A product intelligent recommendation method based on customer credit

Abstract

The present application provides a method for intelligent product recommendation based on customer credit, including: obtaining customer credit score and collateral valuation feature information, judging the customer's credit qualification and initial credit limit according to a preset customer credit threshold and collateral value threshold; obtaining customer credit enhancement method information, including additional mortgage and provision of guarantee, obtaining a risk mitigation score according to the collateral valuation after credit enhancement and the customer credit score, and making a secondary adjustment to the credit limit; presenting an optimized product recommendation list to the customer, the list including product details, risk warnings and purchase channels, and recording customer feedback on the recommended products and purchase behavior, so as to continuously improve the performance of the intelligent recommendation system.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a product intelligent recommendation method based on customer credit. Background Art

[0002] In the research on intelligent product recommendation methods based on customer credit, when focusing on the valuation characteristics of customer collateral, a key technical problem is faced: how to dynamically adjust the product recommendation strategy to adapt to changes in customer credit enhancement methods when the customer's credit limit is adjusted due to changes in the collateral market. The impact mechanism of this dynamic adjustment on product matching also needs in-depth analysis. Specifically, when the value of the customer's collateral fluctuates, its credit limit will also be adjusted accordingly, which requires the product recommendation strategy to be able to respond in a timely manner and recalculate the customer's risk level and product preference based on the customer's current credit enhancement method, such as additional collateral, provision of guarantees, etc., so as to achieve dynamic optimization of recommendation results. However, this recommendation strategy adjustment based on customer credit enhancement methods may face many challenges in actual operation, such as how to obtain and quantify customer credit enhancement behavior data, how to establish and analyze the correlation between credit enhancement methods and product preferences, and how to balance the timeliness and stability of recommendation strategy adjustments. Summary of the invention

[0003] The present invention provides a product intelligent recommendation method based on customer credit, which mainly includes:

[0004] Obtain customer credit score and collateral valuation feature information, and determine the customer's credit qualification and initial credit line based on the preset customer credit threshold and collateral value threshold; monitor changes in the collateral market in real time, and when the collateral market price fluctuation exceeds the price fluctuation threshold, trigger the adjustment of the credit line, and adjust the customer's credit line once according to the current valuation of the collateral, the customer's credit score and risk preference, and generate an adjusted credit line; obtain customer credit enhancement information, including additional mortgages and provision of guarantees, and obtain a risk mitigation score based on the collateral valuation and customer credit score after credit enhancement, and make a secondary adjustment to the credit line; based on the secondary adjusted credit line and customer risk preference, select eligible financial products from the product library to form a candidate product set, and use a collaborative filtering algorithm, combined with customer historical transaction data and similar customer preferences. Well, sort and optimize the candidate products to generate a list of recommended financial products; evaluate the product suitability of the recommended financial products, judge the matching degree between the financial products and the customers according to the customers' financial status and risk tolerance factors, and use the decision tree algorithm to classify and score the product matching degree based on product attributes and customer attributes to obtain the product matching evaluation results; according to the product matching evaluation results, optimize the recommendation list, remove the products with low matching degree from the list, and retain the products with high matching degree; according to the customer feedback and product recommendation effect, dynamically update the product recommendation method to obtain the optimized product recommendation list; present the optimized product recommendation list to the customer, the list includes product details, risk warnings and purchase channels, and records the customer's feedback and purchase behavior on the recommended products to continuously improve the performance of the intelligent recommendation system.

[0005] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0006] The present invention discloses a product intelligent recommendation method based on customer credit. First, obtain the customer's credit score and collateral valuation feature information, and combine the preset credit rating and collateral value threshold to judge the customer's credit qualification and initial credit limit. Based on this, monitor the changes in the collateral market in real time, and trigger the credit limit adjustment process when the market price fluctuation exceeds the preset threshold. By analyzing the current valuation of the collateral, the customer's credit score and risk preference, dynamically adjust the credit limit and generate adjusted credit limit data. Further, obtain the customer's credit enhancement method information, such as additional mortgage or provision of guarantee, calculate the risk mitigation score according to the collateral valuation after credit enhancement and the customer's credit score, and make a secondary adjustment to the credit limit. Based on the adjusted credit limit and customer risk preference, select qualified financial products from the product library to form a candidate product set. Adopt a collaborative filtering algorithm, combine customer historical transaction data and similar customer preferences, sort and optimize the candidate products, and generate a recommended financial product list. The suitability of the recommended products is evaluated. According to the customer's financial status and risk tolerance, the decision tree algorithm is used to classify and score the product matching based on product attributes and customer attributes to obtain the matching evaluation results. According to the evaluation results, the recommendation list is optimized, the products with lower matching degrees are eliminated, and the products with higher matching degrees are retained. Through customer feedback and recommendation effects, the recommendation strategy and model parameters are dynamically updated to form an optimized product recommendation list. Finally, the optimized product recommendation list is presented to the customer, including product details, risk warnings and purchase channels, and the customer's feedback and purchase behavior on the recommended products are recorded as the data basis for optimizing the recommendation strategy and evaluation effects, and continuously improving the performance of the present invention. The present invention realizes personalized and accurate financial product recommendations by combining dynamic credit and intelligent recommendations, thereby improving customer satisfaction and business efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The present invention is a flowchart of a method for intelligently recommending products based on customer credit.

[0008] Figure 2 A schematic diagram of a method for intelligently recommending products based on customer credit according to the present invention.

[0009] Figure 3 This is another schematic diagram of a method for intelligently recommending products based on customer credit according to the present invention. DETAILED DESCRIPTION

[0010] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0011] like Figure 1-3 In this embodiment, a product intelligent recommendation method based on customer credit may specifically include:

[0012] S101. Obtain customer credit score and collateral valuation feature information, and determine the customer's credit eligibility and initial credit limit based on preset customer credit thresholds and collateral value thresholds.

