TWI848850B - Customized financial product recommendation method and system - Google Patents

Abstract

A customized financial product recommendation system includes a processing unit that performs the following operations: generating explicit preferences of the M customers for N different fund types based on first explicit preference data and second explicit preference data of customer data about M customers; A sparse matrix; a matrix composed of the sparse matrix and the predicted values of the plurality of missing elements obtained by the singular value decomposition algorithm; The N values in each column of the matrix are filled, and the N fund types are sorted to obtain a fund type sorting result for each customer; the fund type sorting result for each customer is adjusted so that the currently popular fund types have the highest priority; and the financial product recommendation result for each customer is generated based on the top P fund types extracted from the adjusted fund type sorting result.

Description

Customized financial product recommendation method and system

The present invention relates to financial product recommendation, and more particularly to a customized financial product recommendation method and system.

Most existing financial product recommendation systems use simple statistical analysis results, such as indicating the products that are purchased by the most people, the products with the highest return rate, etc., as the criteria for financial product recommendation. The above recommendation method not only cannot satisfy customers with different product preferences at the same time, but also cannot effectively help customers screen suitable financial products.

Therefore, how to invent a customized financial product recommendation method has become one of the issues that the relevant technical fields want to solve.

Therefore, the purpose of the present invention is to provide a customized financial product recommendation method and system, which can overcome at least one shortcoming of the prior art.

Therefore, the present invention provides a customized financial product recommendation method for recommending N( ) different types of financial products related to funds are implemented using a computer system. The customized financial product recommendation method comprises the following steps: (A) obtaining M ( ) customer data corresponding to M customers, each customer data including fund transaction record data of one of the M customers in a first recent historical period, and digital footprint data of the corresponding customer browsing a financial product network platform related to the N types of funds in a second recent historical period; (B) using the existing RFM model to analyze all fund transaction record data contained in the M customer data and perform standardization on the analysis results to obtain information about the number of days from the last purchase to the current day, the total number of purchases and the total purchase amount of the M customers. (C) analyzing all digital footprint data contained in the M customer data and performing standardization on the analysis results to obtain second explicit preference data on browsing time of the M customers, the second explicit preference data comprising a second preference value corresponding to each fund type that each customer has browsed; (D) generating, based on the first explicit preference data and the second explicit preference data, a matrix representing the explicit preferences of the M customers for the N fund types. Sparse matrix; (E) Analyze the All preference values contained in the sparse matrix are obtained to obtain multiple values corresponding to the The predicted values of multiple missing elements in the sparse matrix; (F) according to the The sparse matrix and the predicted values constitute a (G) adjusting the fund type ranking result of each customer according to a currently popular fund type so that the currently popular fund type has the highest priority; and (H) extracting the top P( ) fund types, and generates a financial product recommendation result corresponding to the customer, which indicates P financial products that belong to the first P fund types and have a higher investment return rate.

In some embodiments, after step (H), the following steps are further included: (I) when a client terminal used by a target customer among the M customers is connected to the computer system, the financial product recommendation result corresponding to the target customer is transmitted to the client terminal for reference by the target customer.

In some embodiments, in step (G), the current popular fund types are generated by analyzing economic news information in a third recent historical period using a known large language model.

In some embodiments, in step (A), for each customer data, the fund transaction record data includes the fund type, transaction date and transaction amount of each transaction, and the digital footprint data includes the total browsing time of each browsing and the browsing time of each browsed fund type; in step (B), the first preference value corresponding to each fund type purchased by each customer is the product of the most recent purchase of the fund type. a first standardized value corresponding to the number of days from the purchase date of the product to the current day, and a second standardized value and a third standardized value corresponding to the total number and total amount of products of the fund type purchased, respectively; and in step (C), the second preference value corresponding to each fund type browsed by each customer is a standardized value corresponding to the sum of the total browsing time of the fund type and the total browsing time unrelated to any fund type.

Therefore, the customized financial product recommendation system provided by the present invention is suitable for providing a system with information about N( ) financial institutions that have a financial product network platform for N different fund types, used to recommend financial products related to the N different fund types, and including a receiving module and a processing unit.

The receiving module is configured to receive M( ) pieces of customer data corresponding to M customers of the financial institution. Each piece of customer data includes fund transaction record data of one of the M customers during a first recent historical period, and digital footprint data of the corresponding customer browsing the financial product online platform during a second recent historical period.

