An insurance intelligent recommendation method and system based on user portrait

Abstract

The present invention discloses an intelligent insurance recommendation method and system based on user portraits, and the method includes: user data collection, construction of user portraits, construction of user risk assessment models, personalized product recommendations and feedback adjustments. The present invention belongs to the field of data processing technology, and specifically refers to an intelligent insurance recommendation method and system based on user portraits. This solution adopts the construction of user portraits, analyzes risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, uses embedding matrices, calculates new hidden state vectors and assigns user data weights, evaluates consumption capacity, summarizes insurance demand tendencies, and improves recommendation accuracy; adopts the construction of user risk assessment models, determines risk assessment indicators and sets weights for each indicator based on its importance, performs empirical risk minimization and structural risk minimization, improves the generalization ability of the model, and transforms the support vector machine problem into an optimization problem, so that the model runs more efficiently.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and specifically relates to an insurance intelligent recommendation method and system based on user portraits. Background Art

[0002] The intelligent insurance recommendation method based on user portrait is a method that uses various characteristics and data information of users to conduct comprehensive analysis and modeling of multiple aspects of information such as age, occupation, health status, financial status, risk preference, etc., to provide targeted insurance recommendations to users. It aims to accurately match and recommend insurance products that better meet users' needs and risk conditions, better meet users' personalized needs for insurance, and improve the efficiency and accuracy of insurance sales. However, the existing intelligent insurance recommendation methods based on user portraits have technical problems such as the inability to accurately assess users' consumption capacity and insurance demand tendencies, the recommended insurance is not suitable for the specific circumstances of different users, and the recommendation error and misjudgment risk are large; it is difficult to systematically consider the impact of different aspects of users on risks, and the model generalization ability is weak, resulting in technical problems such as mismatch between insurance products and user risk conditions. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent insurance recommendation method and system based on user portraits. In view of the technical problems that it is impossible to accurately evaluate the user's consumption capacity and insurance demand tendency, the recommended insurance is not suitable for the specific circumstances of different users, and the recommendation error and misjudgment risk are relatively large, the present invention adopts the method of constructing user portraits, analyzing risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, using embedding matrices, calculating new hidden state vectors and assigning user data weights, evaluating consumption capacity, summarizing insurance demand tendencies, and improving recommendation accuracy; in view of the technical problems that it is difficult to systematically consider the impact of different aspects of users on risks, the generalization ability of the model is weak, and the insurance products do not match the user's risk status, a user risk assessment model is constructed, risk assessment indicators are determined and weights are set for each indicator according to their importance, empirical risk minimization and structural risk minimization are performed, the generalization ability of the model is improved, the support vector machine problem is converted into an optimization problem, and the model runs more efficiently.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an insurance intelligent recommendation method based on user portrait, which includes the following steps:

[0005] Step S1: user data collection, specifically collecting various information of the user, including basic information, financial information, health information and previous insurance status;

[0006] Step S2: Build a user profile, specifically to clarify goals and indicators, analyze risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, use embedding matrices, calculate new hidden state vectors and assign user data weights, evaluate consumption capacity, and summarize insurance demand tendencies;

[0007] Step S3: Build a user risk assessment model, specifically determine the risk assessment indicators and set weights for each indicator according to its importance, perform empirical risk minimization and structural risk minimization, improve the generalization ability of the model, transform the support vector machine problem into an optimization problem, and make the model run more efficiently;

[0008] Step S4: Personalized product recommendations, specifically, detailed analysis and quantitative evaluation of the customer's risk factors, and targeted selection of insurance products that can effectively cover the corresponding risks and meet the customer's financial tolerance;

[0009] Step S5: Feedback adjustment, specifically collecting user feedback and satisfaction information on recommended products and performing model adjustments.

[0010] Furthermore, in step S1, the user data collection specifically collects various information of the user, including basic information, financial information, health information and previous insurance status.

