Potential customer prediction method, device, electronic equipment and program product

Abstract

This application discloses a method, apparatus, electronic device, and program product for predicting potential customers, relating to the field of big data technology. The method includes: classifying customers into a first category and a second category based on transaction information within a first time period; obtaining first transaction information of the first category of customers and second transaction information of the second category of customers from the transaction information; extracting features from the first transaction information to obtain first feature information, and extracting features from the second transaction information to obtain second feature information; predicting potential customers in the first category based on the first feature information, predicting potential customers in the second category based on the second feature information, and pushing product information to the potential customers. This application solves the problem in related technologies where, when using machine learning models to identify potential customers purchasing financial products, the imbalance of sample information corresponding to transaction information of different customer categories affects the accurate delivery of product information.

Description

Technical Field

[0001] This application relates to the field of big data technology, and more specifically, to a method, apparatus, electronic device, and program product for predicting potential customers. Background Technology

[0002] Within a bank's personal customer system, while a segment of high-net-worth clients enjoy exclusive services from wealth management managers, the much larger and more commercially valuable customer base outside of this segment cannot be ignored. This group is often overlooked due to a lack of direct, personalized service, but its sheer number significantly impacts a bank's customer base, business growth, and even market competitiveness. Furthermore, identifying high-potential clients is a frequently used technique by financial institutions in product promotion.

[0003] Currently, the identification of high-potential customers for financial products mainly relies on single-dimensional indicators such as the scale of financial products held by customers, the investment period, and investment behavior. This single-dimensional identification results in low accuracy of potential customer identification, which reduces the accuracy of financial institutions in pushing product information to potential customers.

[0004] There is currently no effective solution to the problem that when using machine learning models to identify potential customers who want to purchase financial products, the imbalance of sample information corresponding to different categories of customer transaction information leads to inaccurate model output results, which in turn affects the accurate delivery of product information. Summary of the Invention

[0005] The main objective of this application is to provide a method, device, electronic device, and program product for predicting potential customers, in order to solve the problem in related technologies where machine learning models are used to identify potential customers who purchase financial products. This problem arises from the imbalance of sample information corresponding to different categories of customer transaction information, leading to inaccurate model output results and affecting the accurate delivery of product information.

[0006] To achieve the above objectives, according to one aspect of this application, a method for predicting potential customers is provided. The method includes: classifying customers based on transaction information within a first time period to obtain a first type of customer and a second type of customer, wherein the first type of customer is a customer who has not traded a target financial product within the first time period, and the second type of customer is a customer who has traded a target financial product within the first time period; obtaining first transaction information of the first type of customer and second transaction information of the second type of customer from the transaction information; performing feature extraction on the first transaction information to obtain first feature information, and performing feature extraction on the second transaction information to obtain second feature information; inputting the first feature information into a first prediction model to predict potential customers in the first type of customer, inputting the second feature information into a second prediction model to predict potential customers in the second type of customer, and pushing product information of the target financial product to the potential customers in the first type of customer and the potential customers in the second type of customer.

[0007] Further, feature extraction is performed on the first transaction information to obtain first feature information, including: statistically analyzing a first type of variable in the first transaction information according to different time dimensions to obtain a first type of data, wherein the first type of variable represents a variable represented by a single numerical value; calculating a second type of variable in the first transaction information using an unsupervised binning algorithm to obtain a second type of data, wherein the second type of variable represents a variable represented by a value within a range; calculating multiple variables in the first transaction information to obtain a third type of data, wherein the multiple variables include: the first type of variable and the second type of variable; and determining the first feature information corresponding to the first transaction information based on the first type of data, the second type of data, and the third type of data.

[0008] Furthermore, before inputting the first feature information into the first prediction model to predict potential customers in the first type of customer, and inputting the second feature information into the second prediction model to predict potential customers in the second type of customer, the method further includes: collecting historical transaction information of historical customers in financial institutions; constructing a first sample set and a second sample set based on the historical transaction information; wherein the historical customers in the first sample set and the second sample set purchased financial products at different times; performing feature extraction on the first sample set to obtain third feature information, and performing feature extraction on the second sample set to obtain fourth feature information; performing data preprocessing on the third feature information to obtain processed third feature information, and performing data preprocessing on the fourth feature information to obtain processed fourth feature information; training a preset model using the processed third feature information to obtain the first prediction model, and training the preset model using the processed fourth feature information to obtain the second prediction model.

[0009] Further, constructing a first sample set and a second sample set based on the historical transaction information includes: determining a first positive sample based on the historical transaction information of customers belonging to the third category, determining a first negative sample based on the historical transaction information of customers belonging to the fourth category, and constructing the first sample set based on the first positive sample and the first negative sample. The third category of customers refers to customers who did not trade the target financial product during the second time period but did so during the third time period; the fourth category of customers refers to customers who did not trade the target financial product during the second time period but did not trade it during the third time period. The third time period is later than the second time period. The number of the first positive samples... The ratio between the number of the first negative sample and the number of the second negative sample is a first preset ratio; a second positive sample is determined based on the historical transaction information of customers belonging to the fifth category, a second negative sample is determined based on the historical transaction information of customers belonging to the sixth category, and a second sample set is constructed based on the second positive sample and the second negative sample, wherein the fifth category of customers are customers who have traded the target financial product in the second time period and in the third time period, and the sixth category of customers are customers who have traded the target financial product in the second time period and in the third time period, and the ratio between the number of the second positive sample and the number of the second negative sample is a second preset ratio.

[0010] Further, training a preset model using the processed third feature information to obtain the first prediction model includes: dividing the processed third feature information into a first training set and a first test set; training N first preset models using the first training set to obtain N third models and their output results; testing the N third models using the first test set to obtain their output results and constructing a second test set based on their output results; constructing a second training set based on their output results and the first training set; training a second preset model using the second training set and optimizing the trained second preset model using the second test set to obtain the first prediction model.

[0011] Further, the third feature information is preprocessed to obtain processed third feature information, including: determining missing information in the third feature information; filling in the missing information in the third feature information according to the data type of the missing information to obtain fifth feature information, wherein the data type of the missing information includes at least one of the following: transaction behavior type, time type, and numerical type; deleting a preset string in the fifth feature information to obtain sixth feature information, wherein the preset string includes at least one of the following: a string corresponding to duplicate information and a string corresponding to semantically unrelated information; and filtering key features in the sixth feature information based on a target indicator value to obtain the processed third feature information, wherein the calculation method of the target indicator value includes at least one of the following: evidence weight, information value, and random forest algorithm.

[0012] Furthermore, the first transaction information of the first type of customer includes at least the following information: the customer's basic information, asset information, liability information, transaction behavior information, and client behavior data.

[0013] To achieve the above objectives, according to another aspect of this application, a potential customer prediction device is provided. The device includes: a segmentation unit, configured to segment customers based on transaction information within a first time period to obtain a first type of customer and a second type of customer, wherein the first type of customer is a customer who has not traded the target financial product within the first time period, and the second type of customer is a customer who has traded the target financial product within the first time period; an acquisition unit, configured to acquire first transaction information of the first type of customer and second transaction information of the second type of customer from the transaction information; a first extraction unit, configured to extract features from the first transaction information to obtain first feature information, and extract features from the second transaction information to obtain second feature information; and a push unit, configured to input the first feature information into a first prediction model to predict potential customers in the first type of customer, input the second feature information into a second prediction model to predict potential customers in the second type of customer, and push product information of the target financial product to the potential customers in the first type of customer and the potential customers in the second type of customer.

