficient, and all products are summed to obtain the final first-product underwriting risk coefficient. This first-product underwriting risk coefficient is added to the underwriting risk matrix, providing crucial risk assessment data for subsequent insurance product recommendations.

[0045] Based on the aforementioned demand coverage risk matrix, the aforementioned protection conflict risk matrix, and the aforementioned underwriting risk matrix, a multi-level optimization matching of the multiple insurance products is performed to obtain an insurance optimization matching map.

[0046] In one embodiment, after obtaining the demand coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix, risk constraint optimization is performed on these three matrices to determine the initial insurance optimization space. Within this initial insurance optimization space, multi-level optimization matching is conducted, including insurance matching optimization, demand coverage optimization, protection conflict optimization, and underwriting risk optimization. By summarizing the optimization results of these four levels, an insurance optimization matching graph is constructed to demonstrate the performance of various insurance products under different risk dimensions, helping users make the optimal choice.

[0047] Furthermore, this application provides a multi-level optimization matching method for the multiple insurance products based on the stated demand coverage risk matrix, the stated protection conflict risk matrix, and the stated underwriting risk matrix, to obtain an insurance optimization matching graph, including:

[0048] Based on the demand coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix, risk constraint optimization is performed on the multiple insurance products to obtain an initial insurance optimization space. Weights are allocated based on multi-dimensional matching risk factors to generate an insurance matching risk analysis model, whereby the multi-dimensional matching risk factors include demand coverage risk, protection conflict risk, and underwriting risk. Based on a predetermined quantity, the insurance matching risk ranking optimization is performed on the initial insurance optimization space according to the insurance matching risk analysis model to obtain a first domain for insurance product optimization. Based on the predetermined quantity, demand coverage risk ranking optimization is performed on the initial insurance optimization space to obtain a second domain for insurance product optimization. Based on the predetermined quantity, protection conflict risk ranking optimization is performed on the initial insurance optimization space to obtain a third domain for insurance product optimization. Based on the predetermined quantity, underwriting risk ranking optimization is performed on the initial insurance optimization space to obtain a fourth domain for insurance product optimization. Based on the first, second, third, and fourth domains of insurance product optimization, an insurance optimization matching graph is generated.

[0049] Preferably, after obtaining the demand coverage risk matrix, protection conflict risk matrix, and underwriting risk matrix, multiple insurance products are mapped based on these matrices. Risk constraints are then used to optimize and filter the mapping results, forming a multi-dimensional initial space for insurance optimization. This initial space demonstrates the performance of each insurance product across multiple risk dimensions. Subsequently, based on expert decisions and historical experience, weights are assigned to each dimension of the multi-dimensional matching risk factors. These factors include the demand coverage risk dimension, protection conflict risk dimension, and underwriting risk dimension. By encapsulating these weights, an insurance matching risk analysis model is constructed. Then, all insurance products are ranked and optimized based on a predetermined number (Q). That is, the demand coverage risk coefficient, protection conflict risk coefficient, and underwriting risk coefficient of each insurance product in the initial optimization space are input into the insurance matching risk analysis model. Combined with the encapsulated weights, these three risk coefficients are merged into an insurance matching risk coefficient through weighted summation. All insurance matching risk coefficients are then ranked from smallest to largest, and the top Q insurance products are selected as the screening results, constituting the first domain for insurance product optimization. Furthermore, the demand coverage risk coefficients of each insurance product in the initial insurance optimization space are sorted according to a predetermined quantity, constructing a second domain for insurance product optimization. Similarly, the protection conflict risk coefficients and product underwriting risk coefficients are sorted, constructing a third and fourth domain for insurance product optimization. Then, the first, second, third, and fourth domains are combined to generate an insurance optimization matching graph. Each point in this graph represents an insurance product, and the size, color, and position of the point reflect the product's strengths and weaknesses across the four risk dimensions. High-quality insurance products typically occupy a superior position in the graph, indicating excellent performance in meeting needs, avoiding conflicts, and mitigating underwriting risks. Moreover, when a new insurance product is added, its demand coverage characteristics, protection conflict characteristics, and underwriting risk characteristics can be quickly extracted based on its terms knowledge set, and the corresponding risk coefficients are calculated separately. The obtained risk coefficients are dynamically added to the demand coverage risk matrix, protection conflict risk matrix, and underwriting risk matrix according to a matrix update method, thereby achieving adaptive matrix expansion. After updating the matrix, only local optimization calculations need to be performed on the dimensions corresponding to the newly added products to complete the incremental update of the initial insurance optimization space, without reordering all products. This effectively improves the response efficiency to new insurance products and ensures the real-time and scalable nature of the optimization matching process. The resulting insurance optimization matching graph will ultimately recommend insurance products with the lowest overall risk and best meeting the user's needs, ensuring that the user selects the insurance product that best suits their requirements, minimizing coverage conflicts and underwriting risks, and improving user satisfaction and the effectiveness of coverage.