[0011] Furthermore, in step S2, constructing the user portrait includes the following steps:

[0012] Step S21: clarify the goals and indicators, and determine the specific aspects that need to be portrayed in the user portrait;

[0013] Step S22: Data preprocessing, cleaning the user data, removing errors and duplicate information, normalizing the data, classifying user characteristics, and identifying them with appropriate labels, including age range, income level, and assets;

[0014] Step S23: Analyze risk characteristics, including the following steps:

[0015] Step S231: Use the user data as input to the feature extraction network, using the embedding matrix W glove , embed the user data into the vector space and calculate the embedding result of the user data. The formula used is as follows:

[0016] ;

[0017] In the formula, t represents the index of the position, i t represents the embedding result of user data at position t, W glove is the embedding matrix, m tis the one-hot vector representation at position t, where a one-hot vector is an encoding method used to represent categorical variables, where only one element is 1 and the rest are 0;

[0018] Step S232: Calculate a new hidden state vector, use a long short-term memory network to further encode the embedding vector of each user data, and output a new hidden state vector. The formula used is as follows:

[0019] ;

[0020] Where b t represents a new hidden state vector obtained after LSTM processing. The LSTM is a recurrent neural network suitable for processing sequence data, which is used to better capture long-term dependencies in sequence data and help the model better understand user data;

[0021] Step S233: Use the soft attention mechanism to calculate and assign weights to each user data to better focus on important user data. The formula used is as follows:

[0022] ;

[0023] ;

[0024] Where pt is the non-normalized attention score used to balance the weight of each user's data, αt represents the attention weight after softmax normalization, Mb represents the parameter matrix used to learn the attention weight in the model, hb represents the additional bias term used for attention calculation, and T represents the total number of positions;

[0025] Step S24: Assess consumption capacity by combining income, assets and consumption habits to determine consumption capacity;

[0026] Step S25: Summarize the insurance demand trends based on past insurance behaviors. As new data is generated, continuously update and optimize the user portrait to more accurately reflect the user's current situation.

[0027] Furthermore, in step S3, constructing the user risk assessment model includes the following steps:

[0028] Step S31: Determine risk assessment indicators and select key factors affecting risk, such as age, income, credit record, consumption habits, industry, and region, based on user portraits;

[0029] Step S32: setting weights for each indicator, assigning different weights according to their importance, and defining risk levels, including low, medium, and high risk levels;

[0030] Step S33: scoring according to the scoring criteria, scoring each indicator according to the user's specific information, calculating the comprehensive total score through the weight and the score of each indicator, determining the risk level, and corresponding to the corresponding risk level according to the total score;

[0031] Step S34: Optimizing the user risk assessment model, including the following steps:

[0032] Step S341: Minimize the empirical risk, calculate the empirical risk, clarify the average loss of the model on the current training sample set, and promote the continuous adjustment and improvement of the model in the direction of reducing the empirical risk. The formula used is as follows:

[0033] ;

[0034] In the formula, f represents the user risk assessment model, represents the empirical risk, which means that the model f is in the training sample set (x i ,y i ), n represents the number of samples, x i represents the input feature of the i-th sample, y i represents the output label of the i-th sample, and L represents the loss function;

[0035] Step S342: Structural risk minimization. The support vector machine adopts the structural risk minimization principle as a learning method to minimize complexity and empirical risk, so that the model still has high output function smoothness and generalization ability when the number of samples is limited. The output function of the support vector machine model is calculated using the following formula:

[0036] ;

[0037] In the formula, f(x) represents the output function of the support vector machine model, is the optimization coefficient in the Lagrangian function, represents the Lagrange multiplier in the support vector machine model, represents the kernel function, i and j both represent the index of the sample, ω represents the hyperplane weight vector, and b represents the offset constant, which is used to adjust the offset between the model's predicted value and the actual label. Represents the feature vector after mapping the input sample x to the high-dimensional feature space;