[0014] Further, the first extraction unit includes: a statistical subunit, used to statistically analyze a first type of variable in the first transaction information according to different time dimensions to obtain a first type of data, wherein the first type of variable represents a variable represented by a single numerical value; a first calculation subunit, used to calculate a second type of variable in the first transaction information using an unsupervised binning algorithm to obtain a second type of data, wherein the second type of variable represents a variable represented by a value within a range; a second calculation subunit, used to calculate multiple variables in the first transaction information to obtain a third type of data, wherein the multiple variables include: the first type of variable and the second type of variable; and a first determination subunit, used to determine the first feature information corresponding to the first transaction information based on the first type of data, the second type of data, and the third type of data.

[0015] Furthermore, the device further includes: a collection unit, configured to collect historical transaction information of historical customers in financial institutions before inputting the first feature information into a first prediction model to predict potential customers in the first type of customers, and inputting the second feature information into a second prediction model to predict potential customers in the second type of customers; a construction unit, configured to construct a first sample set and a second sample set based on the historical transaction information; wherein the purchase times of the historical customers in the first sample set and the second sample set are different; a second extraction unit, configured to extract features from the first sample set to obtain third feature information, and extract features from the second sample set to obtain fourth feature information; a processing unit, configured to preprocess the third feature information to obtain processed third feature information, and preprocess the fourth feature information to obtain processed fourth feature information; and a training unit, configured to train a preset model using the processed third feature information to obtain the first prediction model, and train the preset model using the processed fourth feature information to obtain the second prediction model.

[0016] Further, the construction unit includes: a second determining subunit, configured to determine a first positive sample based on the historical transaction information belonging to a third type of customer, determine a first negative sample based on the historical transaction information belonging to a fourth type of customer, and construct a first sample set based on the first positive sample and the first negative sample, wherein the third type of customer is a customer who did not trade the target financial product during the second time period but traded the target financial product during the third time period, the fourth type of customer is a customer who did not trade the target financial product during the second time period but did not trade the target financial product during the third time period, the third time period being later than the second time period, and the number of the first positive sample and the number of the first negative sample... The ratio between the quantities is a first preset ratio; the third determining subunit is used to determine a second positive sample based on the historical transaction information of customers belonging to the fifth category, determine a second negative sample based on the historical transaction information of customers belonging to the sixth category, and construct a second sample set based on the second positive sample and the second negative sample, wherein the fifth category of customers are customers who have traded the target financial product in the second time period and in the third time period, the sixth category of customers are customers who have traded the target financial product in the second time period and in the third time period, and the ratio between the quantity of the second positive sample and the quantity of the second negative sample is a second preset ratio.

[0017] Further, the training unit includes: a partitioning subunit for dividing the processed third feature information into a first training set and a first test set; a first training subunit for training N first preset models using the first training set to obtain N third models and the output results of the N first preset models; a first construction subunit for testing the N third models using the first test set to obtain the output results of the N third models and constructing a second test set based on the output results of the N third models; a second construction subunit for constructing a second training set based on the output results of the N first preset models and the first training set; and a second training subunit for training a second preset model using the second training set and optimizing the trained second preset model using the second test set to obtain the first prediction model.

[0018] Further, the processing unit includes: a filling subunit, used to determine the missing information in the third feature information, and fill the missing information in the third feature information according to the data type of the missing information to obtain the fifth feature information, wherein the data type of the missing information includes at least one of the following: transaction behavior type, time type, and numerical type; a deletion subunit, used to delete a preset string in the fifth feature information to obtain the sixth feature information, wherein the preset string includes at least one of the following: a string corresponding to duplicate information and a string corresponding to semantically unrelated information; and a filtering subunit, used to filter key features in the sixth feature information based on a target indicator value to obtain the processed third feature information, wherein the calculation method of the target indicator value includes at least one of the following: evidence weight, information value, and random forest algorithm.

[0019] Furthermore, the first transaction information of the first type of customer includes at least the following information: the customer's basic information, asset information, liability information, transaction behavior information, and client behavior data.

[0020] To achieve the above objectives, according to one aspect of this application, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the potential customer prediction method described in any of the above-mentioned embodiments, and the computer program, when executed by a processor, implements the steps of the potential customer prediction method described in various embodiments of this application.

[0021] To achieve the above objectives, according to one aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including stored computer instructions, wherein, when the computer instructions are executed by a processor, the potential customer prediction method described in any one of the above-mentioned methods is implemented.

[0022] To achieve the above objectives, according to one aspect of this application, an electronic device is provided, including one or more processors and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the potential customer prediction method described in any of the above-mentioned embodiments.

[0023] In this embodiment, customers are divided into a first category and a second category based on transaction information within a first time period. The first category consists of customers who have not traded the target financial product within the first time period, while the second category consists of customers who have traded the target financial product within the first time period. First transaction information of the first category and second transaction information of the second category are obtained from the transaction information. First feature information is extracted from the first transaction information, and second feature information is extracted from the second transaction information. The first feature information is input into a first prediction model to predict potential customers in the first category, and the second feature information is input into a second prediction model to predict potential customers in the second category. Product information of the target financial product is then pushed to the potential customers in both the first and second categories. This solves the technical problem that when using machine learning models to identify potential customers purchasing financial products, the imbalance of sample information corresponding to different categories of customer transaction information leads to inaccurate model output, thus affecting the accurate delivery of product information.

[0024] By extracting features from customer transaction data and using the extracted features to train a predictive model, it is possible to predict the likelihood of each customer becoming a potential customer for the target financial product. This allows banks to accurately and effectively identify customers with high purchasing potential. At the same time, by classifying non-customer manager accounts into new and existing customers based on transaction behavior, financial institutions can narrow the scope of recommended customers, reduce the cost and waste of resources from blindly reaching a large number of non-target customers, and push different product information to different types of customers, thereby improving the accuracy of financial product information delivery and customer satisfaction. Attached Figure Description

[0025] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0026] Figure 1 This is a hardware structure block diagram of a computer terminal (or mobile device) for implementing a method for predicting potential customers, according to Embodiment 1 of this application.

[0027] Figure 2 This is a flowchart of an optional potential customer prediction method provided according to Embodiment 1 of this application;

[0028] Figure 3 This is a schematic diagram of the optional training process for two prediction models provided in Embodiment 1 of this application;

[0029] Figure 4This is a flowchart of an optional customer classification provided according to Embodiment 1 of this application;

[0030] Figure 5 This is a schematic diagram of the observation period and performance period in the optional process of predicting potential customers, provided according to Embodiment 1 of this application;

[0031] Figure 6 This is a schematic diagram of an optional ensemble learning process based on a primary learner and a secondary learner, provided according to Embodiment 1 of this application.

[0032] Figure 7 This is a schematic diagram of a potential customer prediction device provided according to Embodiment 2 of this application;

[0033] Figure 8 This is a schematic diagram of a predictive electronic device for potential customers provided according to Embodiment 5 of this application. Detailed Implementation

[0034] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0035] It should be noted that the processing method, apparatus, storage medium, and electronic device specified in this application can be used in the fintech field to improve the accuracy of product information push when pushing financial products to potential customers who are purchasing financial products. It can also be used in any field other than fintech. The application field of the processing method, apparatus, storage medium, and electronic device specified in this application is not limited.

