Financial product recommendation personalization method and system based on data mining and medium

Abstract

The invention discloses a financial product recommendation personalization method and system based on data mining and a medium. The method comprises the steps of determining a user portrait based on user information, determining a corresponding willingness value based on the user portrait, and determining a user type based on the willingness value and a historical purchase record; the financial products are divided into multiple product types, the corresponding product types are matched based on the user types, and the financial products under the matched product types serve as alternative products; determining a first interest value for the alternative product based on the user portrait, and correcting the first interest value based on the user information to obtain a second interest value; selecting a plurality of alternative products in the front based on the second interest value to construct a first recommendation list; and rejecting mutually exclusive products in the first recommendation list to obtain a second recommendation list, and selecting a plurality of financial products in the second recommendation list for recommendation. According to the method, the users can be accurately classified and matched with the corresponding product types, so that the pertinence and accuracy of recommendation are improved.

Description

Technical Field

[0001] This application relates to the technical field of data processing, and particularly to a personalized method, system and medium for financial product recommendation based on data mining. Background Art

[0002] With the popularization of Internet technology and the wide application of big data, financial institutions have accumulated a vast amount of customer data, including transaction records, investment preferences, risk tolerance, etc. These data provide rich materials for personalized recommendation of financial products based on data mining. For example, the recommendation method based on deep reinforcement learning can continuously learn and optimize to recommend the most suitable financial plan for customers.

[0003] For example, the Chinese patent document with the publication number CN115760433A discloses a financial product recommendation method and device. This method obtains the behavior data of users on the mobile banking, including the evaluation information and click data of financial products; then, inputs the evaluation information into the first recommendation model (based on the least squares method) to obtain the first recommendation result, and inputs the click data into the second recommendation model (also based on the least squares method) to obtain the second recommendation result; then, cross - mixes and de - duplicates the two recommendation results to obtain the final third recommendation result, and recommends financial products to users accordingly. Another example is the Chinese patent document with the publication number CN110415123A, which discloses a financial product recommendation method, device, equipment and computer storage medium. This method constructs product recommendation features based on the historical data of the set parameters of each financial product, thereby obtaining the comprehensive product recommendation features of all financial products, and then determines the user recommendation ratio of each financial product according to the deviation degree of the product recommendation feature of each financial product relative to the comprehensive product recommendation feature of the corresponding category, and finally recommends financial products to users based on the user recommendation ratio of each financial product. Higher user recommendation ratios can be assigned to better financial products, so that better financial products can be seen by more users.

[0004] The first technical solution above mainly relies on the evaluation information and click data of users on financial products, but insufficiently mines the deeper preferences and needs of users, and does not further optimize and screen the recommendation results. The second technical solution above does not deeply consider the differences between users and may not accurately match the real needs of users. Summary of the Invention

[0005] To solve the problems raised in the above background art, this application provides a personalized method, system and medium for financial product recommendation based on data mining.

[0006] To achieve the above - mentioned invention purpose, the present invention provides a personalized method for financial product recommendation based on data mining, including:

[0007] Construct multiple user portraits based on historical data, and count the willingness values of each of the user portraits for purchase;

[0008] Obtain the user information of the current user, determine the user portrait based on the user information, determine the corresponding willingness value based on the user portrait, and determine the user type based on the willingness value and the historical purchase record of the current user;

[0009] Divide financial products into multiple product types, match the corresponding product types based on the user type, and use the financial products under the matched product types as alternative products;

[0010] Determine the first interest value of the user in the alternative products based on the user portrait, and correct the first interest value based on the user information to obtain the second interest value;

[0011] Sort the alternative products in descending order based on the second interest value, and select multiple of the top-ranked alternative products to construct a first recommendation list;

[0012] Eliminate mutually exclusive products in the first recommendation list to obtain a second recommendation list, and select multiple of the financial products in the second recommendation list for recommendation

[0013] Furthermore, counting the willingness values of each of the user portraits for purchase includes the following steps:

[0014] The historical data includes the user information of multiple users. The user information includes personal information, asset information, and the historical purchase information. Divide the product types into multiple property types, analyze the asset information to obtain the asset flow information of the user, perform clustering with the personal information and the asset flow information as features to obtain multiple clustering results, predict the actual assets of the user based on the asset flow information, and perform hierarchical division on each of the clustering results based on the actual assets of the user to obtain multiple user portraits;

[0015] Aggregate the historical purchase information corresponding to the user portrait into basic information, analyze the basic information to obtain the attraction value of each financial product to the user portrait and the inclination value of the user portrait to each financial product, and comprehensively calculate the willingness value of the user portrait for purchase based on the attraction value and the inclination value of each financial product.