The processing unit is electrically connected to the receiving module to receive the M pieces of customer data, and includes a first analysis module, a second analysis module, a sparse matrix generation module, a third analysis module, a sorting module, an adjustment module and a recommendation module.

The first analysis module is configured to analyze all fund transaction record data contained in the M customer data using an existing RFM model and perform standardization on the analysis results to obtain first explicit preference data on the number of days from the last purchase to the current day, the total number of purchases, and the total purchase amount of the M customers. The first explicit preference data includes a first preference value corresponding to each fund type that each customer has purchased.

The second analysis module is configured to analyze all digital footprint data contained in the M customer data and perform standardization on the analysis results to obtain second explicit preference data about the browsing time of the M customers. The second explicit preference data includes a second preference value corresponding to each fund type that each customer has browsed.

The sparse matrix generation module is configured to generate a matrix representing the explicit preferences of the M customers for the N fund types according to the first explicit preference data and the second explicit preference data. Sparse matrix.

The third analysis module is configured to analyze the All preference values contained in the sparse matrix are obtained to obtain multiple values corresponding to the Predicted values for multiple missing elements in a sparse matrix.

The sorting module is configured to The sparse matrix and the predicted values constitute a The N fund types are sorted according to the size of the N values in each column of the complement matrix to obtain a fund type sorting result corresponding to each customer, wherein the fund type with a larger value has a higher priority.

The adjustment module is configured to adjust the fund type ranking result of each client according to a currently popular fund type so that the currently popular fund type has the highest priority.

The recommendation module is configured to extract the top P ( ) fund types, and generates a financial product recommendation result corresponding to the customer, which indicates P financial products that belong to the first P fund types and have a higher investment return rate.

In some embodiments, the customized financial product recommendation system further includes a communication module electrically connected to the processing unit. When a client terminal used by a target customer among the M customers is connected to the communication module, the processing unit transmits the financial product recommendation result corresponding to the target customer to the client terminal for reference by the target customer.

In some embodiments, the current popular fund types are generated by analyzing economic news information over a third recent historical period using a known large language model.

In some embodiments, for each customer data, the fund transaction record data includes the fund type, transaction date and transaction amount of each transaction, and the digital footprint data includes the total browsing time of each browsing and the browsing time of each browsed fund type; the first preference value corresponding to each fund type purchased by each customer is the purchase of the product of the fund type purchased most recently. a first standardized value corresponding to the number of days from the current day to the previous day, and the sum of a second standardized value and a third standardized value corresponding to the total number and total amount of products of the fund type purchased respectively; and a second preference value corresponding to each fund type that each customer has browsed is a standardized value corresponding to the sum of the total browsing time of the fund type and the total browsing time unrelated to any fund type.

The utility of the present invention is to analyze the explicit preferences of each customer for some of the N types of funds based on the fund transaction record data and digital footprint data of each customer, and to analyze the explicit preferences of the M customers using the singular value decomposition algorithm. The sparse matrix is used to obtain the predicted value of each customer's implicit preference for other fund types among the N fund types, and the fund type ranking result corresponding to each customer is adjusted according to the current popular fund type, thereby obtaining the financial product recommendation result corresponding to each customer. This recommendation method not only satisfies the customer's known preference for some fund types of financial products, but also helps the customer analyze the potential preference for other fund types of financial products. At the same time, it also considers the current important market information, thereby achieving customized fund product recommendations and effectively improving the efficiency of customer fund screening.

Before the present invention is described in detail, it should be noted that similar elements are represented by the same reference numerals in the following description.

Referring to FIG. 1 , a customized financial product recommendation system 1 according to an embodiment of the present invention is suitable for providing a system having information about N( ) different fund types of financial product network platforms (not shown), and used to recommend financial products related to these N different fund types. For example, the financial products provided by the financial institution through the financial product network platform include, for example, financial products related to fund type code 513899999, such as "First Financial Global Water, Electricity, Gas and Infrastructure Income Fund", "First Financial Global AI Artificial Intelligence Fund" and "First Financial Global Trend Fund", and financial products related to fund type code 523088899, such as "Nomura Quality Fund" and "Nomura Growth Fund". In this embodiment, the customized financial product recommendation system 1 can be implemented as a computer system, and for example includes a receiving module 11, a processing unit 12 electrically connected to the receiving module 11, and a communication module 13 electrically connected to the processing unit 12.