[0036] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, collected data, used data, generated data, processed data, etc.) and the data (including but not limited to data used for analysis, stored data, displayed data, collected information, used information, generated information, processed information, etc.) are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant countries and regions, and necessary confidentiality measures have been taken. These measures do not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations, providing users with corresponding operation entry points for users to choose to agree to or refuse automated decision results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0037] Example 1

[0038] According to an embodiment of this application, a method embodiment for predicting potential customers is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0039] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a method for predicting potential customers is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102n, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (US) port (which may be included as one of the ports of the US bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0040] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0041] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the potential customer prediction method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned potential customer prediction method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0042] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0043] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0044] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for predicting potential customers is shown. Figure 2 This is a flowchart of a method for predicting potential customers according to Embodiment 1 of this application.

[0045] Step S101: Based on the transaction information within the first time period, customers are divided into a first category of customers and a second category of customers. The first category of customers are those who have not traded the target financial product within the first time period, and the second category of customers are those who have traded the target financial product within the first time period.

[0046] In this first embodiment, the first time period can be defined as the past 6 months, or the most recent 180 days. Within this time window, financial institutions analyze non-account manager-managed clients based on transaction information of financial products, identify clients' transaction behaviors, and thus classify non-account manager-managed clients into first-class clients and second-class clients. For example, the target financial product can be a fund product.

[0047] The first type of customer refers to non-account manager-managed customers who have not purchased any funds (target financial products) within the defined first time period (the past 6 months), i.e., new customers, or potential new buyers. The second type of customer refers to non-account manager-managed customers who have purchased funds (i.e., the aforementioned target financial products) at least once within the defined first time period (the past 6 months), i.e., existing customers.

[0048] By segmenting customers based on transaction information, financial institutions can implement differentiated promotional strategies precisely for different customer groups. For example, they can introduce and promote products to new customers, while increasing transaction frequency and investment amount for existing customers to improve the accuracy of product information delivery. Through this precise segmentation, financial institutions can allocate resources more effectively and improve customer satisfaction.

[0049] Step S102: Obtain the first transaction information of the first type of customer and the second transaction information of the second type of customer from the transaction information.

[0050] In this first embodiment, the first transaction information refers to the transaction information of the first type of customer during the observation period (i.e., data from the past 12 months). For example, the basic information of the first type of customer (such as age and occupation), holding status of other financial products in the bank (such as deposit balance and loan balance), fund transaction behavior (such as frequency of use of mobile banking and number of transactions), transaction behavior or customer characteristics related to transactions of multiple financial products, etc.

[0051] The second type of client's second transaction information refers to fund transaction records and other transaction information related to fund transactions during the observation period (i.e., data from the past 12 months). The method for obtaining the second type of client's second transaction information is similar to the method for obtaining the first type of client's first transaction information.

[0052] By acquiring primary transaction information, we can identify customers who, although they have not purchased funds before, have a high probability of becoming new clients in the future. By acquiring secondary transaction information, we can analyze existing fund transaction data and, in conjunction with other transaction behaviors, identify existing clients with the potential to increase the frequency or amount of their fund investments.

[0053] By acquiring and analyzing the transaction information of the first and second types of customers through the above methods, we can more effectively utilize data mining and machine learning technologies to accurately classify and recommend potential customers, while reducing promotion costs and improving the accuracy of financial product information delivery and customer satisfaction.

[0054] Step S103: Extract features from the first transaction information to obtain first feature information, and extract features from the second transaction information to obtain second feature information.

[0055] In this first embodiment, feature extraction is performed on the first transaction information to extract features that reflect the willingness of new customers to purchase financial products. For example, by focusing on information such as the customer's age, asset status, frequency of mobile banking use, whether they have browsed financial product pages, and whether they have transaction records for other financial products (such as loans and wealth management), the first feature information is constructed. This first feature information is then used to train a model to predict potential customers with a high probability of becoming new customers.

[0056] Feature extraction from secondary transaction information involves in-depth analysis of this information. For example, it can analyze a customer's transaction frequency, the distribution of transaction amounts, the timing of transactions, the types of financial products held, browsing behavior on financial product pages on mobile banking, and the customer's service rating. This analysis reflects existing customers' interest in financial products and their investment behavior patterns, forming secondary feature information that is used to train models to predict potential customers among existing customers who have the potential to increase the frequency or amount of their fund investments.

[0057] By transforming raw transaction data into targeted feature information, predictive models can more accurately learn and predict customer behavior patterns, thereby improving the accuracy and effectiveness of financial product information delivery.

[0058] Step S104: Input the first feature information into the first prediction model to predict potential customers in the first type of customers; input the second feature information into the second prediction model to predict potential customers in the second type of customers; and push product information of the target financial product to potential customers in the first and second types of customers.

[0059] In this first embodiment, the first feature information is input into the trained first prediction model (also known as the new customer acquisition model). Based on the first feature information, the probability of fund purchase for each first type of customer is calculated, thereby predicting potential new customers among the first type of customers.

[0060] The second feature information is input into the trained second prediction model (also known as the quality improvement model) to predict that there are old customers in the second type of customers who have the potential to increase their fund investment and become higher-value customers.

[0061] After the first and second prediction models identify potential new customers in the first customer category and potential existing customers in the second customer category, targeted promotional strategies can be implemented. For example, information on financial products can be pushed to these potential customers, including detailed product descriptions, purchase processes, potential returns and risks, and promotional activities, in order to stimulate their investment interest and increase fund sales. The accuracy and personalization of the pushed information usually directly affect customers' understanding of fund products and their purchase decisions; therefore, the pushed content can be designed based on the model prediction results to meet the needs of different potential customers.

[0062] Through the above process, financial institutions can effectively utilize machine learning models to accurately identify potential buyers and potential high-quality customers of fund products based on the characteristics of customer behavior and transaction data, thereby optimizing promotion strategies and improving the accuracy of financial product information delivery and customer satisfaction.

[0063] Optionally, in the potential customer prediction method provided in Embodiment 1 of this application, feature extraction of the first transaction information to obtain the first feature information includes: statistically analyzing the first type of variables in the first transaction information according to different time dimensions to obtain the first type of data, wherein the first type of variables represents variables represented by a single numerical value; calculating the second type of variables in the first transaction information using an unsupervised binning algorithm to obtain the second type of data, wherein the second type of variables represents variables represented by numerical values ​​within a range; calculating multiple variables in the first transaction information to obtain the third type of data, wherein the multiple variables include: the first type of variables and the second type of variables; and determining the first feature information corresponding to the first transaction information based on the first type of data, the second type of data, and the third type of data.

[0064] In this first embodiment, the first type of variable refers to variables that can be directly represented by a single numerical value, such as transaction amount, salary disbursement, number of times financial products are viewed, and number of mobile banking logins in the past year. These variables are statistically analyzed according to different time dimensions (such as monthly, quarterly, and annually) to obtain the first type of data. For example, for mobile banking logins, the bank might collect statistics on the number of logins in the past month, the past three months, and the past twelve months to reflect the customer's recent and long-term mobile banking usage habits. The statistical analysis of this data aims to capture the dynamic behavior and financial status of customers, providing stable and interpretable features for the model.

