An insurance product intelligent matching and dynamic optimization recommendation method and system

Abstract

The application relates to an insurance product intelligent matching and dynamic optimization recommendation method and system, which comprises the following steps: constructing a user panoramic portrait by integrating user static attributes, existing insurance policies and dynamic life cycle data; calculating the current guarantee value of each risk category based on the existing insurance policies in the portrait, calling a predefined risk model to generate a guarantee gap in combination with dynamic data; simultaneously, analyzing insurance product clauses by using natural language processing to form structured feature data and build a product knowledge graph; according to the guarantee gap, the user static attributes and the knowledge graph, calculating the matching score of candidate products by a multi-objective matching algorithm to generate a ranked recommendation list; collecting user interaction behavior data, dynamically adjusting the matching algorithm weight and feeding back optimization. The application realizes personalized intelligent matching and dynamic optimization of insurance products, and improves the recommendation accuracy and user satisfaction through data driving and real-time feedback.

Description

Technical Field

[0001] This application relates to the field of intelligent recommendation technology for insurance products, and in particular to a method and system for intelligent matching and dynamic optimization recommendation of insurance products. Background Technology

[0002] With the deep integration of fintech and the insurance industry, policy management systems have become an important tool for users to manage their insurance assets. These systems initially aimed to address the pain points of paper policies being difficult to store and retrieve, by centrally storing and managing user policy information through digital means, providing basic functions such as policy inquiry and renewal reminders. Building on this foundation, to further enhance service value, the industry has gradually explored introducing product recommendation functions into policy management systems, hoping to proactively provide users with suitable insurance product suggestions, thereby transforming from a passive management tool to a proactive risk advisor. These early recommendation attempts often drew on relatively mature recommendation technologies from the e-commerce field, such as collaborative filtering based on basic user demographic attributes (e.g., age, gender) or content matching based on simple product tags.

[0003] However, insurance products, as complex financial contracts, offer long-term protection against specific risks, fundamentally differing from fast-moving consumer goods or ordinary commodities. Directly applying e-commerce recommendation logic to insurance scenarios remains challenging. For example, a user's true protection needs are not determined solely by static basic information but are deeply intertwined with their family structure, financial situation, and existing insurance portfolio. Static profiles struggle to capture dynamic changes in these needs. Furthermore, insurance product terms are technically complex, and keywords or simple tags alone cannot accurately represent core elements such as coverage and exclusions, resulting in limited product matching accuracy. Moreover, existing recommendation methods are typically one-off, unidirectional pushes, lacking a closed-loop mechanism for continuous optimization based on user feedback and changes in their circumstances. This makes it difficult for recommendation effectiveness to improve over time and adapt to the dynamic needs throughout the user's lifecycle. Summary of the Invention

[0004] To address the aforementioned technical issues, this application provides a method and system for intelligent matching and dynamic optimization recommendation of insurance products.

[0005] Firstly, this application provides a method for intelligent matching and dynamic optimization recommendation of insurance products, employing the following technical solution: A method for intelligent matching and dynamic optimization recommendation of insurance products, the method comprising: Acquire user static attribute data, existing policy data, and dynamic lifecycle data to construct a comprehensive user profile; Based on the existing policy data in the user panoramic profile, calculate the current coverage value for each risk category; Invoke a predefined risk model and generate a protection gap value based on the dynamic life cycle data and the current protection value; Obtain the original terms and conditions data of insurance products, parse them using natural language processing technology to generate structured product feature data, and construct a product knowledge graph containing product relationship mappings; Based on the aforementioned gap value, user static attribute data, and product knowledge graph, a multi-objective matching algorithm is used to generate candidate product matching scores. Based on the candidate product matching scores, a sorted list of recommended products is generated; Collect user interaction data on the recommended product list; Based on the interaction behavior data, the weight parameters of the multi-target matching algorithm are dynamically adjusted, and the adjusted weight parameters are fed back to the execution steps of the multi-target matching algorithm.

[0006] By adopting the above technical solution, insurance product recommendations are upgraded from a crude push based on simple rules or static tags to a highly accurate, personalized, and adaptive intelligent decision support system. This solution not only effectively addresses the core pain points of traditional recommendation methods—such as disconnected coverage, limited dimensions, and insufficient precision—but also, through the introduction of a closed-loop learning mechanism, enables the system to continuously evolve. Ultimately, while improving the scientific nature and satisfaction of user protection planning, it also enables insurance service providers to achieve cost-effective and efficient precision marketing, completing the value leap of the policy management system from a passive management tool to a proactive intelligent advisor.

[0007] Optionally, the step of calculating the current coverage value for each risk category based on existing policy data in the user profile includes: Parse the existing policy data to extract the policy type and corresponding sum insured information; Based on predefined risk classification rules, the insurance policy type is mapped to a preset risk category; Aggregate all valid sums of coverage under the same risk category to generate the current coverage value.

[0008] By adopting the above technical solution, the scattered, heterogeneous, and static original policy data of users is transformed into centralized, standardized, dynamic, and risk-divided quantitative protection indicators. This solution fundamentally overcomes the shortcomings of traditional methods that rely on manual inventory and are difficult to comprehensively and accurately summarize users' protection levels, providing a reliable and computable data foundation for subsequent scientific risk gap analysis and high-precision intelligent insurance product recommendations.

[0009] Optionally, the step of calling a predefined risk model and generating a coverage gap value based on the dynamic lifecycle data and the current coverage value includes: Obtain family structure data and financial data from dynamic lifecycle data; Based on the aforementioned family structure data, a quantitative value for family responsibility is calculated; The financial leverage ratio is determined based on the debt and revenue data in the aforementioned financial data. Call the predefined risk model, input the current protection value, the family responsibility quantification value, and the financial leverage coefficient, and generate the protection gap value for each risk category.

[0010] By adopting the above technical solutions, dynamic life cycle events are obtained, family structural relationships are transformed into calculable economic responsibilities, and a financial leverage coefficient is innovatively introduced to reflect the moderating effect of individual financial status on risk tolerance. Finally, through the integrated calculation of predefined risk models, a precise quantitative protection gap value is output, laying a solid and scientific decision-making foundation for achieving accurate intelligent matching of insurance products.

[0011] Optionally, the steps of obtaining the original terms and conditions data of insurance products, parsing them using natural language processing technology to generate structured product feature data, and constructing a product knowledge graph containing product relationship mappings include: Analyze the original terms and conditions data of the insurance product and extract the key terms and conditions text; Natural language processing technology is used to identify the entities responsible for protection, the entities exempt from liability, and the entities providing health information in the text of the key terms and conditions, thereby generating structured product feature data. Calculate the liability similarity between products based on the liability entities in the structured product feature data; Based on the aforementioned responsibility similarity, a product substitution relationship mapping and a product complement relationship mapping are generated; By aggregating the structured product feature data, product substitution relationship mapping, and product complementarity relationship mapping, a product knowledge graph is constructed.

[0012] By employing the aforementioned technical solutions, natural language processing technology is used to deeply analyze the semantics of insurance terms, transforming product features into computable data. Based on quantified liability similarity, the substitution and complementarity relationships between products are intelligently deduced, ultimately integrating them into a product knowledge graph rich in semantic associations. This process changes the traditional product management approach that relies on human experience or simple labels, enabling machines to understand and reason about the product's connotations. This lays a crucial knowledge foundation for achieving accurate, intelligent, and explainable insurance product recommendations.