[0050] Furthermore, this application provides a method for performing risk constraint optimization on the multiple insurance products based on the stated demand coverage risk matrix, the stated protection conflict risk matrix, and the stated underwriting risk matrix to obtain an initial space for insurance optimization, including:

[0051] The multiple insurance products are mapped according to the demand coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix to obtain the risk matrix of each product; constraints are configured according to the multidimensional matching risk factors to obtain the multidimensional risk constraint matrix; based on the risk matrix of each product, the multiple insurance products are optimized and screened according to the multidimensional constraint matrix to generate the initial space for insurance optimization.

[0052] Optionally, after obtaining the demand coverage risk matrix, protection conflict risk matrix, and underwriting risk matrix, the demand coverage risk coefficient, protection conflict risk coefficient, and underwriting risk coefficient of each insurance product in these matrices are mapped to the corresponding insurance products, constructing risk matrices for each product. These risk matrices comprehensively reflect the product's performance across the three dimensions of demand coverage, protection conflict, and underwriting risk. Subsequently, based on the dimensions of the multidimensional matching risk factors, and according to business needs and historical experience, corresponding constraints are set for each dimension, and these constraints are summarized into a multidimensional risk constraint matrix. Next, the multidimensional risk constraint matrix is ​​applied to the product risk matrices of multiple insurance products to perform preliminary optimization screening, eliminating those insurance products that significantly deviate from the requirements of the multidimensional risk constraint matrix. Finally, the remaining insurance products and their corresponding product risk matrices are stored, constructing an initial insurance optimization space. This initial space contains all screened insurance products and serves as the basis for further screening and optimization.

[0053] In summary, the embodiments of this application have at least the following technical effects:

[0054] This application embodiment first constructs a first user profile based on the target user's user dataset, and optimizes the protection needs based on the first user profile to obtain a second user profile. Then, it extracts and organizes multi-dimensional knowledge from an insurance clause knowledge base to construct profiles for each insurance product. The insurance clause knowledge base includes multiple insurance clause knowledge sets for multiple insurance products. Next, it evaluates the demand coverage risk of each insurance product profile based on the second user profile to obtain a demand coverage risk matrix. Further, it evaluates the protection conflict risk of each insurance product profile based on the second user profile to obtain a protection conflict risk matrix. Then, it introduces an underwriting risk analysis channel and combines the second user profile with the underwriting risk evaluation of each insurance product profile to obtain an underwriting risk matrix. Finally, it performs multi-level optimization matching of the multiple insurance products based on the demand coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix to obtain an insurance optimization matching graph. These technological effects collectively address the technical problems of existing insurance recommendation methods, such as their inability to accurately match user needs, their neglect of conflicts between clauses, and the resulting significant deviations between recommendation results and users' actual risk preferences and protection needs. By performing multi-level optimization matching on a multi-dimensional risk matrix, these methods achieve dynamic and accurate matching between insurance products and user needs, reduce the risk of protection conflicts, improve the relevance of recommendation results, and meet the technical effects of personalized insurance configuration needs.

[0055] Example 2, based on the same inventive concept as the intelligent insurance matching method combining user profiles and a terms knowledge base in the aforementioned examples, such as... Figure 2 As shown, this application provides an intelligent insurance matching system that combines user profiles and a policy knowledge base. The system includes: a user profile construction module 11: constructing a first user profile based on the user dataset of the target user, and optimizing the protection needs based on the first user profile to obtain a second user profile; a knowledge organization module 12: extracting and organizing multi-dimensional knowledge based on the insurance policy knowledge base to construct profiles for each insurance product, wherein the insurance policy knowledge base includes multiple insurance policy knowledge sets for multiple insurance products; a coverage risk assessment module 13: performing a demand coverage risk assessment on each insurance product profile based on the second user profile to obtain a demand coverage risk matrix; a conflict risk assessment module 14: performing a protection conflict risk assessment on each insurance product profile based on the second user profile to obtain a protection conflict risk matrix; an underwriting risk assessment module 15: introducing an underwriting risk analysis channel, and performing an underwriting risk assessment on each insurance product profile based on the second user profile to obtain an underwriting risk matrix; and a multi-level optimization matching module 16: performing multi-level optimization matching on the multiple insurance products based on the demand coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix to obtain an insurance optimization matching graph.