[0065] The second type of variable refers to variables that need to be represented by values ​​within a range, such as customer age and asset information. For these variables, unsupervised binning algorithms can be used for discretization. Unsupervised binning algorithms can transform continuous values ​​into bins or intervals, with each interval representing a category. For example, customer age can be binned into "18-25 years old," "26-35 years old," "36-45 years old," etc. This reduces the impact of outliers and allows the model to better handle non-linear relationships.

[0066] The third type of data is obtained by calculating combinations of the first and second types of variables. For example, one can calculate the combined characteristics of a customer's age and bank balance, or the correlation between the number of logins in recent months and browsing behavior. These combined characteristics often capture the complex relationship between customer behavior and financial status, providing richer information for the model.

[0067] Finally, the first feature information is obtained by combining the first, second, and third types of data. By extracting feature information from the original transaction data or creating new feature information based on the original transaction data, relationships and patterns not directly reflected in the original data can be captured. This helps the model better understand and predict customer behavior, while also increasing the model's generalization ability and reducing overfitting to the training dataset. This helps the prediction model output more accurate prediction results, further improving the accuracy of financial product information delivery and customer satisfaction.

[0068] Optionally, in the potential customer prediction method provided in Embodiment 1 of this application, before inputting the first feature information into the first prediction model to predict potential customers in the first type of customers, and inputting the second feature information into the second prediction model to predict potential customers in the second type of customers, the method further includes: collecting historical transaction information of historical customers in financial institutions; constructing a first sample set and a second sample set based on the historical transaction information; wherein the times when historical customers in the first sample set and the second sample set purchased financial products are different; performing feature extraction on the first sample set to obtain third feature information, and performing feature extraction on the second sample set to obtain fourth feature information; performing data preprocessing on the third feature information to obtain processed third feature information, and performing data preprocessing on the fourth feature information to obtain processed fourth feature information; training a preset model using the processed third feature information to obtain a first prediction model, and training the preset model using the processed fourth feature information to obtain a second prediction model.

[0069] In this first embodiment, financial institutions collect historical transaction information from their customers regarding various transaction activities, including but not limited to fund purchase records, deposit operations, loan information, and mobile banking usage. Based on the time attributes of the historical transaction information, a first sample set and a second sample set are constructed. The first sample set includes customer data that did not purchase funds during the observation period (e.g., the past 12 months) but purchased funds in the next time window (e.g., one month after the observation period); these customers are defined as positive samples in the first sample set. The second sample set includes customer data that purchased funds during the observation period and continued to purchase funds within one month after the observation period; these customers are defined as positive samples in the second sample set.

[0070] Then, feature extraction is performed on the first sample set to obtain the third feature information, which may include the customer's basic information, asset and liability status, product holdings, fund transaction behavior, mobile banking usage habits, etc. Similarly, feature extraction is performed on the second sample set to obtain the fourth feature information. The feature extraction process can adopt the same process as the feature extraction process for the first transaction information of the first type of customer, thereby more accurately describing the customer's behavior and attributes.

[0071] Secondly, the third and fourth feature information are preprocessed accordingly to obtain the processed third and fourth feature information. The purpose of preprocessing is to clean the data, fill in missing information, and transform the data format, thereby improving the prediction accuracy and stability of the model.

[0072] Finally, the pre-set model is trained using the processed third feature information to obtain the first prediction model, namely the new customer acquisition model. The new customer acquisition model is specifically designed to predict potential new customers who will purchase financial products in the future. Similarly, the pre-set model is trained using the processed fourth feature information to obtain the second prediction model, namely the quality improvement model. The quality improvement model is used to predict existing customers who have already purchased funds and are likely to increase their fund investments or investment frequency.

[0073] By collecting historical customer transaction information, two time-distinct sample sets are constructed. Features are then extracted from these two sample sets, and data preprocessing is performed. Finally, the processed feature information is used to train two different prediction models (a new customer acquisition model and a customer quality improvement model). This enables financial institutions to accurately predict potential new and existing customers who will purchase financial products based on machine learning models, thereby implementing more effective promotion strategies and improving the accuracy of financial product information delivery and customer satisfaction.

[0074] Optionally, in the potential customer prediction method provided in Embodiment 1 of this application, constructing a first sample set and a second sample set based on historical transaction information includes: determining a first positive sample based on historical transaction information belonging to the third type of customer, determining a first negative sample based on historical transaction information belonging to the fourth type of customer, and constructing a first sample set based on the first positive sample and the first negative sample. The third type of customer is a customer who did not trade the target financial product during the second time period but traded the target financial product during the third time period; the fourth type of customer is a customer who did not trade the target financial product during the second time period but did not trade the target financial product during the third time period. The third time period is later than the second time period. Within a given time period, the ratio between the number of the first positive samples and the number of the first negative samples is the first preset ratio. Second positive samples are determined based on historical transaction information of customers belonging to the fifth category, and second negative samples are determined based on historical transaction information of customers belonging to the sixth category. A second sample set is constructed based on the second positive samples and the second negative samples. Customers belonging to the fifth category are those who traded the target financial product during both the second and third time periods; customers belonging to the sixth category are those who traded the target financial product during both the second and third time periods. The ratio between the number of the second positive samples and the number of the second negative samples is the second preset ratio.

[0075] In this first embodiment, the third type of customer refers to a customer who did not trade the target financial product (e.g., a fund) during the second time period (e.g., within 6 months), but traded the fund during the subsequent third time period (e.g., within 1 month after 6 months). Then, the first positive sample in the first sample set is formed based on the historical transaction information of the third type of customer during the observation period (e.g., 12 months).

[0076] The fourth type of customer refers to customers who did not trade the target financial product (e.g., funds) during the second time period (e.g., within 6 months), but also did not trade funds during the subsequent third time period (e.g., within 1 month after 6 months). Then, the first negative sample in the first sample set is formed based on the historical transaction information of the fourth type of customers during the observation period (e.g., 12 months).

[0077] The first positive and first negative samples are combined to form the first sample set, which is used to train the new user acquisition model, i.e., the first prediction model mentioned above. It is important to note that when constructing the first sample set, the ratio between the number of positive and negative samples needs to be carefully considered; this is the first preset ratio. For example, through subsequent data preprocessing and feature extraction steps, the ratio of positive to negative samples can be adjusted from its original highly imbalanced state to a more reasonable ratio, such as reducing it from 1:1500 to 1:500, to improve the efficiency of model training and the accuracy of prediction.

[0078] The fifth type of customer refers to customers who traded the target financial product (e.g., a fund) within the second time period (e.g., within 6 months), but then traded the fund again within the subsequent third time period (e.g., within 1 month after 6 months). The second positive sample in the second sample set is then constructed based on the historical transaction information of the fifth type of customers during the observation period (e.g., 12 months).

[0079] The sixth type of customer refers to customers who traded the target financial product (e.g., a fund) during the second time period (e.g., within 6 months), but did not trade the fund during the subsequent third time period (e.g., within 1 month after 6 months). Then, the second negative sample in the second sample set is constructed based on the historical transaction information of the sixth type of customers during the observation period (e.g., 12 months).