[0016] Furthermore, obtaining the second interest value includes the following steps:

[0017] Obtain the number of purchases of each of the financial products by the user portrait based on the historical purchase information, determine the number of active users based on the historical data, and correct the first interest value based on the number of active users and the number of purchases of each of the financial products to obtain the second interest value of the user for each of the financial products.

[0018] Further, eliminating the mutually exclusive products in the first recommendation list includes the following steps:

[0019] Obtain the number of times each user holds various financial products simultaneously based on the historical purchase information, and calculate the holding preference value of the user portrait for different financial products based on the number of simultaneous holdings;

[0020] Take the financial products already purchased by the current user as the purchased products, re - sort the alternative products in the first recommendation list based on the holding preference value for the purchased products to obtain a transition list, locate multiple alternative products with higher rankings in the transition list as the first reserved products, calculate the indirect preference value between the purchased products and the remaining alternative products in the transition list, screen out the alternative products with the indirect preference value greater than the first threshold and define them as the second reserved products, and define the alternative products other than the first reserved products and the second reserved products in the first recommendation list as the mutually exclusive products.

[0021] Further, perform clustering based on the K - means algorithm or the DBSCAN algorithm.

[0022] Further, set a replenishment rule, the replenishment rule includes a high - yield rule and a diversification rule, select the corresponding replenishment rule based on the user type, after eliminating the mutually exclusive products in the first recommendation list, replenish the financial products corresponding to the elimination quantity in the second recommendation list based on the replenishment rule to improve the second recommendation list.

[0023] Further, the user types include a demand type, a stable type, and an aggressive type, and the product types include a credit type, a stable type, and a high - yield type.

[0024] Further, if the current user has never purchased the financial product, determine the current user as the stable type and do not optimize the first recommendation list.

[0025] The present invention also provides a personalized financial product recommendation system based on data mining for implementing the above - mentioned personalized financial product recommendation method based on data mining. The system includes:

[0026] The pre - module constructs multiple user portraits based on historical data and calculates the willingness values for each of the user portraits.

[0027] The matching module obtains the user information of the current user, determines the user portrait based on the user information, determines the corresponding willingness value based on the user portrait, determines the user type based on the willingness value and the historical purchase records of the current user, divides financial products into multiple product types, matches the corresponding product type based on the user type, and uses the financial products under the matched product type as alternative products.

[0028] The generation module determines the first interest value of the user in the alternative products based on the user portrait, corrects the first interest value based on the user information to obtain the second interest value, sorts the alternative products from largest to smallest based on the second interest value, and selects multiple of the top - ranked alternative products to construct the first recommendation list.

[0029] The optimization module eliminates mutually exclusive products from the first recommendation list to obtain the second recommendation list, and selects multiple of the financial products in the second recommendation list for recommendation.

[0030] This application also provides a computer - readable storage medium with instructions stored thereon. When the instructions are executed by a processor, they implement a personalized method for financial product recommendation based on data mining as described above.

[0031] Advantageous effects:

[0032] The present invention first constructs user portraits and divides user types, thereby accurately identifying the characteristics and preferences of different users and providing highly personalized financial product recommendations for users. By calculating willingness values and interest values, the recommendation results are further optimized. In addition, the present invention also eliminates mutually exclusive products, avoiding the user's selection confusion between incompatible products and further improving the accuracy and practicality of the recommendation. Through the recommendation method of the present invention, not only the acceptance and purchase willingness of users for financial products are improved, but also a more efficient means of accurate customer marketing is provided for financial institutions, enhancing customer satisfaction and loyalty. Description of the Drawings

[0033] Figure 1 It is a flow chart of a personalized method for financial product recommendation based on data mining in this application;

[0034] Figure 2 It is a schematic diagram of the principle of the clustering result in this application;

[0035] Figure 3 It is a schematic diagram of the principle of calculating the indirect preference value in this application;

[0036] Figure 4 This is a schematic structural diagram of a personalized system for financial product recommendation based on data mining in this application. Specific implementation manners

[0037] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0038] It can be understood that the terms "first", "second", etc. used in this application can be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish the first element from another element. For example, without departing from the scope of this application, the first xx script can be called the second xx script, and similarly, the second xx script can be called the first xx script.