In this embodiment, the receiving module 11 is configured to receive, for example, M( ) pieces of customer data corresponding to M customers of the financial institution. Each piece of customer data includes fund transaction record data of one of the M customers in a first recent historical period (for example, but not limited to the last three years), and digital footprint data of the corresponding customer browsing the financial product online platform in a second recent historical period (for example, but not limited to the last six months).

In this embodiment, the communication module 13 is connected to a communication network 100 .

The processing unit 12 includes a first analysis module 121, a second analysis module 122, a sparse matrix generation module 123, a third analysis module 124, a sorting module 125, an adjustment module 126 and a recommendation module 127, the operation of each of which will be described in detail below.

1 and 2 are referred to to illustrate how the processing unit 12 executes a customized financial product recommendation program. The customized financial product recommendation program includes the following steps S21 to S29.

In step S21 , the processing unit 12 receives the M pieces of customer data via the receiving module 11 .

In step S22, the first analysis module 121 uses the existing RFM (Recency Frequency Monetary) model to analyze all fund transaction record data contained in the M customer data and standardize the analysis results to obtain the first explicit preference data about the number of days from the last purchase to the current day, the total number of purchases, and the total purchase amount of the M customers. The first explicit preference data includes a first preference value corresponding to each fund type that each customer has purchased. In this embodiment, the first preference value corresponding to each fund type that each customer has purchased is a first standardized value corresponding to the number of days from the last purchase date of the product of that fund type to the current day, and the sum of a second standardized value and a third standardized value corresponding to the total number and total amount of purchases of products of that fund type, respectively.

Specifically, the first analysis module 121 first analyzes all transaction record data contained in the M customer data to obtain a transaction record analysis result. The transaction record analysis result, for each customer, may include, for example, information such as the number of days from the last purchase date corresponding to each fund type purchased by the customer to the current day, the total number of purchases, and the total purchase amount. For example, if the M customers include three customers (i.e., M=3, and are represented by #1, #2, and #3 respectively), and the N fund types include two different fund types (i.e., N=2, and are represented by A and B respectively), the transaction record analysis result may be, for example, as shown in the following Table 1: Table 1 customer Fund Type The number of days between the last purchase date and the current day Total purchases Total purchase amount/yuan #1 A 30 5 200,000 #2 A 90 10 1,000,000 #2 B 30 5 2,000,000 #3 B 15 1 500,000 Then, the first analysis module 121 performs a standardization process on each value in the transaction record analysis result to obtain a standardized result. It is worth noting that in this embodiment, the standardization process is performed according to the following formula (1) to obtain each standardized value in the standardized result: ······(1) For example, the standardized results obtained according to Table 1 can be shown in the following Table 2: Table 2 customer Fund Type The number of days between the last purchase date and the current day Total purchases Total purchase amount First preference value #1 A 0.2 0.4 0.0 0.6 #2 A 1.0 1.0 0.4 2.4 #2 B 0.2 0.4 1.0 1.6 #3 B 0.0 0.0 0.16 0.16 Afterwards, the first analysis module 121 adds up all the standardized values corresponding to each fund type purchased by each customer in the standardized result as the first preference value corresponding to the fund type purchased by the customer, thereby obtaining the first explicit preference data. For example, the first explicit preference data obtained according to Table 2 can be shown in the following Table 3: Table 3 customer First preference value Fund Type A Fund Type B Fund Type C #1 0.6 - - #2 2.4 1.6 - #3 - 0.16 -

Then, in step S23, the second analysis module 122 analyzes all digital footprint data contained in the M customer data and performs standardization processing on the analysis results to obtain second explicit preference data about the browsing time of the M customers. The second explicit preference data includes a second preference value corresponding to each fund type that each customer has browsed. In this embodiment, the second preference value corresponding to each fund type that each customer has browsed is a standardized value corresponding to the sum of the total browsing time of the fund type and the total browsing time of no fund type.