[0080] The determined second positive sample and second negative sample are combined into a second sample set, which is used to train the quality improvement model, i.e., the second prediction model mentioned above. It should also be noted that the ratio between the number of positive and negative samples in the second sample set (and the second preset ratio mentioned above) needs to be reasonably adjusted, for example, reduced from 1:45 to 1:30, to ensure that the model can fully learn the feature differences between positive and negative samples during training, thereby improving the model's generalization ability and prediction accuracy.

[0081] Through the above steps, customer groups can be accurately segmented, and a first sample set and a second sample set can be constructed to train the new customer acquisition model and the quality improvement model, respectively. These two models are optimized for different customer groups (new and existing customers) to predict new customers with potential fund purchases or existing customers with increased investment potential. This achieves the effects of narrowing the potential customer base, reducing promotion costs, and improving the accuracy of financial product information delivery and customer satisfaction. Simultaneously, by adjusting the ratio of positive to negative samples in the first and second sample sets, overfitting of the models can be prevented, and the accuracy of the model prediction results can be improved.

[0082] Optionally, in the potential customer prediction method provided in Embodiment 1 of this application, the processed third feature information is used to train a preset model to obtain a first prediction model, including: dividing the processed third feature information into a first training set and a first test set; training N first preset models using the first training set to obtain N third models and the output results of the N first preset models; testing the N third models using the first test set to obtain the output results of the N third models, and constructing a second test set based on the output results of the N third models; constructing a second training set based on the output results of the N first preset models and the first training set; training a second preset model using the second training set, and optimizing the trained second preset model using the second test set to obtain the first prediction model.

[0083] In this first embodiment, the processed third feature information (feature information for the new user acquisition model) is divided into a first training set and a first test set. The first training set is used to train N first preset models, resulting in N third models. Here, the N first preset models refer to multiple pre-selected machine learning models, such as LightGBM, random forest, and decision tree. The N third models can also be referred to as N primary learners.

[0084] Then, using the N third models trained on the first training set, each model makes predictions on the first test set, resulting in the outputs of the N third models. The outputs of the N third models on the first test set will be used to construct the second test set.

[0085] Next, based on the outputs of N first-preset models and the first training set, a second training set is constructed. In this stage, the second training set not only includes the original first training set but also incorporates the outputs of the primary learner as additional features, along with the label information corresponding to each feature. This combined feature set will be used to train the second-preset model (also known as the secondary learner). Then, the N third models are tested using the first test set to obtain the outputs of the N third models.

[0086] Next, a second training set is used to train the second pre-defined model (also known as a meta-learner or secondary learner; the second pre-defined model can be a logistic regression model), resulting in the trained second pre-defined model. In this process, the second pre-defined model combines the outputs of N first pre-defined models with the original feature information (i.e., the aforementioned first training set) to provide more accurate prediction results. Subsequently, a second test set is used to test and fine-tune the trained second pre-defined model, resulting in the aforementioned first prediction model (new user acquisition model), to evaluate the overall performance of the model and adjust parameters to improve prediction accuracy.

[0087] In one optional embodiment, the processed fourth feature information (feature information for the quality improvement model) is divided into a third training set and a third test set. The third training set is used to train N third preset models to obtain N fourth models. These N third preset models can also be models such as LightGBM, Random Forest, and Decision Tree. The N fourth models can also be referred to as N primary learners.

[0088] Then, N fourth models are used to make predictions on the third test set, resulting in the outputs of the N fourth models. The outputs of the N fourth models on the third test set will be used to construct the fourth test set.

[0089] Next, based on the outputs of the N pre-defined third models and the second training set, a third training set is constructed. In this stage, the third training set not only includes the original third training set but also incorporates the outputs of the primary learner as additional features, along with the label information corresponding to each feature. This combined feature set will be used to train the fourth pre-defined model (also known as the secondary learner). Then, the N fourth models are tested using the fourth test set to obtain their outputs.

[0090] Next, the fourth training set is used to train the fourth preset model (also known as a meta-learner or secondary learner; the fourth preset model can be a logistic regression model), resulting in the trained fourth preset model. In this process, the fourth preset model combines the outputs of N third preset models with the original feature information (i.e., the aforementioned third training set) to provide more accurate prediction results. Subsequently, the trained fourth preset model is tested and fine-tuned using the fourth test set to obtain the aforementioned second prediction model (the improved model), to evaluate the overall performance of the model and adjust parameters to improve prediction accuracy.

[0091] Through the above steps, an integrated and optimized first prediction model (customer acquisition model) and second prediction model (quality improvement model) can be obtained. The prediction performance of this model is usually better than that of a single base model, especially in handling complex data and avoiding overfitting. This results in more accurate predictions, which helps financial institutions implement more effective promotion strategies and improve the accuracy of financial product information delivery and customer satisfaction.

[0092] Optionally, in the potential customer prediction method provided in Embodiment 1 of this application, data preprocessing is performed on the third feature information to obtain processed third feature information, including: determining missing information in the third feature information; filling in the missing information in the third feature information according to the data type of the missing information to obtain fifth feature information, wherein the data type of the missing information includes at least one of the following: transaction behavior type, time type, and numerical type; deleting a preset string in the fifth feature information to obtain sixth feature information, wherein the preset string includes at least: the string corresponding to duplicate information and the string corresponding to semantically unrelated information; and filtering key features in the sixth feature information based on the target indicator value to obtain processed third feature information, wherein the calculation method of the target indicator value includes at least: evidence weight, information value, and random forest algorithm.

[0093] In this first embodiment, missing data in the third feature information (i.e., feature information extracted from historical customers for building the new customer acquisition model) is identified. Since this missing information may involve different data types (e.g., transaction behavior type (such as fund transaction behavior), time type (such as transaction date), numerical type (such as asset balance)), different filling strategies can be adopted according to the data type of the missing information.

[0094] In one optional embodiment, for missing data belonging to the transaction behavior type, i.e., data missing due to the absence of a behavior, it can be filled with 0 (or a specific marker) to indicate that the behavior does not exist. For example, missing data due to a customer not activating a certain service, not holding a certain financial product, or not having performed a behavior can be filled with 0. For missing data belonging to the time type, a special value (such as 999) can be used to fill it. For example, if a customer did not purchase a fund through mobile banking, and the missing data is the time of the most recent login to mobile banking, filling it with 999 means that the customer has not logged into mobile banking for a long time. For missing data belonging to the numerical type, statistical values ​​such as the mean and median can be used to fill it. For example, if the missing data is the customer's age, it can be filled based on the historical average age of customers to maintain the distribution characteristics of the numerical feature. After the above missing value filling process, complete fifth feature information is obtained, ensuring the integrity of the feature information.

[0095] Next, preset strings are removed from the fifth feature information. These strings typically represent meaningless information, such as duplicate field names, empty fields, or field values ​​irrelevant to the prediction target. Preset strings can be predefined to identify specific types of useless information. By removing meaningless information, a more concise sixth feature information is obtained, reducing the impact of noisy data on model training and improving data quality and the efficiency of subsequent feature selection.

[0096] Finally, features in the sixth feature information are filtered based on the target index value to retain feature information that is of great value for predicting potential customers.

[0097] Methods for calculating target index values ​​include, but are not limited to: Weight of Evidence (WOE): measures the strength of the relationship between feature values ​​and positive class target values, used for quantitative evaluation of features; Information Value (IV): calculated based on WOE, reflects the ability of features to distinguish between positive and negative samples, and features with high IV values ​​usually contribute more to prediction; and feature importance scoring during the training process of the random forest model is used to identify the features that have the greatest impact on the model's prediction results.