[0039] As Figure 1 shown, a personalized method for financial product recommendation based on data mining includes:

[0040] S1: Construct multiple user portraits based on historical data, and count the willingness values purchased for each user portrait. The historical data is the user data collected by the platform, including the personal information and purchase records of the user, etc. Then, the user portraits are constructed by analyzing the historical data. The specific method for constructing the user portraits will be introduced later. Each user portrait corresponds to a fixed willingness value. By constructing the user portraits, the characteristics and preferences of different user groups can be accurately identified.

[0041] S2: Obtain the user information of the current user, determine the user portrait based on the user information, determine the corresponding willingness value based on the user portrait, and determine the user type based on the willingness value and the historical purchase records of the current user.

[0042] The user types include demand type, stable type, and aggressive type, and the product types include credit type, stable type, and high-yield type.

[0043] The user information specifically includes personal information, asset information, and historical purchase information. The personal information includes information such as age, gender, occupation, and personal annual income. The asset information includes the holding information of current deposits, time deposits, and other financial products. The historical purchase information includes the financial products purchased by the user in the past period. In this embodiment, the financial products include stocks, funds, gold, credit, and time deposits, etc. After determining the user portrait based on the user information, the corresponding willingness value can be determined.

[0044] In this embodiment, determining the user type includes the following steps. First, set a first numerical range and a second numerical range, where the second numerical range is greater than the first numerical range. The first numerical range corresponds to the first type, and the second numerical range corresponds to the second type. Users of the first type have a lower purchase willingness, while users of the second type have a higher purchase willingness. Determine whether the corresponding user is of the first type or the second type according to the numerical range where the current user willingness value is located. Then, obtain the user's historical purchase records, and determine the user type according to the purchase preferences for various financial products therein. For example, if the user has purchased credit-type financial products more than 5 times in the historical purchase records, then mark the user type as the demand type. The determination methods for other user types are the same.

[0045] For the demand type, it indicates that the user's assets are relatively scarce and may need credit-type financial products. For the stable type, it indicates that the user has certain assets but a low willingness to purchase financial products. Therefore, stable-type financial products need to be provided. The returns of stable-type products are relatively stable to increase their purchase willingness. For the aggressive type, it indicates that the user has certain assets and a high willingness to purchase financial products. Therefore, high-yield-type financial products need to be provided.

[0046] The following method can be used to determine the user portrait. First, extract key features from the corresponding user information according to the existing user portrait. Then, use similarity calculation methods (such as Euclidean distance, cosine similarity, Jaccard similarity, etc.) to measure the feature similarity between the current user and the key features of the existing user portrait. According to the similarity score, match the current user to the most similar user portrait. The calculation method of the specific matching algorithm is prior art and will not be introduced here.

[0047] S3: Divide financial products into multiple product types, match the corresponding product types based on the user type, and use the financial products under the matched product types as alternative products.

[0048] As described above, the product types of financial products include credit type, stable type, and high-yield type. Each product type includes multiple specific products. For example, the stable type includes time deposits, national debts, etc., and the aggressive type includes stocks and funds, etc. The user types include demand type, stable type, and aggressive type. The demand type is matched with the credit type, the stable type is matched with the stable type, and the aggressive type is matched with the high-yield type. For example, if the user is matched to the high-yield type, then stocks and funds are used as alternative products.

[0049] S4: Determine the first interest value of the user for the alternative products based on the user portrait, and modify the first interest value based on the user information to obtain the second interest value.

[0050] The first interest value of each user portrait is also calculated in advance based on historical data and stored in the server. After determining the user portrait, the first interest value of the user for each candidate product can be determined. Then, the first interest value is modified according to the specific user information of the current user to obtain the second interest value. The larger the second interest value for a financial product, the more likely it is that the user is interested in the financial product. The calculation method of the first interest value and the second interest value will be introduced later.

[0051] S5: Sort the candidate products from large to small based on the second interest value, and select multiple candidate products with high rankings to construct a first recommendation list.

[0052] S6: Eliminate mutually exclusive products in the first recommendation list to obtain a second recommendation list, and select multiple financial products from the second recommendation list for recommendation.