[0013] The target customer's credit score record is obtained from the credit assessment database, and the risk rating index is calculated based on the credit score record. The market assessment value of the target customer's collateral is obtained from the assessment agency, and the credit qualification judgment value is obtained by multiplying the risk rating index by the collateral assessment value; if the credit qualification judgment value is greater than the credit qualification threshold, the benchmark credit line is calculated by using the ratio of the risk rating index to the collateral assessment value, and the benchmark credit line is adjusted according to the target customer's repayment ability index to obtain the credit line judgment value; the customer groups are divided according to the proportional relationship between the credit line judgment value and the collateral assessment value, the product adaptation coefficient is calculated using the risk rating index and the collateral discount rate, and the product sequence value is obtained through the collateral liquidity index and the customer risk tolerance index.

[0014] Exemplarily, the credit score records of the target customers in the previous three years are obtained from the credit assessment database and the risk rating index is calculated based on the default history. The collateral market assessment value and depreciation coefficient are obtained from the assessment agency. The initial credit qualification judgment value is obtained by multiplying the risk rating index by the collateral assessment value. If the initial credit qualification judgment value is greater than the preset threshold, the benchmark credit limit is calculated using the ratio of the risk rating index to the collateral assessment value, and the benchmark credit limit is adjusted according to the customer repayment ability index to obtain the credit limit judgment value. Customer groups are divided according to the proportional relationship between the credit limit judgment value and the collateral assessment value, the product adaptation coefficient is calculated using the risk rating index and the collateral discount rate, and the product sequence value is obtained through the collateral liquidity index and the customer risk tolerance index. According to the product sequence value, a list of qualified products is screened from the product database, and a product recommendation list is generated by matching the credit limit judgment value with the product access conditions, and the loan amount of the recommended product is divided into intervals based on the collateral assessment value. The risk rating index is calculated based on the customer's credit record, including factors such as repayment timeliness and credit line utilization rate. By analyzing historical default records to build a scoring matrix, each default behavior of the customer will be recorded and assigned different weights, such as 0.6 for overdue payments of more than 90 days and 0.3 for overdue payments of less than 60 days. The risk coefficient is calculated by combining the number of defaults and the amount. For example, a customer has overdue payments of more than 90 days twice in the past three years, involving an amount of 50,000 yuan, and 1 overdue payment of 30 days, involving an amount of 10,000 yuan. The risk rating index is 78. The market valuation of collateral needs to consider the liquidity and value stability of assets. Different types of collateral use different valuation methods. The basic value of real estate collateral is calculated by the average transaction price and building area of ​​the surrounding market, and then adjusted according to factors such as house age, location, and decoration status to obtain the final valuation. For example, a commercial house with an area of ​​120 square meters and an average transaction price of 15,000 yuan / square meter in the surrounding area, considering the depreciation rate of 0.85 for a 5-year house age, the final valuation is 1,530,000 yuan. In the process of determining the credit limit, the repayment ability index is calculated by the ratio of the customer's average monthly income to fixed expenses. The income sources include wages, rent, investment income, etc., and the expenses include fixed debts such as mortgages and car loans. The total monthly payment does not exceed 50% of the monthly income. A customer has a monthly income of 30,000 yuan and a monthly mortgage payment of 8,000 yuan. According to his repayment ability index of 0.73, he can obtain a benchmark credit limit of 500,000 yuan. The product adaptation coefficient combines the customer's risk characteristics with the product attributes, and recommends different types of credit products to customers with different risk levels. Mortgage products are recommended to customers with high risk rating indexes, and credit products can be recommended to low-risk customers. The collateral discount rate is determined according to the liquidity of the assets, such as 0.7 for residential assets, 0.6 for commercial buildings, and 0.5 for land use rights. The product sequence value 832 indicates that the product is suitable for customers with medium risk and high-quality real estate mortgages.Product access conditions include hard indicators such as minimum credit score requirements and minimum guarantee rate requirements, and the customer's various indicators are calculated and scored for matching. For example, if a customer has a credit score of 810, a guarantee rate of 180%, and an income-liability ratio of 0.35, it is suitable to recommend a housing mortgage loan product with a maximum limit of 70% of the assessed value of the collateral, and a loan limit range of 500,000 to 1 million yuan. The final credit limit is determined within the range based on the customer's actual capital needs.

[0015] S102. Monitor the changes in the collateral market in real time. When the price fluctuation of the collateral market exceeds the price fluctuation threshold, the credit line adjustment is triggered. According to the current valuation of the collateral, the customer's credit score and risk preference, the customer's credit line is adjusted once to generate the adjusted credit line.

[0016] Obtain the regional market price index and transaction volume data of the collateral from the market data source, and obtain the price change rate and market liquidity index based on the market price index and transaction volume data; if the price fluctuation exceeds the price fluctuation threshold, use the latest market price index and price change rate to calculate the current market value of the collateral; calculate the adjusted collateral value based on the price adjustment coefficient and the current market value of the collateral, and obtain the risk tolerance coefficient through the customer risk preference index and the credit score change rate; calculate the credit line adjustment value using the adjusted collateral value and the existing credit line data, and obtain the adjusted credit line by proportionally calculating the credit line adjustment value based on the risk tolerance coefficient.