[0098] In an optional embodiment, a first variable with an information value greater than 0.02 can be selected, and the top 80 variables with an importance index (e.g., feature_importance) greater than 0 can be selected based on the importance ranking results of the first variable given by the random forest model to obtain the above-mentioned processed third feature information.

[0099] In an optional embodiment, the fourth feature information can be preprocessed in the same way as the steps of preprocessing the third feature information to obtain the processed third feature information.

[0100] This series of preprocessing steps ensures that the feature information used for model training is both complete and highly predictive, thereby improving the model's accuracy and stability, reducing the risk of overfitting, and enabling more accurate prediction of potential new fund customers, thus improving the accuracy of financial product information delivery and customer satisfaction.

[0101] Optionally, in the potential customer prediction method provided in Embodiment 1 of this application, the first transaction information of the first type of customer includes at least the following information: basic customer information, asset information, liability information, transaction behavior information, and client behavior data.

[0102] In this first embodiment, the basic information involves the customer's personal attributes, including but not limited to customer service rating, education level, marital status, occupation, age, and region of residence. This basic information helps to understand the customer's background, such as their potential investment risk appetite, income level, and financial stability. It is important to note that the collection of basic customer information is conducted only after obtaining the customer's authorization.

[0103] Asset information reflects a client's asset holdings in banks or other financial institutions, including but not limited to deposit balances, fund and insurance balances (which may refer to funds, insurance, wealth management products, etc.), the amount of low, medium, and high-risk funds held by the client, and the average daily assets over the past 3, 6, and 12 months, RMB and foreign currency wealth management products, third-party deposits, crude oil accounts, credit cards, paper gold activation indicators, personal consumer loans, personal housing loans, and personal business loan activation indicators. Asset information is an important indicator for assessing a client's investment potential and financial condition.

[0104] Debt information includes loan balances, credit card overdraft balances, and other credit card liabilities. Analyzing a customer's debt structure allows for the assessment of their financial health and debt level, which is crucial for predicting their investment behavior and fund purchase potential.

[0105] Transaction behavior information records detailed information about customers' transactions with the bank, including purchases of financial products via mobile banking, the number and amount of debit card transactions, credit card repayments, and the frequency and amount of online purchases of financial products. This information provides direct evidence of customers' trading habits and helps predict their interest and ability to purchase funds.

[0106] Client-side behavioral data primarily originates from customers' mobile banking usage records, including the time elapsed since their last mobile banking login, whether they have logged into mobile banking within the past 30 days, the number of times and duration they viewed their "My Account" section on mobile banking within the past month, the number of times they visited the "Task Center" within the past month, and the number of times they logged into mobile banking within the past 1, 3, and 6 months. This behavioral data reflects customers' activity and interests in banking services, and is of significant value in understanding customers' investment intentions and predicting their fund purchase behavior.

[0107] By collecting and analyzing the aforementioned multi-dimensional first-transaction information, a detailed customer profile can be constructed. This profile not only helps the model understand the customer's basic situation and financial status but also captures the customer's trading habits and preferences for specific financial products, thereby improving the model's accuracy and effectiveness in predicting new fund customers. Ultimately, this information will be used to build and train a customer acquisition model to identify high-potential new fund buyers, thus improving the accuracy of financial product information delivery and customer satisfaction.

[0108] Optionally, in this first embodiment, Figure 3 This is a flowchart illustrating the process of training two predictive models. First, based on whether customers have purchased funds in the past 6 months, individual bank customers not directly managed by bank account managers are segmented into two groups: "old customers" and "new customers".

[0109] The "returning customer" group then enters the "quality improvement model" process. This model aims to identify potential returning customers who may repurchase funds within the next month. Before training the quality improvement model, the returning customer dataset is first screened to narrow the sample range and address the impact of data imbalance, thereby improving model efficiency and prediction accuracy. Next, feature construction is performed on the screened samples, including feature extraction, handling missing and outlier values, and using techniques such as information value (IV) and random forest feature importance to select key features. Finally, predictions from multiple pre-set models (such as LightGBM, random forest, and decision tree) are integrated, and logistic regression is used as the meta-learner for training to improve prediction accuracy.

[0110] The "new customer" group then enters the training process of the "new customer acquisition model," with the goal of identifying which new customers are likely to make their first fund purchase within the next month. The three steps of sample range selection, feature construction, processing and selection, and ensemble learning are similar to the process of the quality improvement model, but they are specifically designed for the new customer group, with the aim of more accurately predicting the likelihood of new customers making fund purchases.

[0111] Optionally, in this first embodiment, Figure 4 This is a flowchart of customer classification. First, all customers within the financial institution not directly managed by a wealth management manager are identified. Then, they are classified based on whether they have purchased funds in the past 180 days. If a customer has purchased funds at least once in the past 180 days, they are classified as a "returning customer." These customers have some knowledge and investment experience with fund products and are potential targets for the bank to increase their investment amount and frequency. Conversely, if a customer has no record of purchasing funds in the past 180 days, they are classified as a "new customer." This may include customers considering investing in funds for the first time, or customers who have not invested in funds in the past period.

[0112] Optionally, in this first embodiment, Figure 5 It is a diagram illustrating the observation and performance periods in the process of predicting potential customers, clearly depicting the concept of time windows in the process of discovering potential customers. Figure 5 This includes the "observation period" on the left and the "performance period" on the right. The observation period can be defined as the past 12 months, a time window during which the model analyzes and learns about the client's trading behavior, investment patterns, and personal characteristics. During this period, the model will be trained based on the client's historical data (such as purchasing behavior, account activity, and personal attributes) to understand and predict the client's behavioral patterns. The performance period can be set within one month after the end of the observation period; this is a time window for the model to predict whether the client will purchase financial products. During the performance period, the model will predict whether the client will make fund purchases in the following month based on the patterns and rules learned during the observation period, thus potentially classifying the client. The observation point is the dividing line between the observation period and the performance period, usually taken as the end of a month.

[0113] In real-world production and daily life, financial institutions manage a vast customer base outside of their account managers. For example, one bank has approximately 700 million non-account holders (i.e., the customer base not managed by account managers). Within the one-month observation period, only 190,000 customers purchased non-regular investment funds, a figure that is extremely imbalanced compared to the non-investment group. Data analysis revealed that 99% of these customers are between 18 and 65 years old, and 95% logged into mobile banking at least once in the past year. Therefore, by performing data preprocessing (e.g., data deletion) and filtering (e.g., filtering based on information values, random forest algorithms, etc.), the overall sample can be filtered. This will not reduce the number of positive samples, but it will effectively reduce the number of negative samples, thus mitigating the extreme imbalance in the sample datasets (e.g., the first and second sample sets mentioned above).

[0114] Optionally, in this first embodiment, Figure 6 This is a flowchart illustrating the ensemble learning process based on primary and secondary learners. In the primary learner stage, model training is divided into three independent but related parts, each handled by a different model (Model 1, Model 2, Model 3). Model 1, Model 2, and Model 3 can be different machine learning models such as LightGBM, Random Forest, or Decision Tree. Each model is trained using the same training set to obtain its own training results. After training, each model makes predictions on the test set, obtaining its own test results. These test results are typically evaluated, such as by calculating accuracy, recall, and precision, to measure the performance of the primary learner.