Specifically, the second analysis module 122 first analyzes all digital footprint data contained in the M customer data to obtain a digital footprint analysis result. The digital footprint analysis result, for each customer, may include, for example, the total browsing time information corresponding to each fund type that the customer has browsed and the total browsing time information not related to any fund type. For example, in the above example, in which customers #1, #2, #3 and the N fund types include, for example, three different fund types (i.e., N=3, and represented by A, B, and C respectively), the digital footprint analysis result may be, for example, as shown in the following Table 4: Table 4 customer Fund Type A Total View Time Fund Type B Total View Time Fund Type C Total View Time Total browsing time for no fund type #1 5 25 0 10 #2 30 30 0 0 #3 80 10 0 0 Then, the second analysis module 122 obtains a summed time information result based on the digital footprint analysis result. The summed time information result includes the summed browsing time information corresponding to each fund type that each customer has browsed. In this embodiment, the summed browsing time information corresponding to each fund type that each customer has browsed is the sum of the total browsing time information of the fund type corresponding to the customer and the total browsing time information of the customer that is not related to any fund type. For example, the summed time information result obtained according to Table 4 can be shown in the following Table 5: Table 5 customer Fund Type A Total browsing time Fund Type B Total browsing time Fund Type C Total browsing time #1 15(5+10) 35(25+10) 10(0+10) #2 30(30+0) 30(30+0) 0(0+0) #3 80(80+0) 10(10+0) 0(0+0) Finally, the second analysis module 122 performs standardization processing on each value in the summed time information result to obtain a standardized value corresponding to each summed browsing time information, and uses each standardized value as a second preference value to obtain the second explicit preference data. In this embodiment, the standardization processing is performed according to formula (1) to obtain each standardized value. For example, the second explicit preference data obtained according to Table 5 can be shown in the following Table 6: Table 6 customer Second preference value Fund Type A Fund Type B Fund Type C #1 0.0 1.0 1.0 #2 0.23 0.8 - #3 1.0 0.0 -

In step S24, the sparse matrix generation module 123 generates a matrix representing the explicit preferences of the M customers for the N fund types according to the first explicit preference data and the second explicit preference data. Sparse matrix. Sparse matrices contain In this embodiment, the preference value corresponding to each fund type of each customer is obtained by adding the first preference value and the second preference value corresponding to the fund type of the customer, and if one of the first preference value and the second preference value is a missing value, the missing value is regarded as "0" for calculation. For example, according to the first explicit preference data in Table 3 and the second explicit preference data in Table 6, the following table 7 can be obtained: Sparse matrix (M=3, N=3). Table 7 customer Preference value Fund Type A Fund Type B Fund Type C #1 0.6(0.6+0.0) 1.0(0+1.0) 1.0(0+1.0) #2 2.63(2.4+0.23) 2.4(1.6+0.8) 0(0+0) #3 1.0(0+1.0) 0.16(0.16+0.0) 0(0+0)

In step S25, the third analysis module 124 uses a singular value decomposition (SVD) algorithm to analyze the All preference values contained in the sparse matrix are obtained to obtain multiple values corresponding to the More specifically, the third analysis module 124 uses the SVD algorithm to decompose the The sparse matrix is constructed to obtain a low-dimensional matrix corresponding to the M customers and a low-dimensional matrix corresponding to the N fund types, and the predicted values of the missing elements are obtained by multiplying the low-dimensional matrices, and the predicted values are filled in to obtain a For example, the value matrix in Table 7 After filling the missing values in the sparse matrix with the predicted values, we get the following table 8: Table 8 customer Preference value Fund Type A Fund Type B Fund Type C #1 0.6 1.0 1.0 #2 2.63 2.4 0.5 (SVD prediction) #3 1.0 0.16 0.8 (SVD prediction)

In step S26, the sorting module 125 sorts the The N values in each column of the complement matrix are used to sort the N fund types to obtain the fund type sorting result corresponding to each customer, where the fund type with a larger value has a higher priority. Taking customer #3 in Table 8 as an example, the fund type sorting result corresponding to customer #3 is shown in Table 9. Table 9 Fund type ranking results for client #3 Sequence Fund Type Preference value 1 A 1.0 2 C 0.8 3 B 0.16

In step S27, the adjustment module 126 adjusts the fund type ranking result of each customer according to a currently popular fund type so that the currently popular fund type has the highest priority. In this embodiment, the currently popular fund type is generated by analyzing the daily economic news information of the past week using a known large language model. For example, assuming that the large language model predicts that fund type C is the currently popular fund type model, the fund type ranking result of customer #3 will be adjusted as shown in Table 10 below. Table 10 Fund type ranking results for client #3 Sequence Fund Type Preference value 1 C 0.8 2 A 1.0 3 B 0.16