[0038] Step S343: Calculate the insensitive function loss, which helps the model to be more robust in the face of some small errors, while focusing on large differences beyond the insensitive interval and increasing attention to key errors. The formula used is as follows:

[0039] ;

[0040] In the formula, represents the insensitive function loss, x and y represent the input features of the sample and the corresponding output label, and ε represents the insensitive interval, which is used to determine the threshold of the loss value of 0;

[0041] Step S344: Convert the support vector machine problem into an optimization problem, and use the optimization method to solve the decision boundary constant of the support vector machine model to better achieve the user risk assessment task. The formula used is as follows:

[0042] ;

[0043] Where b1 represents the decision boundary constant of the support vector machine model, and C represents the penalty coefficient, which is used to control the complexity of the user risk assessment model.

[0044] Furthermore, in step S4, the personalized product recommendation includes the following steps:

[0045] Step S41: According to the user risk assessment model, various risk factors of the customer are analyzed and quantitatively assessed to obtain the risk characteristics and potential risk points of the customer;

[0046] Step S42: Based on the results of the risk assessment model, targeted insurance products are selected that can effectively cover the corresponding risks and are in line with the financial affordability of the customer, and in-depth research is conducted on the terms, protection level and cost-effectiveness of the products to ensure that the most appropriate personalized insurance recommendation plan is provided to the customer.

[0047] Furthermore, in step S5, the feedback adjustment is specifically to collect user usage feedback and satisfaction information on the recommended products, analyze and evaluate the product recommendation mechanism and parameters based on the user feedback information, and optimize and adjust the recommendation model accordingly based on the evaluation results to improve the accuracy and satisfaction of subsequent recommendations.

[0048] The present invention provides an insurance intelligent recommendation system based on user portrait, which includes a user data collection module, a user portrait construction module, a user risk assessment model construction module, a personalized product recommendation module and a feedback adjustment module;

[0049] The user data collection module specifically collects various information of the user, including basic information, financial information, health information and previous insurance status;

[0050] The user portrait module is constructed to clarify goals and indicators, analyze risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, use embedding matrices, calculate new hidden state vectors and assign user data weights, evaluate consumption capacity, and summarize insurance demand tendencies;

[0051] The user risk assessment model module is constructed to determine the risk assessment indicators and set weights for each indicator according to its importance, minimize the empirical risk and structural risk, improve the generalization ability of the model, transform the support vector machine problem into an optimization problem, and make the model run more efficiently;

[0052] The personalized product recommendation module specifically conducts detailed analysis and quantitative evaluation of various risk factors of customers, and specifically selects insurance products that can effectively cover the corresponding risks and meet the financial tolerance of customers;

[0053] The feedback adjustment module is specifically used to collect user feedback and satisfaction information on recommended products and perform model adjustment.

[0054] The beneficial results achieved by the present invention using the above scheme are as follows:

[0055] (1) In order to solve the technical problems that users’ spending power and insurance demand tendency cannot be accurately assessed, the recommended insurance is not suitable for the specific circumstances of different users, and the recommendation error and misjudgment risk are relatively high, we build user portraits, analyze risk characteristics through feature extraction networks, long short-term memory networks, and soft attention mechanisms, use embedding matrices, calculate new hidden state vectors, and assign user data weights to assess spending power, summarize insurance demand tendencies, and improve recommendation accuracy;

[0056] (2) In order to address the technical issues that it is difficult to systematically consider the impact of different aspects of users on risks and the model’s weak generalization ability, which leads to the mismatch between insurance products and users’ risk conditions, we build a user risk assessment model, determine risk assessment indicators and set weights for each indicator based on its importance, perform empirical risk minimization and structural risk minimization, improve the model’s generalization ability, transform the support vector machine problem into an optimization problem, and make the model run more efficiently. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flowchart of an insurance intelligent recommendation method based on user portrait provided by the present invention;

[0058] Figure 2 A schematic diagram of an insurance intelligent recommendation system based on user portrait provided by the present invention;

[0059] Figure 3 is a schematic flow chart of step S2;

[0060] Figure 4 It is a schematic diagram of the process of step S3.