[0013] Optionally, the step of generating candidate product matching scores using a multi-objective matching algorithm based on the protection gap value, user static attribute data, and product knowledge graph includes: Based on the aforementioned coverage gap value, calculate the gap coverage score of the candidate product for the target risk category; Based on the user's static attribute data, calculate the fit score between the candidate product and the user's attributes; Based on the product substitution relationship mapping or product complementarity relationship mapping in the product knowledge graph, the complementarity score between the candidate product and the existing policy combination is calculated. Based on the product price characteristics in the product knowledge graph, calculate the cost-effectiveness score of candidate products; Based on the pre-configured weight parameters, the gap coverage score, fit score, complementarity score, and candidate product cost-effectiveness score are weighted and aggregated to generate a candidate product matching score.

[0014] By employing the aforementioned technical solution, scores across four dimensions—coverage gap, suitability, complementarity, and cost-effectiveness—are calculated step-by-step and then weighted and aggregated, successfully simulating the complex decision-making process of a professional insurance advisor. This solution not only ensures that recommended products accurately address users' protection gaps but also comprehensively considers the feasibility of purchase, synergy with existing protection packages, and economic rationality. This multi-dimensional and quantifiable evaluation mechanism transforms the final recommendation list from a simple ranking of single indicators into a comprehensive and intelligent solution seeking optimal solutions under multiple constraints, significantly improving the accuracy, scientific rigor, and user satisfaction of insurance product recommendations.

[0015] Optionally, the step of dynamically adjusting the weight parameters of the multi-target matching algorithm based on the interaction behavior data and feeding back the adjusted weight parameters to the execution steps of the multi-target matching algorithm includes: The interactive behavior data is categorized into predefined behavior types; Based on the mapping relationship between behavior type and weight parameters, weight adjustment parameters are generated; Based on the weight adjustment parameters, update the set of weight parameters in the multi-target matching algorithm; The updated set of weight parameters is input into the execution steps of the multi-target matching algorithm.

[0016] By adopting the above technical solution, user interaction behaviors are classified, quantified, and transformed into incremental adjustments to algorithm weights. These adjustments are then applied back to the algorithm itself, constructing an adaptive closed-loop system capable of continuously learning from real user feedback. This transforms the originally static, rule-dependent multi-objective matching algorithm into an intelligent agent that can perceive changes in market preferences and dynamically adjust its decision-making priorities. This mechanism ensures that the timeliness and accuracy of the recommendation strategy continuously improve over time and with data accumulation, ultimately achieving continuous optimization of recommendation effectiveness and user satisfaction.

[0017] Optionally, after the step of constructing the user panoramic profile, the method further includes: Identify the relationships among users' family members and generate a family member relationship map; Based on the family member relationship mapping, the user's existing policy data will be associated with the corresponding family members; Call the predefined risk model to calculate the protection gap value of each family member under the preset risk category; Aggregate the protection gap values ​​of all family members to generate a family-level protection gap distribution map.

[0018] By employing the aforementioned technical solutions, not only is the risk exposure of each family member accurately quantified, but the interrelationships and transmission mechanisms of risks among family members are also clearly revealed and quantified, generating an intuitive family-level protection gap distribution map. This transforms insurance recommendations from simple responses to individual needs into collaborative protection plans that optimize overall family financial resilience and provide risk hedging effects, based on a systematic understanding of the family's balance sheet. This fundamentally solves the systemic risk management challenges facing families that traditional methods cannot address.

[0019] Secondly, this application provides an intelligent matching and dynamic optimization recommendation system for insurance products, which adopts the following technical solution: An intelligent matching and dynamic optimization recommendation system for insurance products, the system comprising: The user panoramic profile building module is used to acquire user static attribute data, existing policy data and dynamic life cycle data to build a user panoramic profile. The current coverage value calculation module is used to calculate the current coverage value for each risk category based on the existing policy data in the user panoramic profile. The gap generation module is used to call a predefined risk model and generate a gap value based on the dynamic life cycle data and the current coverage value. The product knowledge graph construction module is used to obtain the original terms and conditions data of insurance products, parse and generate structured product feature data through natural language processing technology, and construct a product knowledge graph containing product relationship mapping. The matching score generation module is used to generate candidate product matching scores based on the protection gap value, user static attribute data, and product knowledge graph through a multi-target matching algorithm. The recommendation list generation module is used to generate a sorted list of recommended products based on the matching scores of the candidate products. The user behavior collection module is used to collect user interaction behavior data on the recommended product list; The weight parameter dynamic adjustment module is used to dynamically adjust the weight parameters of the multi-target matching algorithm based on the interaction behavior data, and feed the adjusted weight parameters back to the execution steps of the multi-target matching algorithm.

[0020] Thirdly, this application provides a computer device, which adopts the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to perform the steps of the method as described in the first aspect.

[0021] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the first process of an intelligent matching and dynamic optimization recommendation method for insurance products according to one embodiment of this application.

[0023] Figure 2 This is a schematic diagram of the second process of the intelligent matching and dynamic optimization recommendation method for insurance products according to one embodiment of this application.

[0024] Figure 3 This is a schematic diagram of the third process of the intelligent matching and dynamic optimization recommendation method for insurance products according to one embodiment of this application.

[0025] Figure 4 This is a schematic diagram of the fourth process of the intelligent matching and dynamic optimization recommendation method for insurance products according to one embodiment of this application.

[0026] Figure 5 This is a schematic diagram of the fifth process of the intelligent matching and dynamic optimization recommendation method for insurance products according to one embodiment of this application.

[0027] Figure 6 This is a schematic diagram of the sixth process of the intelligent matching and dynamic optimization recommendation method for insurance products according to one embodiment of this application.

[0028] Figure 7 This is a schematic diagram of the seventh process of the intelligent matching and dynamic optimization recommendation method for insurance products according to one embodiment of this application. Detailed Implementation

[0029] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figures 1-7 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0030] This application discloses a method for intelligent matching and dynamic optimization recommendation of insurance products.

[0031] Reference Figure 1 A method for intelligent matching and dynamic optimization recommendation of insurance products, specifically including: Step S101: Obtain user static attribute data, existing policy data, and dynamic lifecycle data to construct a comprehensive user profile; The user profile is not simply a collection of information, but a structured data entity. Static attribute data (such as age, occupation, and place of residence) forms the basic framework of the user's risk profile, determining the basic range and premium levels of insurable products. Existing policy data is transformed into a machine-understandable list of protected assets through OCR recognition and structured parsing, clearly outlining the user's current protection baseline. The introduction of dynamic lifecycle data (such as changes in marital status, birth of children, and new large debts) is the essence of dynamic optimization. It enables the system to perceive the unmet and potential new protection needs of users arising from changes in life stages.

[0032] By integrating these three types of data, the generated user profile is upgraded from a flat set of labels into a three-dimensional model that reflects the user's current status and future risk exposure, providing comprehensive and reliable input for subsequent accurate analysis.

[0033] Step S102: Calculate the current coverage value for each risk category based on the existing policy data in the user panoramic profile; While the coverage of insurance products varies greatly, the core risk categories they address (such as death risk, critical illness risk, and medical expense risk) are limited. This step aggregates the corresponding effective sum insured from all relevant policies of the user for each risk category.