In step S28, the recommendation module 127 extracts the top P ( ) fund types, and generates a financial product recommendation result corresponding to the customer. The financial product recommendation result indicates P financial products that belong to the first P fund types and have a higher investment return rate. Continuing with the example of Table 10, assuming that P=2, the recommendation module 127 will select financial products with a higher investment return rate from all financial products of fund type C (for example, financial product c1 and financial product c2) and all financial products of fund type A (for example, financial product a1, financial product a2 and financial product a3). If the financial product with a higher investment return rate in fund type C is financial product c2, and the financial product with a higher investment return rate in fund type A is financial product a1, then the financial product recommendation results for customer #3 are shown in Table 11 below. Table 11 Customer #3's financial product recommendation results Sequence Financial Products 1 c2 2 a1

In step S29, when a client terminal 200 used by a target customer among the M customers is connected to the communication module 13 via a communication network 100, the processing unit 12 transmits the financial product recommendation result corresponding to the target customer to the client terminal 200 for reference by the target customer.

In summary, the explicit preferences of each customer for some of the N fund types are analyzed based on the fund transaction record data and digital footprint data of each customer, and the singular value decomposition algorithm is used to analyze the explicit preferences of the M customers. The sparse matrix is used to obtain the predicted value of each customer's implicit preference for the remaining fund types among the N fund types, and the fund type ranking result corresponding to each customer is adjusted according to the current popular fund type, thereby obtaining the financial product recommendation result corresponding to each customer. This recommendation method not only satisfies the customer's known preference for financial products, but also helps the customer analyze the potential preference for other financial products. At the same time, it also fits the current market situation and achieves the effect of more effectively helping customers screen suitable financial products. Therefore, the purpose of this invention can indeed be achieved.

However, the above is only an example of the implementation of the present invention, and it cannot be used to limit the scope of the implementation of the present invention. All simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the content of the patent specification are still within the scope of the patent of the present invention.

1: Customized financial product recommendation system 11: Receiving module 12: Processing unit 121: First analysis module 122: Second analysis module 123: Sparse matrix generation module 124: Third analysis module 125: Sorting module 126: Adjustment module 127: Recommendation module 13: Communication module 100: Communication network 200: Client S21~S29: Steps

Other features and functions of the present invention will be clearly presented in the implementation method with reference to the drawings, wherein: FIG. 1 is a block diagram, which exemplarily illustrates a customized financial product recommendation system of an embodiment of the present invention, and a client used in conjunction therewith; and FIG. 2 is a flow chart, which exemplarily illustrates how a processing unit of the embodiment executes a customized financial product recommendation program.

1: Customized financial product recommendation system

11: Receiving module

12: Processing unit

121: First analysis module

122: Second analysis module

123: Sparse matrix generation module

124: The third analysis module

125: Sorting module

126: Adjustment module

127: Recommended modules

13: Communication module

100: Communication network

200: Client

Claims (8)