[0053] By selecting some (e.g., 5) candidate products with higher second interest values ​​to form the first recommendation list, it can be easily displayed and convenient for users to click to view. In particular, there may be mutually exclusive products in the first recommendation list. For example, the first recommendation list includes financial product A, and the user already holds financial product B. By analyzing user information, it is determined that financial products A and B are mutually exclusive products. Mutual exclusivity indicates that the user is unlikely to choose to buy financial products A and B at the same time. It is necessary to eliminate one of them and add other financial products for recommendation, so as to further improve the personalized matching degree of financial products.

[0054] The present invention first constructs user portraits and divides user types, so as to accurately identify the characteristics and preferences of different users and provide users with highly personalized financial product recommendations. The recommendation results are further optimized by calculating the willingness value and interest value. In addition, the present invention also avoids the user's choice confusion between incompatible products by eliminating mutually exclusive products, further improving the accuracy and practicality of the recommendation. The recommendation method of the present invention not only improves the user's acceptance and willingness to purchase financial products, but also provides financial institutions with more efficient customer precision marketing methods, thereby improving customer satisfaction and loyalty.

[0055] It is particularly noteworthy that the present invention can accurately classify users and match them with corresponding product types, further improving the pertinence and accuracy of recommendations.

[0056] In this embodiment, counting the purchase intention value of each user profile includes the following steps:

[0057] Historical data includes user information of multiple users. The user information includes personal information, asset information, and historical purchase information. The product types are classified into multiple property types. The asset information is analyzed to obtain the user's asset flow information. Clustering is performed using personal information and asset flow information as features to obtain multiple clustering results. Based on the asset flow information, the user's actual assets are predicted. Based on the user's actual assets, each clustering result is hierarchically divided to obtain multiple user portraits.

[0058] The property types in this embodiment include stocks, funds, gold, time deposits, insurance, and credit. After obtaining the historical data, the asset information of each user is analyzed to obtain the user's asset flow information. The asset information includes the flow of funds, including transfers in and out between different financial products. Based on the asset information, an asset flow sequence for each user is constructed, which is the asset flow information. Then, the personal information and asset flow information are matrixed and clustered based on the K-means algorithm or the DBSCAN algorithm. As Figure 2 shown in the clustering result of the K-means algorithm, users with similar personal information and asset flow information are grouped into one category. The specific clustering method is prior art and will not be introduced here. For example, after clustering, four clustering results are obtained. The first clustering result includes 50 pieces of user information.

[0059] In this embodiment, when determining the user's actual assets, first, the transfer in and out information of the user's funds is extracted from the asset flow information, and a base amount is set. The base amount in this embodiment is 1000. In the asset flow information, transfer-in amounts less than 1000 (salary income is retained) and transfer-out amounts greater than 1000 (large expenditures are excluded) are removed. Then, the transfer-in amounts are accumulated to obtain the total income, the transfer-out amounts are accumulated as the total expenditure, and the difference between the total income and the total expenditure is used as the user's actual assets. In other embodiments, an LSTM neural network model or a decision tree model can also be established, and a more accurate prediction effect can be obtained by inputting the asset flow information into the model.

[0060] As described above, the actual assets of 50 users in the first clustering result can be obtained. Then, the average value and standard deviation of the actual assets of the 50 users are calculated, and the first value and the second value are calculated through the first formula and the second formula. The first formula is Z1 = μ - 2σ, and the second formula is Z2 = μ + 2σ, where Z1 and Z2 are the first value and the second value respectively, and μ and σ are the average value and the standard deviation respectively. Then, the first clustering result is divided into three levels. Among them, the first level is the users with actual assets less than the first value, the second level is the users with actual assets between the first value and the second value, and the third level is the users with actual assets greater than the second value. For example, through the division, it is determined that the first level includes 10 user information, and then the 10 user information is clustered again, and the center of the clustering is selected as the corresponding user portrait.

[0061] By performing hierarchical division on the basis of the clustering result, the construction fineness of the user portrait is further improved.

[0062] Aggregate the historical purchase information corresponding to the user portrait into basic information, analyze the basic information to obtain the attraction value of each financial product to the user portrait, and the preference value of the user portrait for each financial product, and calculate the willingness value of the user portrait to purchase by synthesizing the attraction value and preference value of each financial product.