[0061] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0063] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0064] Example 1, see Figure 1 The present invention provides an insurance intelligent recommendation method based on user portrait, which includes the following steps:

[0065] Step S1: user data collection, specifically collecting various information of the user, including basic information, financial information, health information and previous insurance status;

[0066] Step S2: Build a user profile, specifically to clarify goals and indicators, analyze risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, use embedding matrices, calculate new hidden state vectors and assign user data weights, evaluate consumption capacity, and summarize insurance demand tendencies;

[0067] Step S3: Build a user risk assessment model, specifically determine the risk assessment indicators and set weights for each indicator according to its importance, perform empirical risk minimization and structural risk minimization, improve the generalization ability of the model, transform the support vector machine problem into an optimization problem, and make the model run more efficiently;

[0068] Step S4: Personalized product recommendations, specifically, detailed analysis and quantitative evaluation of the customer's risk factors, and targeted selection of insurance products that can effectively cover the corresponding risks and meet the customer's financial tolerance;

[0069] Step S5: Feedback adjustment, specifically collecting user feedback and satisfaction information on recommended products and performing model adjustments.

[0070] Example 2, see Figure 1This embodiment is based on the above embodiment. In step S1, the user data collection specifically collects various information of the user, including basic information, financial information, health information and previous insurance status.

[0071] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the user portrait is constructed, including the following steps:

[0072] Step S21: clarify the goals and indicators, and determine the specific aspects that need to be portrayed in the user portrait;

[0073] Step S22: Data preprocessing, cleaning the user data, removing errors and duplicate information, normalizing the data, classifying user characteristics, and identifying them with appropriate labels, including age range, income level, and assets;

[0074] Step S23: Analyze risk characteristics, including the following steps:

[0075] Step S231: Use the user data as input to the feature extraction network, using the embedding matrix W glove , embed the user data into the vector space and calculate the embedding result of the user data. The formula used is as follows:

[0076] ;

[0077] In the formula, t represents the index of the position, i t represents the embedding result of user data at position t, W glove is the embedding matrix, m t is the one-hot vector representation at position t, where a one-hot vector is an encoding method used to represent categorical variables, where only one element is 1 and the rest are 0;

[0078] Step S232: Calculate a new hidden state vector, use a long short-term memory network to further encode the embedding vector of each user data, and output a new hidden state vector. The formula used is as follows:

[0079] ;

[0080] Where b t represents a new hidden state vector obtained after LSTM processing. The LSTM is a recurrent neural network suitable for processing sequence data, which is used to better capture long-term dependencies in sequence data and help the model better understand user data;

[0081] Step S233: Use the soft attention mechanism to calculate and assign weights to each user data to better focus on important user data. The formula used is as follows:

[0082] ;

[0083] ;

[0084] Where pt is the non-normalized attention score used to balance the weight of each user's data, αt represents the attention weight after softmax normalization, Mb represents the parameter matrix used to learn the attention weight in the model, hb represents the additional bias term used for attention calculation, and T represents the total number of positions;

[0085] Step S24: Assess consumption capacity by combining income, assets and consumption habits to determine consumption capacity;

[0086] Step S25: Summarize the insurance demand trends based on past insurance behaviors. As new data is generated, continuously update and optimize the user portrait to more accurately reflect the user's current situation.

[0087] By performing the above operations, user portraits are constructed, risk characteristics are analyzed through feature extraction networks, long short-term memory networks and soft attention mechanisms, and embedding matrices are used to calculate new hidden state vectors and assign user data weights to evaluate consumption capacity, summarize insurance demand tendencies, and improve recommendation accuracy. This solves the technical problems of being unable to accurately evaluate users' consumption capacity and insurance demand tendencies, the recommended insurance not being suitable for the specific circumstances of different users, and the high risk of recommendation errors and misjudgments.