A customized financial product recommendation method for recommending and N( ) different types of financial products related to funds, using a computer system to execute, the customized financial product recommendation method comprises the following steps: (A) obtaining M( ) customer data corresponding to M customers respectively, each customer data including fund transaction record data of one of the M customers in a first recent historical period, and digital footprint data of the corresponding customer browsing a financial product network platform related to the N types of funds in a second recent historical period; (B) using an existing RFM model to analyze all fund transaction record data contained in the M customer data and perform standardization on the analysis results to obtain first explicit preference data on the number of days from the last purchase to the current day, the total number of purchases and the total purchase amount of the M customers, the first explicit preference data including a first preference value corresponding to each fund type purchased by each customer; (C) analyzing all digital footprint data contained in the M customer data and performing standardization on the analysis results to obtain second explicit preference data on the browsing time of the M customers, wherein the second explicit preference data includes a second preference value corresponding to each fund type that each customer has browsed; (D) generating a data representing the explicit preferences of the M customers for the N fund types based on the first explicit preference data and the second explicit preference data; (E) Analyze the sparse matrix using the singular value decomposition algorithm All preference values contained in the sparse matrix are obtained to obtain multiple values corresponding to the The predicted values of multiple missing elements in the sparse matrix; (F) according to the The sparse matrix and the predicted values constitute a (G) adjusting the fund type ranking result of each customer according to a currently popular fund type so that the currently popular fund type has the highest priority; and (H) extracting the top P( ) fund types, and generates a financial product recommendation result corresponding to the customer, which indicates P financial products that belong to the first P fund types and have a higher investment return rate. The customized financial product recommendation method as described in claim 1, after step (H), further comprises the following steps: (I) When a client terminal used by one of the M target customers is connected to the computer system, the financial product recommendation result corresponding to the target customer is transmitted to the client terminal for reference by the target customer. In the customized financial product recommendation method as described in claim 1, in step (G), the current popular fund type is generated by analyzing economic news information in a third recent historical period using a known large language model. A customized financial product recommendation method as described in claim 1, wherein: In step (A), for each customer data, the fund transaction record data includes the fund type, transaction date and transaction amount of each transaction, and the digital footprint data includes the total browsing time of each browsing and the browsing time of each browsed fund type; In step (B), the first preference value corresponding to each fund type purchased by each customer is a first standardized value corresponding to the number of days from the last purchase date of the product of the fund type to the current day, and the sum of a second standardized value and a third standardized value corresponding to the total number and total amount of the products of the fund type purchased respectively; and In step (C), the second preference value corresponding to each fund type that each customer has browsed is a normalized value corresponding to the sum of the total browsing time of the fund type and the total browsing time irrelevant to any fund type. A customized financial product recommendation system is suitable for a system that provides a ) different fund types of financial products network platform for financial institutions, for recommending financial products related to the N different fund types, and comprising: a receiving module, configured to receive M ( ) pieces of customer data corresponding to M customers of the financial institution, each piece of customer data including fund transaction record data of one of the M customers in a first recent historical period, and digital footprint data of the corresponding customer browsing the financial product network platform in a second recent historical period; and a processing unit electrically connected to the receiving module to receive the M pieces of customer data, and comprising: a first analysis module configured to analyze all fund transaction record data contained in the M customer data using an existing RFM model and perform standardization processing on the analysis results to obtain first explicit preference data on the number of days from the last purchase to the current day, the total number of purchases and the total purchase amount of the M customers, wherein the first explicit preference data includes a first preference value corresponding to each fund type that each customer has purchased; a second analysis module configured to analyze all digital footprint data contained in the M customer data and perform standardization processing on the analysis results to obtain second explicit preference data on the browsing time of the M customers, wherein the second explicit preference data includes a second preference value corresponding to each fund type that each customer has browsed; a sparse matrix generation module configured to generate a matrix representing the explicit preferences of the M customers for the N fund types according to the first explicit preference data and the second explicit preference data; A sparse matrix; a third analysis module, configured to analyze the All preference values contained in the sparse matrix are obtained to obtain multiple values corresponding to the The predicted values of multiple missing elements in the sparse matrix; a sorting module, assembled according to the The sparse matrix and the predicted values constitute a The N fund types are sorted according to the size of the N values in each column of the fill-in matrix to obtain a fund type sorting result corresponding to each customer, wherein the fund type with a larger value has a higher priority; an adjustment module is configured to adjust the fund type sorting result of each customer according to a current popular fund type so that the current popular fund type has the highest priority; and a recommendation module is configured to extract the top P ( ) fund types, and generates a financial product recommendation result corresponding to the customer, which indicates P financial products that belong to the first P fund types and have a higher investment return rate. The customized financial product recommendation system as described in claim 5 further comprises: A communication module electrically connected to the processing unit; Wherein, when a client terminal used by one of the M target customers is connected to the communication module, the processing unit transmits the financial product recommendation result corresponding to the target customer to the client terminal for reference by the target customer. A customized financial product recommendation system as described in claim 5, wherein the current popular fund type is generated by analyzing economic news information in a third recent historical period using a known large language model. A customized financial product recommendation system as described in claim 5, wherein: For each customer data, the fund transaction record data includes the fund type, transaction date and transaction amount of each transaction, and the digital footprint data includes the total browsing time of each browsing and the browsing time of each browsed fund type; The first preference value corresponding to each fund type purchased by each customer is a first standardized value corresponding to the number of days from the last purchase date of the product of the fund type to the current day, and the sum of a second standardized value and a third standardized value corresponding to the total number and total amount of the products of the fund type purchased respectively; and The second preference value corresponding to each fund type that each customer has browsed is a normalized value corresponding to the sum of the total browsing time for that fund type and the total browsing time for any fund type.