[0063] As before, if the first level includes 10 user information, then there are 10 corresponding historical purchase information, and the 10 historical purchase information are combined into basic information. And the attraction value and preference value are calculated respectively through the following third formula and fourth formula. The third formula is: X ir =(U ir +α) / (U + α·G), where X ir is the attraction value of financial product i to user portrait r, U ir is the total number of times user portrait r purchases financial product i, specifically the total number of times the 10 users in the basic information purchase financial product i, U is the total number of times all users in the historical data purchase financial product i, α is the first smoothing parameter, G is the number of product types of the financial product, and α and G are used to avoid the zero probability problem.

[0064] The fourth formula for calculating the preference value is: Y r =(V r +β) / (V + β·J), where Y r is the preference value of user portrait r for the financial product, V r is the total number of times user portrait r purchases all financial products, V is the total number of times all users in the historical data purchase financial products, β is the second smoothing parameter, J is the number of active users, and the specific obtaining method will be introduced later. β and J are used to avoid the zero probability problem.

[0065] After that, the willingness value P of the user profile r is calculated through the fifth formula r , and the fifth formula is: I is the number of financial products. In the third formula, the more times each user under the user profile purchases financial product i, the greater the attraction value of financial product i. In the fourth formula, the more times each user under the user profile purchases financial products, the greater the inclination value of the user to purchase financial products. After that, the attraction value of each financial product is corrected by the inclination value of the user's purchase, and the willingness value of the user's purchase is obtained by accumulation. The larger the calculation result, the greater the willingness of the user to purchase financial products.

[0066] The steps for obtaining the second interest value in this embodiment include the following:

[0067] Based on historical purchase information, obtain the purchase times of the user profile for various financial products. Based on historical data, determine the active user volume. Based on the active user volume and the purchase times of various financial products, correct the first interest value to obtain the second interest value of the user for each financial product.

[0068] The first interest value here is calculated based on the sixth formula, and the sixth formula is is the first interest value of the user profile r for the financial product i, and the meanings of other parameters are as introduced above. Through the historical data within the past month, count the number of users who have conducted asset operations, and use this number of users as the active user volume. After that, correct the first interest value through the seventh formula to obtain the second interest value. The specific seventh formula is where H ir is the second interest value of the user profile r for the financial product i, U ir is the total number of times the user profile r purchases the financial product i, J is the active user volume. In the seventh formula, first calculate the sum of the interest values of the user profile r for various financial products, and then obtain the proportion of the first interest value of the user profile r for the financial product i. The larger the proportion, the larger the corresponding second interest value. In addition, correct the result through the total number of times the user purchases and the user activity to avoid the calculation result from being wrongly amplified due to differences in user activity.

[0069] In this embodiment, the steps for eliminating mutually exclusive products in the first recommendation list include the following:

[0070] Based on historical purchase information, obtain the simultaneous holding times of each user for various financial products, and calculate the holding preference value of the user profile for different financial products based on the simultaneous holding times.

[0071] Specifically, the holding preference value is calculated in the following way. For example, for user profile A, which includes 10 user data, the number of simultaneous holdings between different financial products is obtained through historical purchase information. For example, if user E holds fund A and buys fund B, then fund A and fund B are counted as one simultaneous holding number. In particular, if user E sells fund A and fund B at the same time, but subsequently buys fund A and fund B, it will be counted as one simultaneous holding number. Finally, the number of simultaneous holdings of fund A and fund B by all users in the user profile is counted as the first number, the second number of fund A purchases, and the third number of fund B purchases, and the holding preference value L of fund A and fund B is calculated by the eighth formula. The eighth formula is: L = N1 / (N2+N3), where N1, N2 and N3 are the first number, the second number and the third number, respectively.

[0072] The financial products purchased by the current user are regarded as purchased products, and the alternative products in the first recommendation list are re-sorted based on the holding preference values ​​of the purchased products to obtain a transition list, and multiple alternative products ranked at the top in the transition list are located as the first retained products, and the indirect preference values ​​of the purchased products and the remaining alternative products in the transition list are calculated, and the alternative products with indirect preference values ​​greater than the first threshold are screened and defined as the second retained products, and the alternative products other than the first retained product and the second retained product in the first recommendation list are defined as mutually exclusive products.