[0017] Exemplarily, the regional market price index and transaction volume data of the collateral are obtained from the market data source, the price change rate and market liquidity index are calculated for different types of collateral, and the price volatility judgment value is obtained according to the collateral depreciation rate and price trend index. If the price volatility judgment value exceeds the preset risk volatility threshold, the latest market price index and price change rate are used to calculate the current market value of the collateral, and the price adjustment coefficient is obtained through the volatility trend during the price monitoring period. The adjusted collateral value is calculated according to the price adjustment coefficient and the current market value of the collateral, and the risk tolerance coefficient is generated based on the customer risk preference index and the credit score change rate, and the collateral value is adjusted by the risk tolerance coefficient. The existing credit limit data is obtained from the customer credit database, and the credit limit adjustment value is calculated using the adjusted collateral value and the existing credit limit data, and the credit limit adjustment value is proportionally calculated according to the risk tolerance coefficient to obtain the adjusted credit limit.

[0018]

[0019] , C final Indicates the adjusted credit limit, C currentRepresents the existing credit line, α represents the risk tolerance coefficient, and ΔC represents the credit line adjustment value. The market price index is calculated by collecting transaction data of the same type of assets in a specific area. For example, the residential market price index of a certain city rose from 15,000 yuan / square meter in January to 16,200 yuan / square meter in February, with a price change rate of 8%. The transaction volume increased from 850 units to 1,100 units during the same period, and the market liquidity index was 1.29. The collateral depreciation rate takes into account factors such as house age, location, and decoration. The annual depreciation rate of ordinary residential houses with a house age of 5 years is 3%. The price trend indicator shows that the housing prices in this area have risen for three consecutive months, and the fluctuation judgment value is 0.86. When the fluctuation judgment value exceeds the preset threshold of 0.75, the price revaluation mechanism is triggered. The original assessed value of a residential collateral is 3 million yuan. The latest market price index shows that the average price in the region has risen by 12%. The current market value is calculated to be 3.36 million yuan. The housing price trend in this area has continued to rise during the monitoring period, and the price adjustment coefficient is 1.15, indicating that market expectations are still bullish. The customer risk preference index is comprehensively assessed based on investment history, trading behavior and other data, with a value range of 0-1. The larger the value, the stronger the risk tolerance. In a customer's recent investment portfolio, fixed-income assets account for 70%, stock assets account for 30%, and the risk preference index is 0.45. The credit score dropped from the original 850 points to 820 points, with a change rate of -3.5%. Combined with the risk preference index, the risk tolerance coefficient is 0.82. In the process of credit line adjustment, the collateral value is used as the core reference. A customer's existing credit line is 1.8 million yuan, and the collateral value has increased by 12%. Considering the risk tolerance coefficient of 0.82, the credit line adjustment value is calculated to be 1.976 million yuan. Based on factors such as customer rating and guarantee rate, the final credit line is determined to be 1.95 million yuan, an increase of 150,000 yuan from the original line. When market conditions change drastically, the dynamic adjustment mechanism of credit line effectively balances asset value fluctuations and credit risks. The price monitoring mechanism adopts differentiated monitoring frequencies for different types of collateral, with residential assets monitored once a month, commercial real estate monitored once a week, and stock pledges monitored daily. When monitoring finds that price fluctuations exceed the threshold, the adjustment process is initiated to ensure that the credit limit matches the actual value of the collateral. For example, a new subway station is built in the surrounding area of ​​a commercial real estate project, which drives up the value of the business district and the value of the collateral increases by 25%, triggering an increase in the credit limit.

[0020] S103. Obtain information on the customer's credit enhancement methods, including additional collateral and provision of guarantees. Based on the valuation of the collateral after credit enhancement and the customer's credit score, obtain a risk mitigation score and make a secondary adjustment to the credit limit.

[0021] Obtain the customer's additional collateral list and guarantor list from the credit enhancement database, calculate the appraisal value for the collateral list, calculate the performance ability score for the guarantor list, and obtain the guarantee coverage coefficient; calculate the collateral value ratio according to the guarantee coverage coefficient, use the performance ability score to calculate the guarantee coverage supplement value, and obtain the risk mitigation coefficient through the collateral value ratio and the guarantee coverage supplement value; calculate the credit enhancement combination score using the risk mitigation coefficient and the customer's credit score, and calculate the credit line adjustment ratio through the credit enhancement combination score; if the guarantee coverage coefficient is greater than the coverage coefficient threshold, multiply the credit line adjustment ratio by the existing credit line to obtain the final credit line.

[0022] For example, the customer's additional collateral list and guarantor list are obtained from the credit enhancement database, the assessment value and liquidity index of the additional collateral are calculated, and the guarantee coverage factor is calculated based on the guarantor's credit rating and performance ability score. If the guarantee coverage factor is greater than the preset threshold, the additional collateral assessment value is used to calculate the collateral value ratio, the guarantee coverage supplement value is calculated based on the guarantor's performance ability score, and the risk mitigation coefficient is obtained based on the collateral value ratio and the guarantee coverage supplement value.

[0023]

[0024] , MVR represents the collateral value ratio, MVa represents the assessed value of the original collateral, MVb represents the assessed value of the additional collateral, and LTV represents the ratio of the total loan amount to the collateral value.

[0025]

[0026] , S gc Indicates the guarantee coverage supplement value, C s Indicates the guarantee amount, S g It indicates the performance score of the guarantor and calculates the supplementary amount of guarantee coverage, that is, the amount of guarantee that still needs to be supplemented after considering the performance ability of the guarantor. The risk mitigation score is calculated based on the risk mitigation coefficient and the guarantee coverage coefficient, and the credit enhancement combination score is calculated using the customer credit score and the guarantor's credit rating. The credit limit adjustment ratio is obtained through the risk mitigation score and the credit enhancement combination score.