[0034] For example, calculating a user's current death benefit requires summing the effective death benefits of all their life insurance, critical illness insurance with death benefits, and accident insurance products at the current point in time. This process quantifies and consolidates the level of protection, integrating the user's scattered protection capabilities across different insurance companies and products into a clear, risk-categorized protection balance sheet, thus laying the foundation for identifying protection gaps in the next step.

[0035] Step S103: Invoke the predefined risk model and generate the protection gap value based on dynamic life cycle data and the current protection value; The predefined risk model essentially embeds a scientific family risk management logic. The model uses the "current protection value" as a benchmark and then inputs "dynamic life cycle data" (such as the family's necessary living expenses, children's education expenses, and total outstanding mortgage payments in the next 5-10 years). Through the model's built-in algorithms (such as income replacement method and demand analysis method), it calculates the ideal risk protection amount that users should have to maintain family financial stability when different risks occur.

[0036] Understandably, the difference between this ideal sum assured and the current coverage value is called the "coverage gap value". This negative quantified gap changes the recommendation objective from selling products to filling the gap, greatly improving the targeting and scientific nature of the recommendation.

[0037] Step S104: Obtain the original terms and conditions data of the insurance product, parse and generate structured product feature data through natural language processing technology, and construct a product knowledge graph containing product relationship mapping; Insurance terms and conditions are lengthy, technical, and highly unstructured natural language texts, which are difficult to utilize deeply using traditional methods. Natural language processing technology, through named entity recognition, relation extraction, and other techniques, automatically extracts key features such as "coverage liability," "exclusion clauses," "health disclosure requirements," and "waiting period" from the terms and conditions, and transforms them into well-organized, indexable, and computable structured product feature data.

[0038] Furthermore, based on these structured features, a product knowledge graph is constructed: nodes in the graph represent products, while edges represent semantic relationships between products. For example, products with highly similar coverage are considered "substitutes," while products with complementary coverage (such as comprehensive medical insurance and critical illness insurance) are considered "complementary." This operation upgrades the fragmented product database into a semantically related network, providing strong data support for subsequent intelligent and interconnected recommendations.

[0039] Step S105: Based on the guarantee gap value, user static attribute data and product knowledge graph, generate candidate product matching scores through a multi-objective matching algorithm; Specifically, the multi-objective matching algorithm is not a simple screening process, but rather an evaluation of the fit between the product and the user from multiple dimensions, calculating a sub-score for each dimension. These dimensions typically include: gap coverage (the degree of matching between the product's coverage liability and the coverage gap value, which is the core objective), user fit (the degree of conformity between the product's age, occupation, region, and other restrictions and the user's static attributes), product complementarity (based on knowledge graphs, determining whether the new product overlaps with or complements the user's existing policy combination), and cost-effectiveness, etc.

[0040] In this embodiment, the algorithm assigns an initial weight parameter to each target, and sums the weighted sub-scores to obtain a comprehensive matching score. This process comprehensively considers the urgency of the need, the feasibility of the purchase, the rationality of the solution, and the economy. The output is no longer a simple product list, but a quantitatively evaluated and ordered set of candidate solutions.

[0041] Step S106: Generate a sorted list of recommended products based on the candidate product matching scores; The system sorts all candidate products and their matching scores from highest to lowest, and typically selects the Top N products to generate the final recommendation list. This sorting mechanism ensures that the products that best match the user's actual protection gaps, personal circumstances, and have the best synergy with existing protection plans are presented to the user first, greatly improving decision-making efficiency and the usability of recommendation results.

[0042] Meanwhile, to achieve explainable recommendations, the system will automatically generate easy-to-understand reasons for recommendations based on the main contributing dimensions of the matching score (such as extremely high gap coverage score or complementarity score), thereby enhancing users' trust and willingness to purchase.

[0043] Step S107: Collect user interaction data on the recommended product list; Among these factors, user interaction data (such as clicking to view details, adding to favorites, ignoring, and ultimately completing the insurance purchase) represents the most direct and authentic vote for the recommendation results. Recording these behaviors is equivalent to collecting labeled training samples for machine learning models.

[0044] For example, completing the insurance purchase is a strong positive feedback signal, indicating that the match was very successful; ignoring or quickly swiping away may mean negative feedback. Behind this behavioral data lies the user's approval or disapproval of the recommendation strategy (i.e., the weights of each objective), which is a key basis for driving the system to make better decisions in the next recommendation.

[0045] Step S108: Based on the interaction behavior data, dynamically adjust the weight parameters of the multi-target matching algorithm, and feed back the adjusted weight parameters to the execution steps of the multi-target matching algorithm.

[0046] The underlying principle of this step is to construct a feedback-based reinforcement learning or online learning mechanism, enabling the recommendation system to have adaptive and evolutionary capabilities.

[0047] Specifically, the system periodically (e.g., weekly or monthly) analyzes accumulated user behavior data and uses machine learning algorithms such as logistic regression to re-evaluate whether the weight parameters of the multi-objective matching algorithm are set reasonably.

[0048] In some embodiments, for example, if data reveals that users who ultimately purchase insurance tend to favor products with "high complementarity" rather than simply those with "high gap coverage," the system automatically increases the weighting of the "complementarity" objective. This adjusted weighting is immediately fed back and applied to subsequent matching calculations for all users (or specific user groups). This forms a complete closed loop of "recommendation-feedback-learning-optimization-re-recommendation," enabling the system to continuously learn from real-world user behavior, fine-tune its recommendation strategy, and ultimately become increasingly intelligent, achieving personalized and accurate recommendations.

[0049] The above implementation upgrades insurance product recommendations from a crude push based on simple rules or static tags to a highly accurate, personalized, and adaptive intelligent decision support system. This technical solution not only effectively addresses the core pain points of traditional recommendation methods—such as disconnected coverage, limited dimensions, and insufficient precision—but also, by introducing a closed-loop learning mechanism, enables the system to continuously evolve. Ultimately, while improving the scientific nature and satisfaction of user protection planning, it also enables insurance service providers to achieve cost-effective and efficient precision marketing, completing the value leap of the policy management system from a passive management tool to a proactive intelligent advisor.

[0050] Reference Figure 2 As one implementation of step S102, the step of calculating the current coverage value for each risk category based on existing policy data in the user panoramic profile includes: Step S201: Parse existing policy data and extract policy type and corresponding sum insured information; The underlying logic of this step is to transform the user's non-standardized original policy information into standardized data units that can be recognized and calculated by machines. The user's existing policy data usually comes from multiple insurance companies, and its format may include electronic policy PDFs, scanned copies of paper policies, or structured data from system integration.

[0051] In some embodiments, for unstructured or semi-structured documents (such as PDF electronic insurance policies), it is necessary to use OCR (Optical Character Recognition) technology or a specific document parsing engine to accurately locate and identify key text information from complex layouts, especially "type of insurance" (such as "whole life insurance" or "critical illness insurance") and "sum insured amount" (such as "RMB 1,000,000"). This process is not a simple text extraction, but involves named entity recognition in natural language processing to accurately distinguish "sum insured amount" from other monetary fields such as "premium" and "cash value".

[0052] Step S202: Map the policy type to a preset risk category according to the predefined risk classification rules; This involves establishing a standardized mapping system from specific insurance products to abstract risk categories, thereby achieving a unified treatment of coverage responsibilities. While the insurance market offers a wide variety of products with diverse names, their essence is the transfer of specific life or health risks.