[0056] Furthermore, the user profile building module 11 is also used to perform the following methods:

[0057] The user dataset is cleaned to obtain a user feature set, which includes a set of basic user information and a set of historical insurance behaviors. Insurance-related sensitivity is evaluated based on the user feature set to obtain a feature sensitivity evaluation sequence. The user feature set is then fitted with a profile based on the feature sensitivity evaluation sequence to generate the first user profile.

[0058] Furthermore, the user profile building module 11 is also used to perform the following methods:

[0059] Based on the first user profile, multi-dimensional protection needs are predicted to obtain a protection needs profile; the first user profile and the protection needs profile are merged to generate the second user profile.

[0060] Furthermore, the knowledge organization module 12 is also used to perform the following methods:

[0061] Activate the insurance knowledge element set, which includes product attributes, coverage, exclusions, and claims conditions; perform multi-dimensional knowledge extraction on the multiple insurance clause knowledge sets based on the insurance knowledge element set to obtain the extraction results of each product element; configure attention based on the insurance knowledge element set to construct a multi-level element attention channel; enhance the extraction results of each product element based on the multi-level element attention channel to obtain the feature matrix of each product element; and generate the profile of each insurance product based on the feature matrix of each product element.

[0062] Furthermore, the coverage risk assessment module 13 is also used to perform the following method:

[0063] Based on the second user profile, an insurance requirement itemset is extracted; based on the profiles of each insurance product, a first insurance product profile is extracted, and a first product insurance liability itemset is constructed based on the first insurance product profile; the first product insurance liability itemset is mapped and compared based on the insurance requirement itemset to determine a first insurance missing itemset; the insurance requirement itemset is evaluated for coverage risk based on the first insurance missing itemset to obtain a first demand coverage risk coefficient, and the first demand coverage risk coefficient is added to the demand coverage risk matrix.

[0064] Furthermore, the conflict risk assessment module 14 is also used to perform the following methods:

[0065] Based on the second user profile, multi-layer conflict identification is performed on the first insurance product profile to obtain a first protection conflict vector; a protection conflict vector sample set and a protection conflict risk sample set are obtained, and the protection conflict risk sample set is clustered based on the protection conflict vector sample set to obtain each conflict risk sample region corresponding to each protection conflict vector sample; support optimization fusion is performed based on each conflict risk sample region to obtain each conflict risk feature value; using each protection conflict vector sample as the detection input node and the each conflict risk feature value as the detection output node, a protection conflict risk detection tree is constructed; the first protection conflict vector is input into the protection conflict risk detection tree to obtain a first protection conflict risk coefficient, and the first protection conflict risk coefficient is added to the protection conflict risk matrix.

[0066] Furthermore, the underwriting risk assessment module 15 is also used to perform the following methods:

[0067] Based on the first insurance product profile, underwriting limitation features are extracted to obtain a first underwriting limitation sequence; based on the first underwriting limitation sequence, sensitive feature mapping is performed on the second user profile to obtain a first underwriting sensitivity mapping map; the first underwriting sensitivity mapping map is input into the underwriting risk analysis channel, and multiple underwriting risk coefficients are output based on multiple underwriting risk assessment models within the underwriting risk analysis channel; multiple risk assessment accuracies of the multiple underwriting risk assessment models are obtained, and output incentive configuration is performed on the multiple underwriting risk assessment models based on the multiple risk assessment accuracies to obtain multiple output incentive coefficients; output fusion calculation is performed on the multiple underwriting risk coefficients based on the multiple output incentive coefficients to obtain the first product underwriting risk coefficient, and the first product underwriting risk coefficient is added to the underwriting risk matrix.

[0068] Furthermore, the multi-layer optimization matching module 16 is also used to perform the following method:

[0069] Based on the demand coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix, risk constraint optimization is performed on the multiple insurance products to obtain an initial insurance optimization space. Weights are allocated based on multi-dimensional matching risk factors to generate an insurance matching risk analysis model, whereby the multi-dimensional matching risk factors include demand coverage risk, protection conflict risk, and underwriting risk. Based on a predetermined quantity, the insurance matching risk ranking optimization is performed on the initial insurance optimization space according to the insurance matching risk analysis model to obtain a first domain for insurance product optimization. Based on the predetermined quantity, demand coverage risk ranking optimization is performed on the initial insurance optimization space to obtain a second domain for insurance product optimization. Based on the predetermined quantity, protection conflict risk ranking optimization is performed on the initial insurance optimization space to obtain a third domain for insurance product optimization. Based on the predetermined quantity, underwriting risk ranking optimization is performed on the initial insurance optimization space to obtain a fourth domain for insurance product optimization. Based on the first, second, third, and fourth domains of insurance product optimization, an insurance optimization matching graph is generated.