[0027]

[0028] , where R f represents the risk mitigation factor, R m represents the collateral value ratio, S grepresents the guarantee coverage supplement value, and g represents a combined function, which combines the collateral value ratio and the guarantee coverage supplement value to obtain the risk mitigation coefficient. The function can be set according to risk preference and business needs, such as weighted average function, minimum function or product function, etc., reflecting the risk mitigation effect under the joint action of collateral and guarantee. The credit line adjustment value is calculated by multiplying the credit line adjustment ratio by the existing credit line, and the credit line adjustment value is proportionally restricted according to the guarantee coverage coefficient to obtain the final credit line. The credit enhancement methods include two main types: adding collateral and introducing guarantors. When calculating the guarantee coverage coefficient, the value of the collateral and the strength of the guarantor need to be comprehensively considered. Taking a corporate loan as an example, the customer added a commercial property worth 8 million yuan as collateral, and introduced a listed company with a market value of 5 billion yuan to provide joint and several guarantees. After evaluation, the collateral liquidity index was 0.75, the guarantor's credit rating was AAA, and the performance ability score was 0.95. The calculated guarantee coverage coefficient was 2.8. When the guarantee coverage coefficient exceeds the preset threshold of 2.0, the risk mitigation assessment process is initiated. The assessed value of the additional collateral of the enterprise accounts for 160% of the loan amount, and the calculated collateral value ratio is 1.6. The guarantor's performance ability score of 0.95 is converted into a guarantee coverage supplement value of 0.85. The risk mitigation coefficient calculated by combining the two indicators is 1.35, indicating that the credit enhancement measures have significantly reduced the credit risk. The risk mitigation score reflects the overall risk resistance of the credit enhancement measures. The risk mitigation score of 3.78 is obtained by multiplying the guarantee coverage coefficient of 2.8 with the risk mitigation coefficient of 1.35. The customer's own credit score is 750 points, and the guarantor's credit rating is AAA, corresponding to 950 points. The weighted credit enhancement combination score of the two is 875 points. Based on the risk mitigation score of 3.78 and the credit enhancement combination score of 875, the credit line adjustment ratio is determined to be 1.4. The original credit line is 5 million yuan, and the credit line adjustment value of 7 million yuan is obtained after calculating the credit line adjustment ratio of 1.4 times. Considering the guarantee coverage factor of 2.8, the credit line adjustment ratio is required not to exceed 50% of the original credit line, and the final credit line is determined to be 6.5 million yuan. In another case, a certain enterprise's credit enhancement measure is to add accounts receivable as collateral. Due to the long realization cycle and large value fluctuations, the liquidity index is only 0.4, and the guarantee coverage factor of 1.2 did not trigger the credit line adjustment process. The risk mitigation effect of credit enhancement measures is closely related to the characteristics of the credit enhancement method. The combination of real estate mortgage and high-quality enterprise guarantee has a strong risk mitigation effect, while accounts receivable, inventory and other assets with poor liquidity require a stricter discount rate when used as credit enhancement measures. A certain enterprise used 2.5 million yuan of inventory as additional collateral. After considering the realization discount of 0.5, the guarantee coverage factor was only 1.25, which did not meet the standard for initiating credit line adjustment. In practice, by dynamically monitoring the guarantee coverage factor, the risk mitigation effect of credit enhancement measures is continuously evaluated, and the credit plan is adjusted in a timely manner.

[0029] S104. Based on the credit limit and customer risk preference after the second adjustment, select eligible financial products from the product library to form a candidate product set. Use a collaborative filtering algorithm, combine customer historical transaction data and similar customer preferences, sort and optimize the candidate products, and generate a recommended financial product list.

[0030] A list of financial products is obtained from the product database. If the credit limit and risk preference of the product meet the entry threshold, the product is stored in the candidate product set. For the historical transaction data in the customer transaction record library, the customer activity level is calculated based on the transaction frequency and transaction amount, and the customer preference characteristics are clustered to obtain similar user groups. The product purchase weight is calculated based on the historical transaction records of the similar user groups, and the candidate product set is collaboratively filtered to obtain a product recommendation index. For each product in the candidate product set, the final recommendation coefficient is calculated using the product recommendation index and the customer activity level, and the candidate product set is sorted and filtered by the final recommendation coefficient to obtain a product recommendation list.

[0031] Exemplarily, a list of financial products is obtained from a product database, the product access threshold is calculated based on the credit limit and risk preference, eligible candidate products are screened based on the product risk coefficient and yield index, and the candidate products are classified and grouped by product type attributes to obtain a candidate product set. Historical transaction data is extracted from the customer transaction record library, customer activity is calculated based on transaction frequency and amount, customer preference characteristics are counted based on transaction product types, and similar user groups are obtained by calculating the similarity of customer groups through the nearest neighbor algorithm. Product purchase weights are calculated based on historical transaction records of similar user groups, user preference scores are calculated using customer historical transaction data, and recommendation priority values ​​are calculated for the candidate product set based on the collaborative filtering method to obtain a product recommendation index. The final recommendation coefficient is calculated using the product recommendation index and customer activity, and the candidate product set is sorted and screened by the final recommendation coefficient to obtain a product recommendation list. The product access threshold is set based on the customer's risk tolerance and financial strength. For example, for customers with a credit line of 5 million yuan and a risk preference index of 0.6, the upper limit of the product risk coefficient is set to 0.75, and the annualized yield range is 4%-8%. Fixed-income wealth management products and structured deposit products are selected to form a candidate product set, including 15 products that meet the access conditions. Customer activity reflects the characteristics of transaction behavior. The transaction frequency and amount scores in the past year are calculated. The activity level of customers with an average of 2 transactions per month and a single amount of 300,000 yuan is 0.65. Based on the characteristics of transaction types, terms, and yields, the customer preference feature vector is summarized, and the nearest neighbor algorithm is used to find 300 customer groups with a similarity of more than 85% to form a similar user group. The product purchase weight reflects the group's transaction preference. In the similar user group, the purchase of structured deposit products accounts for 35%, and the purchase of fixed-income wealth management products accounts for 45%. The corresponding product weights are 0.35 and 0.45, respectively. In the target customer's historical transactions, the number of structured deposit purchases is 8 times, and the number of fixed-income products is 6 times. The calculated user preference scores are 0.57 and 0.43, respectively. The recommendation priority value is calculated for each product in the candidate product set to obtain the product recommendation index. Taking a certain structured deposit product as an example, with a term of 180 days and an expected rate of return of 4.5%, combined with a product weight of 0.35 and a user preference score of 0.57, the recommendation priority value is calculated to be 0.48. Considering the customer activity level of 0.65, the final recommendation coefficient is 0.62. The recommendation coefficients of all products in the candidate product set are calculated and sorted, and the top 5 products are selected to form the recommendation list. Product recommendation strategies vary according to the characteristics of different customer groups. High-net-worth customer groups prefer medium- and long-term, higher-yield products, and the yield weight is higher in the calculation of the recommendation coefficient. Ordinary retail customers are more concerned about liquidity, and the recommendation weight of short-term products is increased accordingly. The selection criteria for similar user groups of retail customers also pay more attention to the similarity of income level and professional background, enhancing the adaptability of recommended products.The market environment also needs to be considered in the product recommendation process. In the downward interest rate cycle, the yield of fixed-income products generally declines, and structured deposit products have a higher recommendation weight due to their yield advantage. In the period of increased market volatility, the recommendation weight of principal-guaranteed products is increased to reflect the risk control orientation. The product recommendation list is dynamically updated to ensure that the recommended products always meet the characteristics of customer needs.