[0053] In this embodiment, the predefined risk classification rule is a configurable mapping rule library or lookup table constructed from domain expert knowledge. The core function of this rule is to map the specific "insurance type" extracted in the first step (e.g., "Company X's Worry-Free Lifetime Critical Illness Insurance") to several limited, pre-defined, and standardized "risk categories" (e.g., death risk, critical illness risk, medical risk, etc.). For example, regardless of the product name, as long as its core responsibility is to pay insurance benefits when the insured is diagnosed with a critical illness stipulated in the contract, it will be mapped to the "critical illness risk" category.

[0054] Understandably, this mapping relationship resolves the computational confusion caused by the diversification of product names, categorizing thousands of specific products into a limited number of risk dimensions, making it possible to compare and aggregate the protection capabilities across products. Especially for composite policies containing multiple types of protection liabilities (such as a life insurance policy with additional critical illness and medical coverage), this step also requires the splitting of liabilities, rationally allocating the sum insured of a policy to multiple risk categories according to liability ratios or agreements, ensuring the comprehensiveness and accuracy of the calculations.

[0055] Step S203: Aggregate all valid sums of coverage under the same risk category to generate the current coverage value.

[0056] In this approach, all of a user's protection capabilities are linearly superimposed under the same risk dimension to generate an absolute value that represents their current risk resistance capability.

[0057] Specifically, the aggregation operation performs an arithmetic summation on all valid sum insured values ​​belonging to the same risk category. The term "valid" is a key qualifier, requiring that expired policies be filtered out through logical checks (such as comparing the policy expiry date with the current system time) before aggregation, ensuring that the sum insured included in the calculation truly reflects the user's current coverage status.

[0058] For example, if a user has purchased 500,000 yuan of critical illness insurance from Company A and 300,000 yuan of critical illness insurance from Company B, after this step of aggregation, their "current coverage value" under the "critical illness risk" category will be 800,000 yuan. The significance of this value is that it is no longer isolated policy information, but a comprehensive and quantifiable risk protection indicator.

[0059] In the above implementation method, the scattered, heterogeneous, and static original policy data of users is transformed into centralized, standardized, dynamic, and risk-divided quantitative protection indicators. This solution fundamentally overcomes the shortcomings of traditional methods that rely on manual inventory and are difficult to comprehensively and accurately summarize the protection level of users, providing a reliable and calculable data foundation for subsequent scientific risk gap analysis and high-precision intelligent recommendation of insurance products.

[0060] Reference Figure 3 As one implementation of step S103, the step of calling a predefined risk model and generating a protection gap value based on dynamic lifecycle data and the current protection value includes: Step S301: Obtain family structure data and financial data from the dynamic life cycle data; The core logic of this step lies in capturing significant structural changes in a user's family lifecycle and correlating them with their financial situation, thereby transforming abstract needs into quantifiable economic impacts. Unlike static basic information (such as age and gender), dynamic lifecycle data is essentially the key events driving leaps in security needs.

[0061] Specifically, family structure data (such as newlywed status, birth of children, and the need to support elderly parents) directly alters the scope and duration of a user's economic responsibilities, for example, shifting from personal responsibility when single to responsibility for the entire family's future. Meanwhile, financial data (such as new mortgage or car loan debts, or significant changes in income levels) precisely depicts a user's current and foreseeable future financial burdens and cash flow situation.

[0062] Step S302: Calculate the quantitative value of family responsibility based on family structure data; In this process, qualitative family roles and dependencies are transformed into a comprehensive value that represents the overall economic responsibility of the family through a standardized weighting system.

[0063] Specifically, applying the fundamental principles of "demand analysis" in actuarial science and financial planning, a simple count of family members cannot accurately reflect differences in economic responsibility. For example, the economic burden of supporting a parent with no income is different from that of supporting a parent with a pension; the future expenditure curves for raising an infant and a college student are also drastically different. Therefore, calculating the "quantitative value of family responsibility" is not a simple count, but rather involves consulting a predefined responsibility weight table, assigning different weight coefficients to different roles (such as the spouse who is the breadwinner, minor children, and elderly people without income), and then performing a weighted calculation based on the specific quantities.

[0064] Understandably, this calculation process is essentially building a model to estimate the total amount of funds necessary to maintain the basic living, education, and debt repayment of other family members for a future period (such as children until they become financially independent, and spouses until retirement) when the breadwinner of the family faces risks (such as death or disability). This quantitative value of family responsibility thus becomes a standardized and comparable benchmark for measuring family economic responsibility.

[0065] Step S303: Determine the financial leverage coefficient based on the debt data and income level data in the financial data; The underlying logic of this step is to introduce a risk adjustment factor to characterize the amplification or reduction effect of the user's current financial situation on their protection needs, making the recommended protection amount more risk-sensitive.

[0066] Specifically, the financial leverage ratio is a key financial concept used to measure the vulnerability or resilience of a user's financial structure. It is calculated as the ratio of debt to income (i.e., the debt-to-income ratio). A family with a high debt ratio and relatively tight income has a more leveraged financial structure, meaning the impact of any income interruption risk is amplified, and the family's financial system is less stable. Therefore, a higher level of protection is needed to hedge against risk, and a leverage ratio greater than 1 should be assigned to amplify the demand. Conversely, a family with no debt and stable income has strong risk resistance, and its protection needs are more based on basic responsibilities; its leverage ratio may be close to or equal to 1.

[0067] Understandably, by mapping the debt-to-income ratio to a specific leverage coefficient within a preset range, a quantitative assessment of the user's financial risk status is achieved, and this assessment is dynamically incorporated into the calculation of protection needs, so that the output results match the user's actual financial pressure.

[0068] Step S304: Call the predefined risk model, input the current protection value, the family responsibility quantification value and the financial leverage coefficient, and generate the protection gap value for each risk category.

[0069] The logical principle behind this step is to comprehensively assess a family's economic responsibilities, financial risks, and existing protection levels within a unified, parameterized model framework, and then accurately calculate the protection gap that urgently needs to be filled through arithmetic operations.

[0070] In this embodiment, a predefined risk model encapsulates the professional logic of risk assessment. This model first multiplies the quantified family responsibility value (representing the total responsibility that should ideally be covered) by the financial leverage coefficient (representing the risk multiple that needs to be adjusted due to financial structure). This multiplication can then be combined with a preset per capita responsibility cost base based on the regional cost of living to generate a personalized risk exposure benchmark value for the user. That is, risk exposure benchmark value = quantified family responsibility value × financial leverage coefficient × preset per capita responsibility cost base. This risk exposure benchmark value represents the ideal level of protection required to maintain family financial security when a specific risk occurs.

[0071] The model then performs a subtraction operation: subtracting the calculated current coverage value (i.e., the total effective coverage amount already held by the user) from this risk exposure baseline value, the resulting coverage gap value is a quantitative indicator that can be positive or negative. A positive gap indicates the specific amount of insufficient coverage, zero indicates adequate coverage, and a negative gap may mean over-coverage.

[0072] In the above implementation, dynamic life cycle events are acquired, family structural relationships are transformed into calculable economic responsibilities, and a financial leverage coefficient is innovatively introduced to reflect the moderating effect of individual financial status on risk tolerance. Finally, through the integrated calculation of predefined risk models, a precise quantitative protection gap value is output, laying a solid and scientific decision-making foundation for achieving accurate intelligent matching of insurance products.