[0115] In the secondary learner stage, the task is to perform higher-level learning and prediction based on the output of the primary learner and the features of the original training set. Unlike traditional ensemble learning models (Stacking models), the original training set is also used as part of the input to the meta-learner. This helps the meta-learner not only learn the output of the primary learner, but also directly utilize the features of the original data, thereby better understanding the relationship between model predictions and actual customer behavior.

[0116] The training results of Model 1, Model 2, and Model 3, along with the first training set and label information, are combined to form a new training set, namely the second training set mentioned above. The test results of Model 1, Model 2, and Model 3 are then combined to form a new test set, namely the second test set mentioned above. The second test set is used to evaluate the performance of the secondary learner on unseen data after training.

[0117] Finally, the meta-learner (usually a logistic regression model) will be trained using a second training set to learn how to combine the predictions of the primary learner with the original features to make more accurate fund purchase predictions. After training, the meta-learner will be evaluated using a second test set to ensure its generalization ability.

[0118] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0119] In summary, the potential customer prediction method provided in this application divides customers into a first category and a second category based on transaction information within a first time period. The first category consists of customers who have not traded the target financial product within the first time period, while the second category consists of customers who have traded the target financial product within the first time period. The method obtains first transaction information for the first category and second transaction information for the second category from the transaction information. It then extracts features from the first transaction information to obtain first feature information and from the second transaction information to obtain second feature information. The first feature information is input into a first prediction model to predict potential customers within the first category, and the second feature information is input into a second prediction model to predict potential customers within the second category. Finally, it pushes product information of the target financial product to the potential customers in both categories. This solves the problem in related technologies where, when using machine learning models to identify potential customers purchasing financial products, the imbalance of sample information corresponding to different categories of customer transaction information leads to inaccurate model output, thus affecting the accurate delivery of product information.

[0120] By extracting features from customer transaction data and using the extracted features to train a predictive model, it is possible to predict the likelihood of each customer becoming a potential customer for the target financial product. This allows banks to accurately and effectively identify customers with high purchasing potential. At the same time, by classifying non-customer manager accounts into new and existing customers based on transaction behavior, financial institutions can narrow the scope of recommended customers, reduce the cost and waste of resources from blindly reaching a large number of non-target customers, and push different product information to different types of customers, thereby improving the accuracy of financial product information delivery and customer satisfaction.

[0121] Example 2

[0122] This application also provides a potential customer prediction device. It should be noted that the potential customer prediction device of this application can be used to execute the potential customer prediction method provided in this application. The potential customer prediction device provided in this application is described below.

[0123] According to embodiments of this application, an apparatus for implementing the above-described method for predicting potential customers is also provided, such as... Figure 7 As shown, the device includes:

[0124] Specifically, the segmentation unit 701 is used to segment customers based on transaction information within the first time period to obtain a first type of customer and a second type of customer. The first type of customer is a customer who has not traded the target financial product within the first time period, and the second type of customer is a customer who has traded the target financial product within the first time period.

[0125] The acquisition unit 702 is used to acquire the first transaction information of the first type of customer and the second transaction information of the second type of customer from the transaction information.

[0126] The first extraction unit 703 is used to extract features from the first transaction information to obtain first feature information, and to extract features from the second transaction information to obtain second feature information.

[0127] The push unit 704 is used to input the first feature information into the first prediction model to predict potential customers in the first type of customers, input the second feature information into the second prediction model to predict potential customers in the second type of customers, and push product information of the target financial product to potential customers in the first type of customers and potential customers in the second type of customers.

[0128] The potential customer prediction device provided in this application embodiment divides customers into a first category and a second category based on transaction information within a first time period using a segmentation unit 701. The first category consists of customers who have not traded the target financial product within the first time period, while the second category consists of customers who have traded the target financial product within the first time period. An acquisition unit 702 acquires first transaction information of the first category and second transaction information of the second category from the transaction information. A first extraction unit 703 extracts features from the first transaction information to obtain first feature information and extracts features from the second transaction information to obtain second feature information. A push unit 704 inputs the first feature information into a first prediction model to predict potential customers in the first category, inputs the second feature information into a second prediction model to predict potential customers in the second category, and pushes product information of the target financial product to potential customers in both categories. This solves the problem in related technologies where, when using machine learning models to identify potential customers purchasing financial products, the imbalance of sample information corresponding to transaction information of different customer categories leads to inaccurate model output results, thus affecting the accurate push of product information.

[0129] By extracting features from customer transaction data and using the extracted features to train a predictive model, it is possible to predict the likelihood of each customer becoming a potential customer for the target financial product. This allows banks to accurately and effectively identify customers with high purchasing potential. At the same time, by classifying non-customer manager accounts into new and existing customers based on transaction behavior, financial institutions can narrow the scope of recommended customers, reduce the cost and waste of resources from blindly reaching a large number of non-target customers, and push different product information to different types of customers, thereby improving the accuracy of financial product information delivery and customer satisfaction.

[0130] Optionally, in the potential customer prediction device provided in Embodiment 2 of this application, the first extraction unit 703 includes: a statistical subunit, used to perform statistics on a first type of variable in the first transaction information according to different time dimensions to obtain a first type of data, wherein the first type of variable represents a variable represented by a single numerical value; a first calculation subunit, used to calculate a second type of variable in the first transaction information using an unsupervised binning algorithm to obtain a second type of data, wherein the second type of variable represents a variable represented by a numerical value within an interval; a second calculation subunit, used to calculate multiple variables in the first transaction information to obtain a third type of data, wherein the multiple variables include: the first type of variable and the second type of variable; and a first determination subunit, used to determine the first feature information corresponding to the first transaction information based on the first type of data, the second type of data and the third type of data.

[0131] Optionally, in the potential customer prediction device provided in Embodiment 2 of this application, the device further includes: a collection unit, used to collect historical transaction information of historical customers in financial institutions before inputting the first feature information into the first prediction model to predict potential customers in the first type of customers and inputting the second feature information into the second prediction model to predict potential customers in the second type of customers; a construction unit, used to construct a first sample set and a second sample set based on the historical transaction information; wherein the purchase times of historical customers in the first sample set and the second sample set are different; a second extraction unit, used to extract features from the first sample set to obtain third feature information and extract features from the second sample set to obtain fourth feature information; a processing unit, used to preprocess the third feature information to obtain processed third feature information and preprocess the fourth feature information to obtain processed fourth feature information; and a training unit, used to train a preset model using the processed third feature information to obtain a first prediction model and to train the preset model using the processed fourth feature information to obtain a second prediction model.

[0132] Optionally, in the potential customer prediction device provided in Embodiment 2 of this application, the aforementioned construction unit includes: a second determining subunit, used to determine a first positive sample based on historical transaction information belonging to the third type of customer, determine a first negative sample based on historical transaction information belonging to the fourth type of customer, and construct a first sample set based on the first positive sample and the first negative sample, wherein the third type of customer is a customer who did not trade the target financial product in the second time period but traded the target financial product in the third time period, the fourth type of customer is a customer who did not trade the target financial product in the second time period but did not trade the target financial product in the third time period, the third time period being later than the second time period, and the first positive sample... The ratio between the number of positive samples and the number of negative samples is the first preset ratio; the third determining subunit is used to determine the second positive samples based on the historical transaction information of customers belonging to the fifth category, determine the second negative samples based on the historical transaction information of customers belonging to the sixth category, and construct a second sample set based on the second positive samples and the second negative samples. The fifth category of customers are customers who have traded the target financial product in the second time period and in the third time period, and the sixth category of customers are customers who have traded the target financial product in the second time period and in the third time period. The ratio between the number of positive samples and the number of negative samples is the second preset ratio.