[0032] S105. Conduct product suitability assessment on recommended financial products, determine the matching degree between financial products and customers according to the customers’ financial status and risk tolerance factors, and use decision tree algorithm to classify and score product matching based on product attributes and customer attributes to obtain product matching assessment results.

[0033] Obtain the customer's monthly income data and fixed expenditure data, and calculate the financial strength score, investment experience value and risk tolerance index based on the monthly income data and fixed expenditure data; if the risk tolerance index exceeds the preset risk threshold, use the customer's rate of return requirement and liquidity demand to calculate the investment characteristic index, and use the investment characteristic index to screen the product list to obtain a preliminary set of products; calculate the basic matching score, cognitive matching score and characteristic matching score based on the preliminary set of products and the investment experience value, and obtain the comprehensive matching degree of the products through the three matching score calculations; obtain product rating data from the product rating library, use the comprehensive matching degree of the product and the product rating data to calculate the final suitability score, and classify the products according to the final suitability score to obtain a product matching evaluation result.

[0034] For example, financial status indicators and investment years records are obtained from the customer database, and the financial strength score is calculated based on the monthly income level and fixed expenditure ratio. The investment experience value is obtained through the statistics of investment transaction frequency and amount, and the risk tolerance index is obtained according to the customer age level and risk assessment questionnaire. If the risk tolerance index exceeds the preset threshold, the investment characteristic index is calculated using the rate of return requirement value and liquidity demand, and the customer classification rules are constructed based on the decision tree algorithm. The product recommendation list is preliminarily screened through the customer classification rules to obtain the product preliminary set. The basic matching score is calculated based on the product preliminary set, the cognitive matching score is calculated using the investment experience value and the product complexity index, and the characteristic matching score is calculated based on the investment characteristic index and the product attributes. The comprehensive matching degree of the product is calculated through the three matching scores.

[0035]

[0036] , S cm represents the cognitive matching score, S ie represents the investment experience score, S pcRepresents the product complexity score, α represents the weight of the investment experience score in the cognitive matching score, and (1-α) represents the weight of the product complexity score in the cognitive matching score. Obtain product comprehensive rating data from the product rating database, use the product comprehensive matching degree and product rating data to calculate the final suitability score, and classify the products according to the final suitability score to obtain the product matching evaluation result. The financial strength score is calculated by analyzing the customer's financial situation. For example, for a customer with a monthly income of 50,000 yuan and a fixed expenditure of 40%, the monthly investment amount is 30,000 yuan. Combined with the current asset scale of 3 million yuan, the financial strength score is 0.75. The investment experience value is evaluated based on historical transaction data. The customer has completed 26 investment transactions in the past three years, with an average single amount of 150,000 yuan, and the investment experience value is 0.68. The risk assessment questionnaire contains 20 assessment dimensions, covering factors such as investment goals, loss tolerance, and investment period. The customer scored 85 points, corresponding to a risk tolerance index of 0.72. The decision tree algorithm classifies customers based on their investment characteristics. Customers with a required rate of return of 8% and a liquidity requirement of 0.6 are calculated to have an investment characteristic index of 0.65, which means they are conservative investors. The preliminary product selection set includes 5 products, covering fixed income and structured financial products, with product yields of 5%-9% and investment periods ranging from 3 to 12 months. The basic matching score reflects the degree of match between the risk-return characteristics of the product and the risk preferences of the customer. Taking a certain structured deposit as an example, the product risk rating is medium, the customer risk tolerance index is 0.72, and the basic matching score is calculated to be 0.85. The cognitive matching score measures the customer's understanding of the product. The complexity index of the product is 0.45, which is lower than the customer's investment experience value of 0.68, and the cognitive matching score is 0.82. The feature matching score compares the product attributes with the customer's investment characteristics. The product term is 6 months and the expected rate of return is 7%, which is highly consistent with the customer's investment characteristic index of 0.65, and the feature matching score is 0.88. The comprehensive rating data of products includes three dimensions: liquidity rating, credit rating, and market rating. The three ratings of a certain structured deposit product are 4 stars, 5 stars, and 4 stars respectively. Combined with the comprehensive matching degree of 0.85, the final suitability score is 0.83, and the product is rated as the most suitable. The product matching assessment is set at four levels, 0.8 or above is the most suitable, 0.6-0.8 is suitable, 0.4-0.6 is relatively suitable, and 0.4 or below is not suitable. Compared with another wealth management product with an investment period of 2 years and an expected rate of return of 10%, although the rate of return is higher, the term does not match the customer's liquidity demand of 0.6, and the feature matching score is only 0.62. At the same time, the product structure is relatively complex, with a cognitive matching score of 0.71 and a final suitability score of 0.65, and the assessment result is suitable. The matching assessment of different types of products has different focuses. Fixed income products focus on term matching, structured products pay more attention to cognitive matching, and net value products focus on risk matching.