[0073] Reference Figure 4 As one implementation of step S104, the steps of obtaining the original terms and conditions data of the insurance product, parsing and generating structured product feature data through natural language processing technology, and constructing a product knowledge graph containing product relationship mappings include: Step S401: Parse the original terms and conditions data of the insurance product and extract the key terms and conditions text. Insurance terms are legal documents with complex content, but the core information that determines the characteristics of the product is usually concentrated in specific chapters, such as "Coverage", "Exclusions", and "Health Disclosure".

[0074] In this embodiment, the parsing process first relies on a pre-defined rule base or model to automatically locate these key chapters. This avoids indiscriminately sending the entire clause text (including the preface, definitions, general provisions, etc.) into subsequent analysis, greatly improving the efficiency and accuracy of data processing. Next, the extraction operation specifically involves segmenting the located chapter text into independent sentence units. Because a chapter may contain multiple parallel or progressive guarantee liabilities or disclaimers, splitting them into independent sentences or short phrases helps the subsequent entity recognition model accurately understand and label each smallest-granular guarantee or restriction condition.

[0075] Step S402: Using natural language processing technology, identify the entities responsible for protection, the entities exempt from liability, and the entities providing health information in the key clause elements text, and generate structured product feature data. Specifically, the guarantees, liability protections, and exemption conditions are not presented in a standard format, but are hidden in the natural language description.

[0076] In the embodiments of this application, natural language processing technology, particularly named entity recognition models based on pre-trained models (such as BERT), is used. These models can understand the professional semantics of the insurance field, thereby accurately identifying entities covered by insurance in sentences. These entities include disease names (such as "malignant tumor" and "coronary artery bypass surgery"), medical behaviors (such as "inpatient treatment" and "outpatient surgery"), and conditions for compensation (such as "confirmed for the first time" and "reaching the state of disease"). In addition, entities such as exclusion conditions (such as "driving under the influence of alcohol" and "congenital disease") can also be identified.

[0077] Next, the identified entities are organized according to a predefined data model to form a complete and well-structured product feature profile. This is achieved by establishing a structured data model, for example, creating a data object for each product with attribute fields corresponding to "Insurance Liability List" (containing all identified insurance liability entities), "Disclaimer List," and "Health Disclosure Requirements List," respectively. This process transforms unstructured text information into a well-structured data structure that can be directly queried, compared, and calculated by the program.

[0078] Step S403: Calculate the liability similarity between products based on the liability entities in the structured product feature data; The calculation of product liability similarity includes: converting the set of entities with liability for protection into a word vector matrix, and calculating the distance between the matrices using a cosine similarity algorithm.

[0079] Specifically, by using a vector space model and a similarity measurement algorithm, the semantic similarity of the core warranty responsibilities of products is quantified, providing an objective and calculable basis for defining the relationship between products. Determining whether two products are similar cannot rely on subjective feelings but requires an objective comparison of their warranty responsibilities.

[0080] In this embodiment, word vector technology is first used to transform the "set of entities responsible for insurance coverage" for each product into a numerical vector (i.e., a "word vector matrix") in a high-dimensional space. The geometric features of this vector represent the overall semantic information of the product's insurance coverage. Then, a cosine similarity algorithm is used to calculate the cosine of the angle between the corresponding vectors of two products in the high-dimensional space. This value ranges from -1 to 1; the closer the value is to 1, the more consistent the directions of the two vectors, meaning the more semantically similar the insurance coverage of the two products. This calculation method can capture the deep semantic connections between liability descriptions, rather than simply matching keywords, thus more accurately reflecting the functional substitutability of products.

[0081] Step S404: Based on responsibility similarity, generate product substitution relationship mapping and product complement relationship mapping; Based on quantified similarity values, relationships are determined using preset threshold rules, thereby establishing semantically clear associations between products. The calculated responsibility similarity is a continuous value, which needs to be converted into discrete, business-meaning relationship tags.

[0082] In this embodiment, this conversion is achieved by setting two key thresholds: when the similarity of the liabilities of two products is higher than a high first threshold (e.g., 0.8), they are determined to have a high degree of overlap in actual protection functions, and one product can largely substitute for the other, thus generating a "product substitution relationship mapping". Conversely, when the similarity is lower than a low second threshold (e.g., 0.3), and their protection liability categories have almost no overlap, they are determined to not only not conflict, but can also jointly constitute a more comprehensive protection, thus generating a "product complementarity relationship mapping" (e.g., basic critical illness insurance and specialized cancer insurance). The application of this rule enables the system to automatically and on a large scale discover the implicit substitution and complementarity network in the product library.

[0083] Step S405: Aggregate structured product feature data, product substitution relationship mapping, and product complementarity relationship mapping to construct a product knowledge graph.

[0084] The construction of the product knowledge graph involves using each product as a node, with its complete structured product feature data attached as attributes. Related product nodes are connected by edges representing substitution and complementarity relationships. This graph-structured representation method significantly surpasses traditional relational database tables, enabling the system to easily answer complex business questions such as "What products have similar functions to product A?" (traversing along substitution edges) and "Which products can be paired with product A to form a more comprehensive protection solution?" (traversing along complementarity edges).

[0085] Understandably, this product knowledge graph has become a powerful semantic knowledge base, providing solid data support and reasoning capabilities for subsequent intelligent matching algorithms to make multi-dimensional and interpretable product recommendations.

[0086] In the above implementation, natural language processing technology is used to deeply analyze the semantics of insurance terms, transforming product features into computable data. Based on quantified liability similarity, the substitution and complementarity relationships between products are intelligently deduced, ultimately integrating them into a product knowledge graph rich in semantic associations. This process changes the traditional product management method that relies on human experience or simple labels, achieving machine-understandable and reasonable understanding of product connotations, laying a crucial knowledge foundation for accurate, intelligent, and explainable insurance product recommendations.

[0087] Reference Figure 5 As one implementation of step S105, the step of generating candidate product matching scores using a multi-objective matching algorithm based on the protection gap value, user static attribute data, and product knowledge graph includes: Step S501: Based on the coverage gap value, calculate the gap coverage score of the candidate product for the target risk category; The calculation of the gap coverage score is essentially an assessment of the extent to which a candidate product can fill the quantitative guarantee gap precisely calculated by the preceding steps. This calculation is not a simple binary judgment, but a refined quantitative process.

[0088] Specifically, the system performs a deep match between the coverage characteristics of candidate products and the specific risk categories (such as critical illness risk) corresponding to the coverage gap value. This not only requires the product to be in the correct broad category, but also to determine whether its coverage scope covers the key sub-risks under that risk category. Furthermore, the system evaluates the relationship between the product's sum assured and the gap value: when the sum assured is greater than or equal to the gap value, it is considered to provide full coverage and is assigned the highest score; when the sum assured is less than the gap value, it is assigned a linearly decreasing score proportionally. This mechanism ensures that the recommendation system prioritizes products that effectively address the user's insufficient coverage in terms of both scope and amount, thus shifting the recommendation logic from "what the user might like" to "what the user truly needs," which is one of the core innovative points that distinguishes this application from existing technologies.

[0089] Step S502: Calculate the fit score between candidate products and user attributes based on user static attribute data; One key feature is the introduction of a feasibility filter. This ensures that recommended products not only match user needs but are also practically suitable for purchase, thereby improving the feasibility and conversion rate of recommendations. Even if a product perfectly covers a coverage gap, the recommendation is invalid if the user does not meet the product's underwriting rules due to age, occupation, or location. The suitability score calculation addresses this issue.