[0070] Furthermore, the multi-layer optimization matching module 16 is also used to perform the following method:

[0071] The multiple insurance products are mapped according to the demand coverage risk matrix, the protection conflict risk matrix, and the underwriting risk matrix to obtain the risk matrix of each product; constraints are configured according to the multidimensional matching risk factors to obtain the multidimensional risk constraint matrix; based on the risk matrix of each product, the multiple insurance products are optimized and screened according to the multidimensional constraint matrix to generate the initial space for insurance optimization.

[0072] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0073] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0074] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. An intelligent insurance matching method combining user profiling and clause knowledge base, characterized in that, The method comprises: According to the user data set of the target user, a user first portrait is constructed, and a guarantee demand optimization is performed based on the user first portrait to obtain a user second portrait; According to the insurance clause knowledge base, multi-dimensional knowledge extraction and sorting are performed to construct each insurance product portrait, and the insurance clause knowledge base comprises a plurality of insurance clause knowledge sets of a plurality of insurance products; According to the user second portrait, a demand coverage risk matrix is obtained by performing demand coverage risk evaluation on each insurance product portrait, comprising: according to the user second portrait, a guarantee demand item set is extracted; according to each insurance product portrait, a first insurance product portrait is extracted, and a first product guarantee liability item set is constructed according to the first insurance product portrait; the first product guarantee liability item set is mapped and compared according to the guarantee demand item set, a first guarantee missing item set is determined, a first demand coverage risk coefficient is obtained by performing coverage risk evaluation on the guarantee demand item set according to the first guarantee missing item set, and the first demand coverage risk coefficient is added to the demand coverage risk matrix; According to the user second portrait, a guarantee conflict risk matrix is obtained by performing guarantee conflict risk evaluation on each insurance product portrait, comprising: according to the user second portrait, a first insurance product portrait is performed multi-layer conflict identification, and a first guarantee conflict vector is obtained; a guarantee conflict vector sample set and a guarantee conflict risk sample set are obtained, and each conflict risk sample area corresponding to each guarantee conflict vector sample is obtained by clustering the guarantee conflict risk sample set according to the guarantee conflict vector sample set; each conflict risk characteristic value is obtained by support optimization fusion according to each conflict risk sample area; the guarantee conflict risk detection tree is constructed by taking each guarantee conflict vector sample as a detection input node and taking each conflict risk characteristic value as a detection output node; the first guarantee conflict risk coefficient is obtained by inputting the first guarantee conflict vector into the guarantee conflict risk detection tree, and the first guarantee conflict risk coefficient is added to the guarantee conflict risk matrix; An underwriting risk analysis channel is introduced, and an underwriting risk matrix is obtained by performing underwriting risk evaluation on each insurance product portrait in combination with the user second portrait, comprising: according to the first insurance product portrait, a first underwriting limited sequence is obtained by extracting the underwriting limited features; a first underwriting sensitive mapping diagram is obtained by mapping the sensitive features of the user second portrait according to the first underwriting limited sequence; the first underwriting sensitive mapping diagram is input into the underwriting risk analysis channel, a plurality of underwriting risk coefficients are output according to a plurality of underwriting risk evaluation models in the underwriting risk analysis channel; a plurality of risk evaluation accuracies of the plurality of underwriting risk evaluation models are obtained, and the plurality of underwriting risk evaluation models are output and excited according to the plurality of risk evaluation accuracies to obtain a plurality of output excitation coefficients; a first product underwriting risk coefficient is obtained by output fusion calculation of the plurality of underwriting risk coefficients according to the plurality of output excitation coefficients, and the first product underwriting risk coefficient is added to the underwriting risk matrix; According to the demand coverage risk matrix, the guarantee conflict risk matrix and the underwriting risk matrix, multi-level optimization matching is performed on the plurality of insurance products to obtain an insurance optimization matching graph. 2.The intelligent insurance matching method of combining user portrait with clause knowledge base according to claim 1, wherein, According to the multi-dimensional knowledge extraction and analysis of the insurance clause knowledge base, an image of each insurance product is constructed, including: An insurance knowledge element set is activated, including product attributes, guarantee responsibilities, exemption ranges and claim conditions; According to the insurance knowledge element set, multi-dimensional knowledge extraction is performed on the plurality of insurance clause knowledge sets to obtain product element extraction results; According to the insurance knowledge element set, a multi-level element attention channel is constructed; According to the multi-level element attention channel, the product element extraction results are strengthened to obtain product element feature matrices; According to the product element feature matrices, the images of the insurance products are generated. 3.The intelligent insurance matching method of combining user portrait with clause knowledge base according to claim 1, wherein, According to the demand coverage risk matrix, the guarantee conflict risk matrix and the underwriting risk matrix, multi-level optimization matching is performed on the plurality of insurance products to obtain an insurance optimization matching graph, including: According to the demand coverage risk matrix, the guarantee conflict risk matrix and the underwriting risk matrix, risk constraint optimization is performed on the plurality of insurance products to obtain an insurance optimization initial space; According to the multi-dimensional matching risk factors, a weight distribution is performed to generate an insurance matching risk analysis model, including demand coverage risk, guarantee conflict risk and underwriting risk; Based on a predetermined number, according to the insurance matching risk analysis model, an insurance matching risk sorting optimization is performed on the insurance optimization initial space to obtain an insurance product optimization first domain; Based on the predetermined number, according to the insurance optimization initial space, a demand coverage risk sorting optimization is performed to obtain an insurance product optimization second domain; Based on the predetermined number, according to the insurance optimization initial space, a guarantee conflict risk sorting optimization is performed to obtain an insurance product optimization third domain; Based on the predetermined number, according to the insurance optimization initial space, an underwriting risk sorting optimization is performed to obtain an insurance product optimization fourth domain; According to the insurance product optimization first domain, the insurance product optimization second domain, the insurance product optimization third domain and the insurance product optimization fourth domain, the insurance optimization matching graph is generated. 4.The intelligent insurance matching method of combining user portrait with clause knowledge base according to claim 3, characterized in that, According to the demand coverage risk matrix, the guarantee conflict risk matrix and the underwriting risk matrix, risk constraint optimization is performed on the plurality of insurance products to obtain an insurance optimization initial space, including: According to the demand coverage risk matrix, the guarantee conflict risk matrix and the underwriting risk matrix, mapping is performed on the plurality of insurance products to obtain product risk matrices; According to the multi-dimensional matching risk factors, a weight distribution is performed to generate an insurance matching risk analysis model, including demand coverage risk, guarantee conflict risk and underwriting risk; Based on the product risk matrices, according to the multi-dimensional risk constraint matrix, the plurality of insurance products are optimized and screened to generate the insurance optimization initial space. 5.The intelligent insurance matching method of combining user portrait with clause knowledge base according to claim 1, wherein, According to the user data set of the target user, a user first image is constructed, including: According to the user data set, feature cleaning is performed to obtain a user feature set, and the user data set includes a user basic information set and a historical insurance behavior set; According to the user feature set, an insurance association sensitivity evaluation is performed to obtain a feature sensitivity evaluation sequence; Based on the feature sensitivity evaluation sequence, the user feature set is fitted to generate a user first portrait. 6.The intelligent insurance matching method of combining user portrait with clause knowledge base according to claim 1, wherein, Based on the user first portrait, a guarantee demand optimization is performed to obtain a user second portrait, including: According to the user first portrait, a multi-dimensional guarantee demand prediction is performed to obtain a guarantee demand portrait; Fusion of the user first portrait and the guarantee demand portrait generates the user second portrait.