[0037] S106. According to the product matching evaluation results, the recommendation list is optimized, the products with low matching are removed from the list, and the products with high matching are retained. According to customer feedback and product recommendation effects, the product recommendation method is dynamically updated to obtain an optimized product recommendation list.

[0038] Obtain matching evaluation data in the product recommendation list, screen and grade the recommended products according to the matching evaluation data and the product matching threshold to obtain a screened product set; for the screened product set, use the customer feedback index and the purchase intention index to calculate the product recommendation score, and re-sort the screened product set by the product recommendation score to obtain a sorted product set; if the product recommendation score is lower than the product recommendation score threshold, calculate the product acceptance level, and optimize the sorted product set according to the product acceptance level to obtain an optimized product set; for the optimized product set, calculate the recommendation accuracy according to the product recommendation records and the selection preference data, use the recommendation accuracy and the product acceptance level to calculate the product conversion rate, and screen the optimized product set by the product conversion rate to obtain an optimized product recommendation list.

[0039] Exemplarily, the matching evaluation data is obtained from the product recommendation list, the recommended products are screened and graded based on the product matching threshold, the product recommendation score is calculated using the customer feedback index and the purchase intention index, and the product list is reordered by the product recommendation score. If the product recommendation score is lower than the preset threshold, the product acceptance is calculated using the feedback satisfaction and the user response rate, the product acceptance is optimized and adjusted based on the gradient boosting tree algorithm, and the unqualified products are screened and eliminated by the product acceptance. The recommendation accuracy is calculated based on the product recommendation record and the selection preference data, the product conversion rate is calculated using the recommendation accuracy and the product acceptance, and the recommended parameters are dynamically updated based on the product conversion rate to obtain the parameter optimization set. The recommended products are recalculated using the parameter optimization set, and the product list is optimized and screened by the calculation results to obtain the final recommendation list. The product recommendation list initially contains 10 financial products, each of which has a matching evaluation result, such as a structured deposit product with a matching degree of 0.85, a customer feedback index of 0.78, and a purchase intention index of 0.82, and the product recommendation score is calculated to be 0.82. The product matching threshold is set to 0.75, and the recommended products are graded and screened, retaining 7 products above the threshold. The product acceptance reflects the customer's overall evaluation of the product. The recommendation score of a fund product is 0.73, which is lower than the preset threshold of 0.75. Further analysis shows that the customer feedback satisfaction is 0.65, the user response rate is only 0.55, and the calculated product acceptance is 0.62. Through the gradient boosting tree algorithm analysis, it is found that the mismatch between the product term and the customer's liquidity demand is the main reason for the low acceptance. The product is removed from the recommendation list. The recommendation accuracy measures the accuracy of product recommendations. A financial product has been recommended 100 times recently, customers have viewed the details 80 times, and 35 transactions have been completed. The recommendation accuracy is 0.75. Combined with the product acceptance of 0.82, the product conversion rate is calculated to be 0.78. Based on the conversion rate, the recommendation parameters are adjusted, the product income weight is increased from 0.4 to 0.5, and the term weight is reduced from 0.3 to 0.2, forming a new parameter optimization set. Different optimization strategies are adopted for different types of products. Fixed income products mainly focus on the matching of yield and term, while net value products pay more attention to risk-return characteristics. The original weight configuration of a net value financial product was risk 0.4, return 0.3, and liquidity 0.3. After optimization, it was adjusted to risk 0.5, return 0.2, and liquidity 0.3, which emphasized risk control attributes. The product recommendation effect is also closely related to the market environment. During periods of increased market volatility, the acceptance of stable products has increased significantly. For a principal-guaranteed structured deposit product, the recommendation accuracy rate was 0.65 in the early stage, which increased to 0.82 after market fluctuations, and the product acceptance rate increased from 0.72 to 0.85. In the process of parameter optimization, the weight of the security indicator is appropriately increased to make the recommendation results more in line with the market environment. Through dynamic optimization, the overall matching degree of the recommendation list has been significantly improved.The original recommendation list had an average recommendation score of 0.76 for the 10 products, and the average score of the 6 products retained after optimization increased to 0.83. Product types are also more diversified, including structured deposits, fixed income wealth management, net value products, etc., to meet the investment needs of customers at different levels. We continue to track customer feedback data, adjust the parameter optimization set in a timely manner, and maintain dynamic optimization of recommendation effects.

[0040] S107. Present the optimized product recommendation list to the customer, the list including product details, risk warnings and purchase channels, and record the customer's feedback and purchase behavior on the recommended products to continuously improve the performance of the intelligent recommendation system.

[0041] The risk warning content is graded and labeled according to the product feature information and product rating classification in the product information library to obtain product information with risk levels; the product display layout is personalized using the product information with risk levels and user interaction habit data to obtain a product display page; if the user triggers the interaction of the product display page, the user browsing trajectory is recorded, the user behavior characteristics are extracted from the user browsing trajectory, and the user attention index is calculated according to the user behavior characteristics; the content feedback score is calculated according to the user attention index and the page click record, and the display layout is optimized and sorted according to the content feedback score and the channel conversion rate to obtain a final display plan.