[0090] In this embodiment, the purchase condition characteristics of candidate products (such as the insured age range, the list of insurable occupational categories, and sales area restrictions) can be compared one by one with the user's static attribute data (such as actual age, occupation, and city of residence). For each condition, a complete match results in full marks for that condition, while partial or non-matches result in a deduction of corresponding marks according to preset rules (for example, a larger deduction is given for a hard condition like occupational category mismatch). This score calculation essentially labels each candidate product with "purchaseability," preventing the system from recommending products that the user cannot purchase, thus ensuring the practical value of the recommendation results.

[0091] Step S503: Based on the product substitution relationship mapping or product complementarity relationship mapping in the product knowledge graph, calculate the complementarity score between the candidate product and the existing policy combination; Specifically, the system queries the relationship between candidate products and all the insurance policies currently held by the user in the policy graph. If there is a complementary mapping between the candidate product and existing products (for example, the user already has hospitalization insurance, and the candidate product is a fixed-benefit critical illness insurance), it means that the product can make up for the deficiencies of the existing protection system and form a better risk protection network, thus receiving the highest score. Conversely, if there is a substitution mapping (for example, the candidate product and existing products have a high degree of overlap in coverage), recommending the product would result in duplicate protection and wasted funds, thus receiving the lowest score. This mechanism ensures that the recommendation system can act like a professional insurance advisor, recommending products that can improve the health of the protection portfolio from the perspective of the user's overall financial planning and protection efficiency.

[0092] Step S504: Calculate the cost-effectiveness score of candidate products based on the product price characteristics in the product knowledge graph; By incorporating economic considerations, the system can be guided to prioritize recommending more competitively priced options among products with similar functions, thereby improving user satisfaction and the overall value of the solution.

[0093] Specifically, the cost-effectiveness score is not calculated by simply comparing absolute prices, but rather based on a reasonable market benchmark. The system retrieves price characteristics (such as annual premiums) of candidate products from the product knowledge graph and compares them with the average market price data of similar products.

[0094] In some embodiments, the scoring rules are typically as follows: products priced significantly below the market average (e.g., below 80%) are assigned high scores to reflect their price advantage; products priced significantly above the market average (e.g., above 120%) are assigned low scores; and products in the middle range are assigned gradient scores through linear interpolation. This scoring mechanism enables the system to identify and prioritize options that save users money among multiple candidate products with similar protection effects, achieving a balance between protection effectiveness and economic burden.

[0095] Step S505: Based on the pre-configured weight parameters, the gap coverage score, fit score, complementarity score, and candidate product cost-effectiveness score are weighted and aggregated to generate a candidate product matching score.

[0096] Understandably, the four dimensions of gap coverage, fit, complementarity, and cost-effectiveness are sometimes synergistic and sometimes contradictory (for example, a product with excellent coverage may be expensive). It is difficult to directly compare the four independent scores, so a weighted aggregation is needed for comprehensive decision-making.

[0097] Specifically, a pre-configured weight parameter (such as weights A, B, C, and D) is assigned to each dimension, with the sum of the weights being 1. The weight reflects the relative importance of that dimension in the current recommendation strategy. For example, in scenarios emphasizing filling coverage gaps, the weight for gap coverage can be set higher. Finally, a weighted summation formula is used to calculate the comprehensive matching score for each candidate product. This score is no longer a reflection of a single dimension, but rather an optimal compromise and balance achieved by the four dimensions under specific weight configurations. It provides a quantitative and comparable metric for the overall matching quality of a product relative to a specific user, enabling the system to generate a scientific and reasonable recommendation order.

[0098] In the above implementation, scores for four dimensions—coverage gap, suitability, complementarity, and cost-effectiveness—are calculated step-by-step and then weighted and aggregated, successfully simulating the complex decision-making process of a professional insurance advisor. This technical solution not only ensures that recommended products accurately address users' protection gaps but also comprehensively considers the feasibility of purchase, synergy with existing protection portfolios, and economic rationality. This multi-dimensional and quantifiable evaluation mechanism ensures that the final recommendation list is not merely a simple ranking based on a single indicator, but a comprehensive and intelligent solution seeking optimal solutions under multiple constraints, significantly improving the accuracy, scientific rigor, and user satisfaction of insurance product recommendations.

[0099] Reference Figure 6 As one implementation of step S108, the step of dynamically adjusting the weight parameters of the multi-target matching algorithm based on the interaction behavior data and feeding back the adjusted weight parameters to the execution steps of the multi-target matching algorithm includes: Step S601: Classify the interaction behavior data into predefined behavior types; Among them, user interaction data is the only real source for the system to perceive the recommendation effect, but it is itself fragmented and ambiguous in meaning, such as a "click" or a "ignore".

[0100] In this application embodiment, a classification standard can be established, namely, a predefined behavior type. The most typical division is positive feedback behavior type (such as clicking on details, adding to favorites, completing insurance purchase) and negative feedback behavior type (such as quickly swiping over, closing the recommendation page).

[0101] Understandably, this classification process is not simply data grouping, but a crucial step in giving semantic meaning to raw behavioral data. It interprets users' unconscious clicks as feedback signals with clear direction, such as "approval" or "disinterest" in the recommendation results.

[0102] Step S602: Generate weight adjustment parameters based on the mapping relationship between behavior type and weight parameters; The categorized behavior types are still symbolic and cannot be directly used for mathematical operations. This step achieves this crucial transformation through a pre-defined mapping relationship.

[0103] Specifically, the essence of this mapping relationship is to assign a specific, signed, preset adjustment coefficient to each type of behavior. For example, each positive feedback (such as successful insurance purchase) may correspond to a positive adjustment coefficient (such as +0.01), while each negative feedback (such as ignoring) corresponds to a negative adjustment coefficient (such as -0.005).

[0104] Understandably, the process of generating weight adjustment parameters involves statistically analyzing the number of times various behaviors are triggered within a specific time period (such as the past week) and then summing them up based on their corresponding adjustment coefficients. This calculated weight adjustment parameter is a specific numerical value that comprehensively reflects the overall tendency implied by the recent user group's behavior. Its sign represents the direction of adjustment (strengthening or weakening), and its absolute value represents the intensity of the adjustment.

[0105] Step S603: Update the set of weight parameters in the multi-objective matching algorithm based on the weight adjustment parameters; Among them, a smooth incremental update strategy can be adopted to fine-tune the weight parameters of each dimension in the multi-objective matching algorithm, so that the algorithm's recommendation strategy can adapt to changes in user group preferences while maintaining the stability of the system.

[0106] In this embodiment, the core of the multi-target matching algorithm is the set of weight parameters (i.e., weights for dimensions such as gap coverage, fit, complementarity, and cost-effectiveness), which defines what constitutes a "good" recommendation. The "update" in this step is not simply replacing old values ​​with new ones, but rather a cautious incremental update, such as multiplying the original weight parameters by a scaling factor based on the weight adjustment parameter. That is, if users have recently shown a stronger preference for products with high "cost-effectiveness" (more positive feedback), the adjustment parameter for that dimension's weight will be positive, and its weight will be slightly increased; conversely, it will be slightly decreased.