[0088] Example 4, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S3, the construction of the user risk assessment model includes the following steps:

[0089] Step S31: Determine risk assessment indicators and select key factors affecting risk, such as age, income, credit record, consumption habits, industry, and region, based on user portraits;

[0090] Step S32: setting weights for each indicator, assigning different weights according to their importance, and defining risk levels, including low, medium, and high risk levels;

[0091] Step S33: scoring according to the scoring criteria, scoring each indicator according to the user's specific information, calculating the comprehensive total score through the weight and the score of each indicator, determining the risk level, and corresponding to the corresponding risk level according to the total score;

[0092] Step S34: Optimizing the user risk assessment model, including the following steps:

[0093] Step S341: Minimize the empirical risk, calculate the empirical risk, clarify the average loss of the model on the current training sample set, and promote the continuous adjustment and improvement of the model in the direction of reducing the empirical risk. The formula used is as follows:

[0094] ;

[0095] In the formula, f represents the user risk assessment model, represents the empirical risk, which means that the model f is in the training sample set (x i ,y i ), n represents the number of samples, x i represents the input feature of the i-th sample, y i represents the output label of the i-th sample, and L represents the loss function;

[0096] Step S342: Structural risk minimization. The support vector machine adopts the structural risk minimization principle as a learning method to minimize complexity and empirical risk, so that the model still has high output function smoothness and generalization ability when the number of samples is limited. The output function of the support vector machine model is calculated using the following formula:

[0097] ;

[0098] In the formula, f(x) represents the output function of the support vector machine model, is the optimization coefficient in the Lagrangian function, represents the Lagrange multiplier in the support vector machine model, represents the kernel function, i and j both represent the index of the sample, ω represents the hyperplane weight vector, and b represents the offset constant, which is used to adjust the offset between the model's predicted value and the actual label. Represents the feature vector after mapping the input sample x to the high-dimensional feature space;

[0099] Step S343: Calculate the insensitive function loss, which helps the model to be more robust in the face of some small errors, while focusing on large differences beyond the insensitive interval and increasing attention to key errors. The formula used is as follows:

[0100] ;

[0101] In the formula, represents the insensitive function loss, x and y represent the input features of the sample and the corresponding output label, and ε represents the insensitive interval, which is used to determine the threshold of the loss value of 0;

[0102] Step S344: Convert the support vector machine problem into an optimization problem, and use the optimization method to solve the decision boundary constant of the support vector machine model to better achieve the user risk assessment task. The formula used is as follows:

[0103] ;

[0104] Where b1 represents the decision boundary constant of the support vector machine model, and C represents the penalty coefficient, which is used to control the complexity of the user risk assessment model.

[0105] By performing the above operations, a user risk assessment model is constructed, risk assessment indicators are determined and weights are set for each indicator based on their importance, empirical risk minimization and structural risk minimization are performed, the generalization ability of the model is improved, the support vector machine problem is transformed into an optimization problem, and the model runs more efficiently. This solves the technical problems of difficulty in systematically considering the degree of impact of different aspects of users on risks, weak model generalization ability, and the mismatch between insurance products and user risk conditions.

[0106] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the personalized product recommendation includes the following steps:

[0107] Step S41: According to the user risk assessment model, various risk factors of the customer are analyzed and quantitatively assessed to obtain the risk characteristics and potential risk points of the customer;

[0108] Step S42: Based on the results of the risk assessment model, targeted insurance products are selected that can effectively cover the corresponding risks and are in line with the financial affordability of the customer, and in-depth research is conducted on the terms, protection level and cost-effectiveness of the products to ensure that the most appropriate personalized insurance recommendation plan is provided to the customer.