[0073] If the current user has not purchased any financial product, the current user is determined to be a stable type, and the first recommendation list is not optimized.

[0074] If the current user has not purchased a financial product, the first recommendation list will not be optimized. If the user has purchased a financial product, the purchased financial product will be defined as a purchased product. Then, the holding preference values ​​of each candidate product and the purchased product in the first recommendation list are determined based on the user portrait, and sorted from large to small. For example, the transition list after sorting is candidate products A, B, C, D, and E. The first two candidate products A and B are selected as the first retained products. The holding preference values ​​of the two first retained products and the purchased product are 0.9 and 0.8 respectively. Then, the holding preference values ​​of candidate product A and candidate product CE, as well as the holding preference values ​​of candidate product B and candidate product CE are obtained.

[0075] The indirect preference value is calculated using the following method, such as Figure 3As shown, there are two paths to calculate the indirect preference value between the purchased product and alternative product C (purchased product - alternative product A - alternative product C, purchased product - alternative product B - alternative product C). Each path will obtain a result. If one of the two results is greater than the first threshold, then alternative product C is retained. For example, to calculate the indirect preference value between the purchased product and alternative product C, define the holding preference value between the purchased product and alternative product A as the first preference value, and the holding preference value between alternative product A and alternative product C as the second preference value, with a value of 0.7. Then the indirect preference value between the purchased product and alternative product C is 0.9 * 0.7 = 0.63. If the first threshold is 0.6, then alternative product C is used as the second retained product, indicating that the user may have the willingness to hold the purchased product, alternative product A, and alternative product C simultaneously. If the indirect preference values between the purchased product, alternative product A and alternative product C, and between the purchased product, alternative product B and alternative product C are all less than the first threshold, then alternative product C is not retained, that is, alternative product C is used as the mutually exclusive product.

[0076] In this embodiment, a replenishment rule is also set. The replenishment rule includes a high - return rule and a diversification rule. Based on the user type, the corresponding replenishment rule is selected. After excluding the mutually exclusive products in the first recommendation list, financial products corresponding to the exclusion quantity are replenished in the second recommendation list based on the replenishment rule to improve the second recommendation list.

[0077] Specifically, according to the user type, the replenishment rule is selected. If the user type is a demand type or a high - return type, then the diversification rule is selected. If the user type is a stable type, then the high - return rule is selected. For the second recommendation list provided for users of the demand type, the types of credit products included are more diverse, such as providing different loan amounts and terms. For the second recommendation list provided for users of the aggressive type, the types of high - return products included are more diverse, facilitating their diversified investment. For the second recommendation list provided for users of the stable type, high - return financial products are added to it, facilitating their understanding of relevant content of high - risk investment while making stable investments.

[0078] During actual replenishment, for example, if financial products B and C are deleted from the second recommendation list, and the deletion quantity is 2, then according to the replenishment rule, two more financial products are selected and added to the second recommendation list to improve the second recommendation list.

[0079] As Figure 4 shown, the present invention also provides a personalized financial product recommendation system based on data mining for implementing the above - mentioned personalized financial product recommendation method based on data mining. The system includes:

[0080] A pre - processing module that constructs multiple user portraits based on historical data and statistically calculates the willingness values of each user portrait for purchase;

[0081] A matching module obtains user information of the current user, determines a user profile based on the user information, determines a corresponding willingness value based on the user profile, determines a user type based on the willingness value and the current user's historical purchase record, divides financial products into multiple product types, matches corresponding product types based on user types, and selects financial products under the matching product types as candidate products;

[0082] A generation module, which determines a first interest value of the user for the candidate products based on the user portrait, modifies the first interest value based on the user information, obtains a second interest value, sorts the candidate products from large to small based on the second interest value, and selects a plurality of candidate products with the highest sorting to construct a first recommendation list;

[0083] The optimization module removes mutually exclusive products from the first recommendation list to obtain a second recommendation list, and selects multiple financial products from the second recommendation list for recommendation.

[0084] The present application also provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, a method for personalized financial product recommendation based on data mining as described above is implemented.