[0107] This gradual adjustment approach allows the recommendation strategy to evolve with changing circumstances while preventing drastic weight fluctuations caused by a single, accidental action, thus ensuring the consistency and stability of the recommendation system's experience. Furthermore, the update process is typically accompanied by pre-set boundary constraints to prevent excessive expansion or contraction of weights in any dimension, ensuring that the balance of multi-objective decision-making is not disrupted.

[0108] Step S604: Input the updated set of weight parameters into the multi-target matching algorithm execution steps.

[0109] This new set of parameters is fed back and loaded into the running multi-objective matching algorithm. Starting with the next recommendation request, the algorithm will use this updated weight to comprehensively score and rank all candidate products. This means that the system's recommendation strategy has been fine-tuned based on the collective wisdom of historical users.

[0110] For example, if the system detects that users are recently paying more attention to the "complementarity" between products and existing insurance policies, then the weight of the complementarity dimension will be increased. Consequently, products that effectively fill existing protection gaps will naturally rank higher in subsequent recommendations. This forms a complete, self-driven optimization cycle: user behavior influences weights, changes in weights guide recommendation strategies, and new recommendation strategies generate new user behaviors. This cycle repeats continuously, enabling the system to adapt to changes in user preferences and market demands, becoming increasingly intelligent with use.

[0111] In the above implementation, user interaction behaviors are categorized, quantified, and transformed into incremental adjustments to algorithm weights. These adjustments are then applied back to the algorithm itself, constructing an adaptive closed-loop system capable of continuously learning from real user feedback. This transforms the originally static, rule-dependent multi-objective matching algorithm into an intelligent agent that can sense changes in market preferences and dynamically adjust its decision-making priorities. This mechanism ensures that the timeliness and accuracy of the recommendation strategy continuously improve over time and with data accumulation, ultimately achieving continuous optimization of recommendation effectiveness and user satisfaction.

[0112] Reference Figure 7As a further implementation of the intelligent matching and dynamic optimization recommendation method for insurance products, after the step of constructing a comprehensive user profile, it also includes: Step S701: Identify the relationships among the user's family members and generate a family member relationship mapping; Traditional user profiling, centered on the individual, fails to depict the economic dependence and responsibility relationships among family members. This step identifies core family member entities (such as the user, spouse, children, and dependent parents) by analyzing information provided by the user, including household registration, social security, or self-reported information. Moving beyond simple list records, it formally expresses these relationships using a node-edge model from graph theory.

[0113] Specifically, each family member is abstracted as an independent "node," while the economic responsibility relationships among members, such as raising, supporting, and jointly providing for each other, are abstracted as "directed edges" with direction and weight. For example, an edge from the "family economic pillar" node to the "minor children" node represents a strong, one-way economic responsibility for support. The resulting "family member relationship mapping" is a structured relationship graph that allows the system to clearly describe the flow of economic responsibility and the intensity of dependence within the family, thereby elevating the analytical perspective from isolated individuals to an interconnected family system.

[0114] Step S702: Based on the family member relationship mapping, associate the user's existing policy data with the corresponding family members; Based on the legal relationship of the insurance contract, the coverage provided by each policy is precisely attributed to the family member node that is actually insured, thus overlaying the coverage currently available at each node onto the constructed relationship graph. A family policy may involve multiple insured persons or beneficiaries, and traditional methods often simply classify them under the policyholder's name, leading to confusion regarding coverage attribution.

[0115] In this embodiment, the system parses the policy terms, identifies key legal entities such as the "insured," "survival benefit beneficiary," and "death benefit beneficiary," and precisely associates these entities with the corresponding nodes in the family member relationship mapping generated in the first step. For example, a critical illness insurance policy with the husband as the insured should be associated with the husband's node; an education savings insurance policy with the child as the insured should be associated with the child's node; and a life insurance policy with the wife as the death beneficiary, although the insured is the husband, the beneficiary of its coverage (death benefit) is the wife, reflecting the economic responsibility protection for the wife, thus requiring an association. For policies with shared coverage (such as family shared medical insurance), the system will allocate the sum insured to relevant member nodes proportionally according to the terms and through a liability-splitting algorithm.

[0116] Step S703: Call the predefined risk model to calculate the protection gap value of each family member under the preset risk category; The underlying logic of this step is to quantify the difference between the ideal protection needs and the current protection level at the individual member level, based on their family role and specific circumstances, thereby identifying the specific risk exposure faced by each family member.

[0117] Specifically, the predefined risk model goes deeper than the individual analysis model, taking into account the specific roles of members within the family. The model calculates the "protection gap" for each member in key risk categories such as death, critical illness, medical care, education, and retirement. The calculation is based not only on the member's personal circumstances (such as age and health status), but more importantly, on the economic responsibilities inherent in their family role.

[0118] For example, calculating the "death risk protection gap" for the family's breadwinner requires that their ideal coverage amount should cover their family responsibilities, such as remaining mortgage payments, children's education expenses until independence, and their spouse's living expenses for the next few years, not just their personal debts. For children, the "education risk protection gap" needs to estimate future higher education costs. The model aggregates the "current protection value" from the existing related policies of this member node, and then subtracts this value from the calculated, personalized "ideal risk protection amount" to obtain the gap. This value quantifies the insufficient protection this member has as a component of the family system.

[0119] Step S704: Aggregate the protection gap values ​​of all family members to generate a family-level protection gap distribution map.

[0120] This step involves weighted aggregation of each member's individual risk exposure based on the relationship mapping established in the first step, and reveals the transmission effect of individual risk in the family network, thereby generating a global and systematic risk view.

[0121] Specifically, simply adding up the gap values ​​of each member is insufficient to reflect the full picture of family risk, because the risks among members are interconnected. The "aggregation" in this application's embodiments is an organic aggregation based on relational weights, which considers the direction and strength of edges in the family member relationship mapping (i.e., "failure impact factors"). For example, the death risk gap of the primary income earner has a much higher weight in impacting the overall financial stability of the family than other members. Therefore, the total family risk is not an arithmetic sum, but a weighted sum.

[0122] The system then generates a family-level protection gap distribution map, typically a heatmap-style visualization. The axes of the map may be set for different risk categories and different family members, with the heatmap color depth of each cell representing the severity of the gap for that member within that risk category. This visualization method can intuitively expose the vulnerabilities of the family protection system: for example, it may clearly show that the father's death risk gap (red high-temperature zone) is currently the biggest source of risk for the family, and because the father is the breadwinner, this risk will transmit and amplify the mother's future retirement risk (shown as the associated orange area). This map makes hidden risk transmission paths visible, providing a scientific basis for developing a comprehensive family protection plan that prioritizes addressing key risks and considers risk hedging.

[0123] The above implementation not only precisely quantifies the risk exposure of each family member, but also clearly reveals and quantifies the interrelationships and transmission mechanisms of risks among family members, generating an intuitive family-level protection gap distribution map. This makes insurance recommendations no longer a simple response to single-point needs, but rather, based on a systematic understanding of the family's balance sheet, propose synergistic protection solutions that optimize the family's overall financial resilience and have a risk hedging effect, fundamentally solving the family systemic risk management challenges that traditional methods cannot address.

[0124] This application also discloses an intelligent matching and dynamic optimization recommendation system for insurance products.