[0042] Exemplarily, characteristic information of recommended products is obtained from the product information library, risk warning content is graded and labeled based on product rating classification, user interaction habit data is used to personalize product display layout, and display page layout is generated through product details and purchase channel information. If the user triggers page interaction, the interactive data collection tool is used to record the user's browsing trajectory, and the browsing behavior is feature extracted based on the deep learning algorithm. The user attention index is obtained through the stay time and click frequency data. The content feedback score is calculated based on the user attention index and the page click record, and the channel conversion rate is calculated using the user evaluation data and the channel selection record. The display layout is optimized and updated based on the content feedback score and the channel conversion rate. The optimized display layout is used to calculate the product display weight, and the recommended list is re-sorted by the product display weight to obtain the final display plan. The personalized layout of the product display layout is optimized based on the user's reading and interaction habits. For example, if the risk level of a financial product is medium, it is highlighted with a yellow warning sign on the page, and the risk warning content occupies 25% of the page area. For users who are used to browsing quickly, the core information of the product such as the rate of return, term, and minimum purchase amount are arranged in a prominent position at the top of the page, and the detailed terms are displayed through a folding panel. The characteristics of user browsing behavior reflect the information acquisition preference. The interactive data collection shows that the user stays for 45 seconds in the income details, 15 seconds in the risk warning part, and 30 seconds in the product description. The calculated user attention index distribution is 0.5 for income, 0.17 for risk, and 0.33 for description. The click behavior record shows that the user repeatedly viewed the income calculation example 3 times, and the corresponding section received a high degree of content attention. The content feedback score reflects the effectiveness of information display. When users view a certain structured deposit product on the mobile terminal, the average page stay time is 180 seconds, and the number of clicks on each functional module is 12 times. The calculated content feedback score is 0.85. The product purchase channels include online banking, mobile banking, and offline outlets. Users stay in the mobile banking channel for the longest time, complete 3 purchase operations, and the channel conversion rate is 0.75. Based on the analysis of user interaction data, the display layout is optimized in a targeted manner. The area related to income is expanded to 40%, and the historical income trend chart is highlighted. The risk warning content is displayed in a layered manner, with important reminders displayed fixedly and detailed instructions folded and retracted. The purchase channel recommendation is sorted based on the conversion rate, with the mobile banking channel ranked first, and a quick purchase entrance is added. There are obvious differences in the interaction characteristics of different customer groups. Young users prefer concise information display, with an average page stay time of 90 seconds, and they are mainly concerned about yield and operational convenience. Middle-aged users pay more attention to product details, with an average stay time of 240 seconds, and the click-through rate of risk warning content is relatively high. The display layout is designed differently for users of different age groups to highlight the key information they are concerned about. Product type also affects the optimization direction of the display plan. Fixed-income products highlight the return calculation method and historical repayment records, net value products emphasize risk-return characteristics and investment strategy descriptions, and structured products focus on displaying the return composition and triggering conditions.Based on user feedback, we continuously optimize display effects and improve the accuracy of product information delivery.

[0043] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A product intelligent recommendation method based on customer credit, characterized in that: The method comprises: Obtain customer credit score and collateral valuation feature information, and determine the customer's credit qualification and initial credit line based on the preset customer credit threshold and collateral value threshold; monitor changes in the collateral market in real time, and when the collateral market price fluctuation exceeds the price fluctuation threshold, trigger the adjustment of the credit line, and adjust the customer's credit line once according to the current valuation of the collateral, the customer's credit score and risk preference, and generate an adjusted credit line; obtain customer credit enhancement information, including additional mortgages and provision of guarantees, and obtain a risk mitigation score based on the collateral valuation and customer credit score after credit enhancement, and make a secondary adjustment to the credit line; based on the secondary adjusted credit line and customer risk preference, select eligible financial products from the product library to form a candidate product set, and use a collaborative filtering algorithm, combined with customer historical transaction data and similar customer preferences. Well, sort and optimize the candidate products to generate a recommended list of financial products; evaluate the suitability of the recommended financial products, judge the matching degree between the financial products and the customers according to the customers' financial status and risk tolerance factors, and use the decision tree algorithm to classify and score the product matching degree based on product attributes and customer attributes to obtain the product matching evaluation results; optimize the recommendation list according to the product matching evaluation results, remove the products with low matching degree from the list, and retain the products with high matching degree; dynamically update the product recommendation method according to customer feedback and product recommendation effects to obtain the optimized product recommendation list; present the optimized product recommendation list to the customers, the list includes product details, risk warnings and purchase channels, and records the customers' feedback and purchase behavior on the recommended products to continuously improve the performance of the intelligent recommendation system; The product suitability assessment of the recommended financial products is conducted to determine the matching degree between the financial products and the customers according to the customers' financial status and risk tolerance factors. The decision tree algorithm is used to classify and score the matching degree of the products based on the product attributes and customer attributes to obtain the product matching evaluation results, which specifically include: Obtaining monthly income data and fixed expenditure data of the customer, and calculating a financial strength score, an investment experience value, and a risk tolerance index based on the monthly income data and fixed expenditure data; If the risk tolerance index exceeds the preset risk threshold, the investment characteristic index is calculated using the customer's required rate of return and liquidity requirement, and the product list is screened by the investment characteristic index to obtain a preliminary selection of products; The basic matching score, cognitive matching score and feature matching score are calculated based on the product preliminary selection set and the investment experience value, and the comprehensive matching degree of the product is calculated through the three matching scores; The product rating data is obtained from the product rating database, the final suitability score is calculated using the comprehensive product matching degree and the product rating data, and the product is graded according to the final suitability score to obtain a product matching evaluation result.