[0109] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the feedback adjustment is specifically to collect user feedback and satisfaction information on the recommended products, analyze and evaluate the product recommendation mechanism and parameters based on the user feedback information, and optimize and adjust the recommendation model accordingly based on the evaluation results to improve the accuracy and satisfaction of subsequent recommendations.

[0110] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, and the present invention provides an insurance intelligent recommendation system based on user portrait, including a user data collection module, a user portrait construction module, a user risk assessment model construction module, a personalized product recommendation module and a feedback adjustment module;

[0111] The user data collection module specifically collects various information of the user, including basic information, financial information, health information and previous insurance status;

[0112] The user portrait module is constructed to clarify goals and indicators, analyze risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, use embedding matrices, calculate new hidden state vectors and assign user data weights, evaluate consumption capacity, and summarize insurance demand tendencies;

[0113] The user risk assessment model module is constructed to determine the risk assessment indicators and set weights for each indicator according to its importance, minimize the empirical risk and structural risk, improve the generalization ability of the model, transform the support vector machine problem into an optimization problem, and make the model run more efficiently;

[0114] The personalized product recommendation module specifically conducts detailed analysis and quantitative evaluation of various risk factors of customers, and specifically selects insurance products that can effectively cover the corresponding risks and meet the financial tolerance of customers;

[0115] The feedback adjustment module is specifically used to collect user feedback and satisfaction information on recommended products and perform model adjustment.

[0116] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0117] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0118] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent insurance recommendation method based on user portrait, characterized by: The method comprises the following steps: Step S1: user data collection; Step S2: Build a user profile, specifically to clarify goals and indicators, analyze risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, use embedding matrices, calculate new hidden state vectors and assign user data weights, evaluate consumption capacity, and summarize insurance demand tendencies; Step S3: Construct a user risk assessment model, specifically to determine the risk assessment indicators and set weights for each indicator according to its importance, define the risk level, and then score and calculate the total score to determine the risk level, perform empirical risk minimization and structural risk minimization, and transform the support vector machine problem into an optimization problem; Step S4: Personalized product recommendation; Step S5: feedback adjustment; In step S2, constructing the user portrait includes the following steps: Step S21: clarify the goals and indicators, and determine the specific aspects that need to be portrayed in the user portrait; Step S22: Data preprocessing, cleaning the user data, removing errors and duplicate information, normalizing the data, classifying user characteristics, and identifying them with labels, including age range, income level, and assets; Step S23: Analyze risk characteristics, including the following steps: Step S231: Use the user data as input to the feature extraction network, using the embedding matrix W glove , embed the user data into the vector space and calculate the embedding result of the user data. The formula used is as follows: ； In the formula, t represents the index of the position, i t represents the embedding result of user data at position t, W glove is the embedding matrix, m t is the one-hot vector representation at position t, where a one-hot vector is an encoding method used to represent categorical variables, where only one element is 1 and the rest are 0; Step S232: Calculate a new hidden state vector, use a long short-term memory network to further encode the embedding vector of each user data, and output a new hidden state vector. The formula used is as follows: ； Where b t Represents the new hidden state vector obtained after LSTM processing; Step S233: Use the soft attention mechanism to calculate and assign the weight of each user data. The formula used is as follows: ； ； Where pt is the non-normalized attention score used to balance the weight of each user's data, αt represents the attention weight after softmax normalization, Mb represents the parameter matrix used to learn the attention weight in the model, hb represents the additional bias term used for attention calculation, and T represents the total number of positions; Step S24: Assess consumption capacity by combining income, assets and consumption habits to determine consumption capacity; Step S25: Summarize insurance demand trends; In step S3, the construction of the user risk assessment model includes the following steps: Step S31: determine the risk assessment indicators and select the key factors affecting the risk based on the user portrait; Step S32: setting weights for each indicator, assigning different weights according to the importance of the indicator, and defining risk levels, including low, medium, and high risk levels; Step S33: scoring according to the scoring criteria, scoring each indicator according to the user's specific information, calculating the comprehensive total score through the weight and the score of each indicator, determining the risk level, and corresponding to the corresponding risk level according to the total score; Step S34: Optimizing the user risk assessment model, including the following steps: Step S341: Minimize the empirical risk and calculate the empirical risk. The formula used is as follows: ； In the formula, f represents the user risk assessment model, represents the empirical risk, which represents the model f in the training sample set (x i ,y i ), n represents the number of samples, x i represents the input feature of the i-th sample, y i represents the output label of the i-th sample, and L represents the loss function; Step S342: Minimize the structural risk and calculate the output function of the support vector machine model. The formula used is as follows: ； Where f(x) represents the output function of the support vector machine model, is the optimization coefficient in the Lagrangian function, represents the Lagrange multiplier in the support vector machine model, represents the kernel function, x j represents the input feature of the jth sample, ω represents the hyperplane weight vector, and b represents the offset constant, which is used to adjust the offset between the model's predicted value and the actual label. Represents the feature vector after mapping the input sample x to the high-dimensional feature space; Step S343: Calculate the insensitive function loss, the formula used is as follows: ； In the formula, represents the insensitive function loss, x and y represent the input features of the sample and the corresponding output label, and ε represents the insensitive interval, which is used to determine the threshold of the loss value of 0; Step S344: Convert the support vector machine problem into an optimization problem. The formula used is as follows: ； Where b1 represents the decision boundary constant of the support vector machine model, y j represents the output label of the jth sample, and C represents the penalty coefficient, which is used to control the complexity of the user risk assessment model.