[0125] An intelligent matching and dynamic optimization recommendation system for insurance products, specifically including: The user panoramic profile building module is used to acquire user static attribute data, existing policy data and dynamic life cycle data to build a user panoramic profile. The current coverage value calculation module is used to calculate the current coverage value for each risk category based on the existing policy data in the user's panoramic profile. The gap generation module is used to call a predefined risk model and generate a gap value based on dynamic lifecycle data and the current coverage value. The product knowledge graph construction module is used to obtain the original terms and conditions data of insurance products, parse and generate structured product feature data through natural language processing technology, and construct a product knowledge graph containing product relationship mapping. The matching score generation module is used to generate candidate product matching scores based on the guarantee gap value, user static attribute data and product knowledge graph through a multi-target matching algorithm. The recommendation list generation module is used to generate a sorted list of recommended products based on the matching scores of candidate products. The user behavior collection module is used to collect user interaction data on the recommended product list; The dynamic weight parameter adjustment module is used to dynamically adjust the weight parameters of the multi-target matching algorithm based on the interaction behavior data, and feed the adjusted weight parameters back to the execution steps of the multi-target matching algorithm.

[0126] The insurance product intelligent matching and dynamic optimization recommendation system of this application embodiment can implement any of the above methods, and the specific working process of each module in the system can refer to the corresponding process in the above method embodiment.

[0127] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0128] This application also discloses a computer device.

[0129] A computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement an intelligent matching and dynamic optimization recommendation method for insurance products as described above.

[0130] This application also discloses a computer-readable storage medium.

[0131] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the methods of intelligent matching and dynamic optimization recommendation of insurance products.

[0132] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0133] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for intelligent matching and dynamic optimization recommendation of insurance products, characterized in that, The method includes: Acquire user static attribute data, existing policy data, and dynamic lifecycle data to construct a comprehensive user profile; Based on the existing policy data in the user panoramic profile, calculate the current coverage value for each risk category; Invoke a predefined risk model and generate a protection gap value based on the dynamic life cycle data and the current protection value; Obtain the original terms and conditions data of insurance products, parse them using natural language processing technology to generate structured product feature data, and construct a product knowledge graph containing product relationship mappings; Based on the aforementioned gap value, user static attribute data, and product knowledge graph, a multi-objective matching algorithm is used to generate candidate product matching scores. Based on the candidate product matching scores, a sorted list of recommended products is generated; Collect user interaction data on the recommended product list; Based on the interaction behavior data, the weight parameters of the multi-target matching algorithm are dynamically adjusted, and the adjusted weight parameters are fed back to the execution steps of the multi-target matching algorithm.

2. The method for intelligent matching and dynamic optimization recommendation of insurance products according to claim 1, characterized in that, The steps for calculating the current coverage value for each risk category based on the existing policy data in the user profile include: Parse the existing policy data to extract the policy type and corresponding sum insured information; Based on predefined risk classification rules, the insurance policy type is mapped to a preset risk category; Aggregate all valid sums of coverage under the same risk category to generate the current coverage value.

3. The method for intelligent matching and dynamic optimization recommendation of insurance products according to claim 1, characterized in that, The steps of generating a coverage gap value by invoking a predefined risk model, based on the dynamic lifecycle data and the current coverage value, include: Obtain family structure data and financial data from dynamic lifecycle data; Based on the aforementioned family structure data, a quantitative value for family responsibility is calculated; The financial leverage ratio is determined based on the debt and revenue data in the aforementioned financial data. Call the predefined risk model, input the current protection value, the family responsibility quantification value, and the financial leverage coefficient, and generate the protection gap value for each risk category.

4. The method for intelligent matching and dynamic optimization recommendation of insurance products according to claim 3, characterized in that, The steps involved in obtaining the original terms and conditions of insurance products, parsing them using natural language processing technology to generate structured product feature data, and constructing a product knowledge graph containing product relationship mappings include: Analyze the original terms and conditions data of the insurance product and extract the key terms and conditions text; Natural language processing technology is used to identify the entities responsible for protection, the entities exempt from liability, and the entities providing health information in the text of the key terms and conditions, thereby generating structured product feature data. Calculate the liability similarity between products based on the liability entities in the structured product feature data; Based on the aforementioned responsibility similarity, a product substitution relationship mapping and a product complement relationship mapping are generated; By aggregating the structured product feature data, product substitution relationship mapping, and product complementarity relationship mapping, a product knowledge graph is constructed.

5. The method for intelligent matching and dynamic optimization recommendation of insurance products according to claim 4, characterized in that, The steps for generating candidate product matching scores using a multi-objective matching algorithm based on the aforementioned security gap value, user static attribute data, and product knowledge graph include: Based on the aforementioned coverage gap value, calculate the gap coverage score of the candidate product for the target risk category; Based on the user's static attribute data, calculate the fit score between the candidate product and the user's attributes; Based on the product substitution relationship mapping or product complementarity relationship mapping in the product knowledge graph, the complementarity score between the candidate product and the existing policy combination is calculated. Based on the product price characteristics in the product knowledge graph, calculate the cost-effectiveness score of candidate products; Based on the pre-configured weight parameters, the gap coverage score, fit score, complementarity score, and candidate product cost-effectiveness score are weighted and aggregated to generate a candidate product matching score.

6. The method for intelligent matching and dynamic optimization recommendation of insurance products according to claim 1, characterized in that, The steps of dynamically adjusting the weight parameters of the multi-target matching algorithm based on the interaction behavior data and feeding back the adjusted weight parameters to the execution steps of the multi-target matching algorithm include: The interactive behavior data is categorized into predefined behavior types; Based on the mapping relationship between behavior type and weight parameters, weight adjustment parameters are generated; Based on the weight adjustment parameters, update the set of weight parameters in the multi-target matching algorithm; The updated set of weight parameters is input into the execution steps of the multi-target matching algorithm.

7. The method for intelligent matching and dynamic optimization recommendation of insurance products according to any one of claims 1 to 6, characterized in that, Following the step of constructing the user panoramic profile, the following is also included: Identify the relationships among users' family members and generate a family member relationship map; Based on the family member relationship mapping, the user's existing policy data will be associated with the corresponding family members; Call the predefined risk model to calculate the protection gap value of each family member under the preset risk category; Aggregate the protection gap values ​​of all family members to generate a family-level protection gap distribution map.

8. An intelligent matching and dynamic optimization recommendation system for insurance products, characterized in that, The system includes: The user panoramic profile building module is used to acquire user static attribute data, existing policy data and dynamic life cycle data to build a user panoramic profile. The current coverage value calculation module is used to calculate the current coverage value for each risk category based on the existing policy data in the user panoramic profile. The gap generation module is used to call a predefined risk model and generate a gap value based on the dynamic life cycle data and the current coverage value. The product knowledge graph construction module is used to obtain the original terms and conditions data of insurance products, parse and generate structured product feature data through natural language processing technology, and construct a product knowledge graph containing product relationship mapping. The matching score generation module is used to generate candidate product matching scores based on the protection gap value, user static attribute data, and product knowledge graph through a multi-target matching algorithm. The recommendation list generation module is used to generate a sorted list of recommended products based on the matching scores of the candidate products. The user behavior collection module is used to collect user interaction behavior data on the recommended product list; The weight parameter dynamic adjustment module is used to dynamically adjust the weight parameters of the multi-target matching algorithm based on the interaction behavior data, and feed the adjusted weight parameters back to the execution steps of the multi-target matching algorithm.

9. A computer device, characterized in that: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images