7. An intelligent insurance matching system that combines user profiling with a clause knowledge base, characterized in that, The system is used to execute the intelligent insurance matching method combining user portrait and clause knowledge base according to any one of claims 1-6, and the system includes: A user portrait construction module: according to the user data set of the target user, a user first portrait is constructed, and based on the user first portrait, a guarantee demand optimization is performed to obtain a user second portrait; A knowledge carding module: according to the insurance clause knowledge base, multi-dimensional knowledge extraction carding is performed to construct each insurance product portrait, and the insurance clause knowledge base includes multiple insurance product knowledge sets; An overlay risk evaluation module: according to the user second portrait, a demand overlay risk evaluation is performed on the each insurance product portrait to obtain a demand overlay risk matrix; A conflict risk evaluation module: according to the user second portrait, a guarantee conflict risk evaluation is performed on the each insurance product portrait to obtain a guarantee conflict risk matrix; A risk evaluation module: an underwriting risk analysis channel is introduced, and combined with the user second portrait, an underwriting risk evaluation is performed on the each insurance product portrait to obtain an underwriting risk matrix; A multi-layer optimization matching module: according to the demand overlay risk matrix, the guarantee conflict risk matrix and the underwriting risk matrix, a multi-layer optimization matching is performed on the multiple insurance products to obtain an insurance optimization matching graph.

Images