2. The method for intelligent insurance recommendation based on user portrait according to claim 1, characterized in that: In step S1, the user data is collected, specifically, various information of the user is collected, including basic information, financial information, health information and previous insurance status.

3. The method for intelligent insurance recommendation based on user portrait according to claim 1, characterized in that: In step S4, the personalized product recommendation includes the following steps: Step S41: According to the user risk assessment model, various risk factors of the customer are analyzed and quantitatively assessed to obtain the risk characteristics and potential risk points of the customer; Step S42: Based on the results of the risk assessment model, targeted insurance products are selected that can effectively cover the corresponding risks and are in line with the financial affordability of the customer, and in-depth research is conducted on the terms, protection level and cost-effectiveness of the products to ensure that the most appropriate personalized insurance recommendation plan is provided to the customer.

4. The method for intelligent insurance recommendation based on user portrait according to claim 1, characterized in that: In step S5, the feedback adjustment specifically collects user feedback and satisfaction information on the recommended products, analyzes and evaluates the product recommendation mechanism and parameters based on the user feedback information, and optimizes and adjusts the recommendation model based on the evaluation results.

5. An insurance intelligent recommendation system based on user portrait, used to implement an insurance intelligent recommendation method based on user portrait as described in any one of claims 1 to 4, characterized in that: It includes user data collection module, user portrait building module, user risk assessment model building module, personalized product recommendation module and feedback adjustment module.

6. The user portrait-based insurance intelligent recommendation system according to claim 5, characterized in that: The user data collection module is specifically used to collect various information of users; The user portrait module is constructed to clarify goals and indicators, analyze risk characteristics through feature extraction networks, long short-term memory networks and soft attention mechanisms, use embedding matrices, calculate new hidden state vectors and assign user data weights, evaluate consumption capacity, and summarize insurance demand tendencies; The user risk assessment model module is constructed, specifically determining risk assessment indicators and setting weights for each indicator according to its importance, defining the risk level and calculating the total score to determine the risk level, performing empirical risk minimization and structural risk minimization, and converting the support vector machine problem into an optimization problem; The personalized product recommendation module specifically conducts detailed analysis and quantitative evaluation of various risk factors of customers, and specifically selects insurance products that can effectively cover the corresponding risks and meet the financial tolerance of customers; The feedback adjustment module is specifically used to collect user feedback and satisfaction information on recommended products and perform model adjustment.

Images