2. The method according to claim 1, characterized in that The obtaining of customer credit score and collateral valuation feature information, and judging the customer's credit qualification and initial credit limit according to preset customer credit threshold and collateral value threshold, includes: Obtain the target customer's credit score record from the credit evaluation database, calculate the risk rating index based on the credit score record, obtain the target customer's collateral market assessment value from the assessment agency, and obtain the credit qualification determination value by multiplying the risk rating index by the collateral assessment value; If the credit qualification determination value is greater than the credit qualification threshold, the benchmark credit line is calculated using the risk rating index and the collateral assessment value ratio, and the benchmark credit line is adjusted according to the target customer's repayment ability index to obtain the credit line determination value; The customer groups are divided according to the proportional relationship between the credit limit determination value and the collateral assessment value, the product adaptation coefficient is calculated using the risk rating index and the collateral discount rate, and the product sequence value is obtained through the collateral liquidity index and the customer risk tolerance index.

3. The method according to claim 1, characterized in that The real-time monitoring of the changes in the collateral market, when the collateral market price fluctuation exceeds the price fluctuation threshold, triggers the adjustment of the credit line, and adjusts the customer's credit line once according to the current valuation of the collateral, the customer's credit score and risk preference, and generates an adjusted credit line, including: Obtaining the regional market price index and transaction volume data of the collateral from the market data source, and obtaining the price change rate and market liquidity index based on the market price index and transaction volume data; If the price fluctuation exceeds the price fluctuation threshold, the latest market price index and price change rate are used to calculate the current market value of the collateral; Calculate the adjusted collateral value based on the price adjustment coefficient and the current market value of the collateral, and obtain the risk tolerance coefficient through the customer risk preference index and the credit score change rate; The adjusted collateral value and the existing credit line data are used to calculate the credit line adjustment value, and the adjusted credit line is obtained by proportionally calculating the credit line adjustment value according to the risk tolerance coefficient.

4. The method according to claim 1, characterized in that The information on the customer's credit enhancement method is obtained, including additional collateral and provision of guarantees. According to the valuation of the collateral after credit enhancement and the customer's credit score, the risk mitigation score is obtained, and the credit limit is adjusted twice, including: Obtain the customer's additional collateral list and guarantor list from the credit enhancement database, calculate the assessed value for the collateral list, calculate the performance ability score for the guarantor list, and obtain the guarantee coverage coefficient; Calculate the collateral value ratio according to the guarantee coverage coefficient, calculate the guarantee coverage supplement value using the performance ability score, and obtain the risk mitigation coefficient through the collateral value ratio and the guarantee coverage supplement value; The risk mitigation coefficient and the customer credit score are used to calculate the credit enhancement combination score, and the credit line adjustment ratio is calculated based on the credit enhancement combination score; If the guarantee coverage factor is greater than the coverage factor threshold, the credit line adjustment ratio is multiplied by the existing credit line to obtain the final credit line.

5. The method according to claim 1, characterized in that According to the credit limit and customer risk preference after the secondary adjustment, eligible financial products are screened from the product library to form a candidate product set, and a collaborative filtering algorithm is used to sort and select the candidate products in combination with the customer's historical transaction data and similar customer preferences to generate a recommended financial product list, including: Obtain a list of financial products in the product database. If the credit limit and risk appetite of the product meet the entry threshold, store the product in the candidate product set. Based on the historical transaction data in the customer transaction record database, the customer activity level is calculated based on the transaction frequency and transaction amount, and the customer preference characteristics are clustered to obtain similar user groups; Calculate product purchase weights based on historical transaction records of the similar user group, and perform collaborative filtering on the candidate product set to obtain a product recommendation index; For each product in the candidate product set, the product recommendation index and the customer activity level are used to calculate a final recommendation coefficient, and the candidate product set is sorted and screened by the final recommendation coefficient to obtain a product recommendation list.

6. The method according to claim 1, characterized in that The recommendation list is optimized according to the product matching evaluation results, products with low matching are removed from the list, and products with high matching are retained. According to customer feedback and product recommendation effects, the product recommendation method is dynamically updated to obtain an optimized product recommendation list, including: Obtaining matching evaluation data in the product recommendation list, and screening and grading the recommended products according to the matching evaluation data and the product matching threshold to obtain a screening product set; For the screened product set, a customer feedback index and a purchase intention index are used to calculate a product recommendation score, and the screened product set is re-sorted according to the product recommendation score to obtain a sorted product set; If the product recommendation score is lower than the product recommendation score threshold, the product acceptance degree is calculated, and the ranked product set is optimized according to the product acceptance degree to obtain an optimized product set; For the optimized product set, the recommendation accuracy is calculated based on the product recommendation records and selection preference data, the product conversion rate is calculated using the recommendation accuracy and the product acceptance level, and the optimized product set is screened by the product conversion rate to obtain an optimized product recommendation list.

7. The method according to claim 1, characterized in that The optimized product recommendation list is presented to the customer, the list includes product details, risk warnings and purchase channels, and the customer's feedback and purchase behavior on the recommended products are recorded to continuously improve the performance of the intelligent recommendation system, including: According to the product feature information and product rating classification in the product information database, the risk warning content is graded and labeled to obtain product information with risk levels; The product information with risk levels and the user interaction habit data are used to personalize the product display layout to obtain a product display page; If the user triggers the interaction of the product display page, the user browsing trajectory is recorded, the user behavior characteristics are extracted from the user browsing trajectory, and the user attention index is calculated according to the user behavior characteristics; The content feedback score is calculated according to the user attention index and the page click record, and the display layout is optimized and sorted according to the content feedback score and the channel conversion rate to obtain the final display plan.

Images