[0085] It should be understood that the various technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0086] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A personalized method for recommending financial products based on data mining, characterized in that: Build multiple user profiles based on historical data, and count the purchase intention values ​​of each user profile; Acquire user information of the current user, determine the user portrait based on the user information, determine the corresponding willingness value based on the user portrait, and determine the user type based on the willingness value and the historical purchase record of the current user; Classify financial products into multiple product types, match the corresponding product types based on the user types, and use the financial products matching the product types as candidate products; Determine a first interest value of the user for the candidate product based on the user portrait, and modify the first interest value based on the user information to obtain a second interest value; sorting the candidate products from large to small based on the second interest value, and selecting a plurality of the candidate products with the highest sorting to construct a first recommendation list; The mutually exclusive products in the first recommendation list are eliminated to obtain a second recommendation list, and a plurality of the financial products are selected from the second recommendation list for recommendation.

2. The method according to claim 1, characterized in that Counting the purchase intention value of each user portrait includes the following steps: The historical data includes the user information of multiple users, the user information includes personal information, asset information and the historical purchase information, the product types are divided into multiple property types, the asset information is analyzed to obtain the asset flow information of the user, clustering is performed based on the personal information and the asset flow information to obtain multiple clustering results, the actual assets of the user are predicted based on the asset flow information, and each of the clustering results is hierarchically divided based on the actual assets of the user to obtain multiple user portraits; Aggregate the historical purchase information corresponding to the user portrait as basic information, analyze the basic information to obtain the attractiveness value of each financial product to the user portrait, and the inclination value of the user portrait to each financial product, and calculate the willingness value of the user portrait to purchase by combining the attractiveness value and the inclination value of each financial product.

3. The method according to claim 2, characterized in that Obtaining the second interest value comprises the following steps: The number of purchases of the various financial products by the user profile is obtained based on the historical purchase information, the number of active users is determined based on the historical data, and the first interest value is corrected based on the number of active users and the purchase times of the various financial products to obtain the user's second interest value for each of the financial products.

4. The method according to claim 3, characterized in that Eliminating the mutually exclusive products in the first recommendation list includes the following steps: Based on the historical purchase information, the number of times each user simultaneously holds various financial products is obtained, and based on the number of times each user simultaneously holds, the holding preference value of the user profile for different financial products is calculated; The financial product purchased by the current user is regarded as a purchased product, and the candidate products in the first recommendation list are reordered based on the holding preference value of the purchased product to obtain a transition list, and multiple candidate products ranked at the top in the transition list are located as the first retained products, and the indirect preference value between the purchased product and the remaining candidate products in the transition list is calculated, and the candidate products whose indirect preference value is greater than a first threshold are screened and defined as the second retained products, and the candidate products other than the first retained product and the second retained product in the first recommendation list are defined as the mutually exclusive products.

5. The method according to claim 2, characterized in that: Clustering is performed based on the K-means algorithm or the DBSCAN algorithm.

6. The method according to claim 1, characterized in that Set supplementary rules, which include high-yield rules and diversification rules. Select the corresponding supplementary rules based on the user type. After removing the mutually exclusive products in the first recommendation list, add the financial products corresponding to the removed quantity to the second recommendation list based on the supplementary rules to complete the second recommendation list.

7. The method according to claim 1, characterized in that The user types include demand type, stable type and aggressive type, and the product types include credit type, stable type and high-yield type.

8. The method according to claim 7, characterized in that If the current user has not purchased the financial product, the current user is determined to be the stable type, and the first recommendation list is not optimized.

9. A financial product recommendation personalized system based on data mining, used to implement a financial product recommendation personalized method based on data mining as claimed in any one of claims 1 to 8, characterized in that: The front-end module builds multiple user portraits based on historical data and counts the purchase intention value of each user portrait; A matching module, which obtains user information of the current user, determines the user portrait based on the user information, determines the corresponding willingness value based on the user portrait, determines the user type based on the willingness value and the historical purchase record of the current user, divides financial products into multiple product types, matches the corresponding product type based on the user type, and uses the financial products matching the product type as candidate products; A generating module, determining a first interest value of the user for the candidate products based on the user portrait, modifying the first interest value based on the user information to obtain a second interest value, sorting the candidate products from large to small based on the second interest value, and selecting a plurality of the candidate products with the highest sorting to construct a first recommendation list; The optimization module removes mutually exclusive products in the first recommendation list to obtain a second recommendation list, and selects multiple financial products from the second recommendation list for recommendation.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, a personalized method for recommending financial products based on data mining as described in any one of claims 1 to 8 is implemented.

Images