[0133] Optionally, in the potential customer prediction device provided in Embodiment 2 of this application, the training unit includes: a partitioning subunit, used to partition the processed third feature information into a first training set and a first test set; a first training subunit, used to train N first preset models using the first training set respectively to obtain N third models and the output results of the N first preset models; a first construction subunit, used to test the N third models using the first test set respectively to obtain the output results of the N third models, and to construct a second test set based on the output results of the N third models; a second construction subunit, used to construct a second training set based on the output results of the N first preset models and the first training set; and a second training subunit, used to train a second preset model using the second training set, and to optimize the trained second preset model using the second test set to obtain a first prediction model.

[0134] Optionally, in the potential customer prediction device provided in Embodiment 2 of this application, the above-mentioned processing unit includes: a filling subunit, used to determine the missing information in the third feature information, and fill the missing information in the third feature information according to the data type of the missing information to obtain the fifth feature information, wherein the data type of the missing information includes at least one of the following: transaction behavior type, time type, and numerical type; a deletion subunit, used to delete a preset string in the fifth feature information to obtain the sixth feature information, wherein the preset string includes at least: the string corresponding to duplicate information and the string corresponding to semantically unrelated information; and a filtering subunit, used to filter key features in the sixth feature information based on the target index value to obtain the processed third feature information, wherein the calculation method of the target index value includes at least: evidence weight, information value, and random forest algorithm.

[0135] Optionally, in the potential customer prediction device provided in Embodiment 2 of this application, the first transaction information of the first type of customer mentioned above includes at least the following information: basic customer information, asset information, liability information, transaction behavior information, and client behavior data.

[0136] It should be noted that the division unit 701, acquisition unit 702, first extraction unit 703, and push unit 704 mentioned above correspond to steps S201 to S204 in Embodiment 1. The two modules and the corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102potential customer prediction, ..., 102n). The above modules can also be part of the device and run in the computer terminal 10 provided in Embodiment 1.

[0137] Example 3

[0138] Embodiments of this application may provide an electronic device. Figure 8 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 8 As shown, the electronic device may include: one or more ( Figure 8 Only one of the components is shown: processor 802, memory 804, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module, and display.

[0139] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0140] The processor can access information and applications stored in the memory via a transmission device to perform the following steps: classifying customers into a first category and a second category based on transaction information within a first time period, wherein the first category consists of customers who have not traded the target financial product within the first time period, and the second category consists of customers who have traded the target financial product within the first time period; obtaining first transaction information of the first category and second transaction information of the second category from the transaction information; extracting features from the first transaction information to obtain first feature information, and extracting features from the second transaction information to obtain second feature information; inputting the first feature information into a first prediction model to predict potential customers in the first category, inputting the second feature information into a second prediction model to predict potential customers in the second category, and pushing product information of the target financial product to the potential customers in the first and second categories.

[0141] The processor can access information and applications stored in the memory via a transmission device to execute the following steps: extracting features from the first transaction information to obtain first feature information, including: statistically analyzing a first type of variable in the first transaction information according to different time dimensions to obtain first type data, wherein the first type of variable represents a variable represented by a single numerical value; calculating a second type of variable in the first transaction information using an unsupervised binning algorithm to obtain second type data, wherein the second type of variable represents a variable represented by a value within a range; calculating multiple variables in the first transaction information to obtain third type data, wherein the multiple variables include: first type variables and second type variables; and determining the first feature information corresponding to the first transaction information based on the first type data, the second type data, and the third type data.

[0142] The processor can invoke information and application programs stored in the memory via a transmission device to execute the following steps: Before inputting the first feature information into the first prediction model to predict potential customers in the first type of customer, and inputting the second feature information into the second prediction model to predict potential customers in the second type of customer, the above method further includes: collecting historical transaction information of historical customers in financial institutions; constructing a first sample set and a second sample set based on the historical transaction information; wherein the times when historical customers in the first sample set and the second sample set purchased financial products are different; performing feature extraction on the first sample set to obtain third feature information, and performing feature extraction on the second sample set to obtain fourth feature information; performing data preprocessing on the third feature information to obtain processed third feature information, and performing data preprocessing on the fourth feature information to obtain processed fourth feature information; training a preset model using the processed third feature information to obtain a first prediction model, and training the preset model using the processed fourth feature information to obtain a second prediction model.

[0143] The processor can access information and applications stored in the memory via a transmission device to execute the following steps: constructing a first sample set and a second sample set based on historical transaction information, including: determining a first positive sample based on historical transaction information belonging to a third type of customer, determining a first negative sample based on historical transaction information belonging to a fourth type of customer, and constructing a first sample set based on the first positive sample and the first negative sample. The third type of customer refers to customers who did not trade the target financial product during the second time period but did trade the target financial product during the third time period; the fourth type of customer refers to customers who did not trade the target financial product during the second time period but did not trade the target financial product during the third time period. Later than the second time period, the ratio between the number of the first positive samples and the number of the first negative samples is the first preset ratio; the second positive samples are determined based on the historical transaction information of customers belonging to the fifth category, and the second negative samples are determined based on the historical transaction information of customers belonging to the sixth category. A second sample set is constructed based on the second positive samples and the second negative samples. Customers belonging to the fifth category are those who have traded the target financial product in both the second and third time periods, and customers belonging to the sixth category are those who have traded the target financial product in both the second and third time periods. The ratio between the number of the second positive samples and the number of the second negative samples is the second preset ratio.

[0144] The processor can access information and applications stored in memory via a transmission device to execute the following steps: training a preset model using processed third feature information to obtain a first prediction model, including: dividing the processed third feature information into a first training set and a first test set; training N first preset models using the first training set to obtain N third models and the output results of the N first preset models; testing the N third models using the first test set to obtain the output results of the N third models, and constructing a second test set based on the output results of the N third models; constructing a second training set based on the output results of the N first preset models and the first training set; training a second preset model using the second training set, and optimizing the trained second preset model using the second test set to obtain the first prediction model.

[0145] The processor can invoke information and application programs stored in the memory via a transmission device to perform the following steps: preprocessing the third feature information to obtain processed third feature information, including: determining missing information in the third feature information; filling in the missing information in the third feature information according to the data type of the missing information to obtain fifth feature information, wherein the data type of the missing information includes at least one of the following: transaction behavior type, time type, and numerical type; deleting a preset string from the fifth feature information to obtain sixth feature information, wherein the preset string includes at least one of the following: strings corresponding to duplicate information and strings corresponding to semantically unrelated information; filtering key features in the sixth feature information based on the target indicator value to obtain processed third feature information, wherein the calculation method of the target indicator value includes at least one of the following: evidence weight, information value, and random forest algorithm.

[0146] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: The first transaction information of the first type of customer includes at least the following information: customer basic information, asset information, liability information, transaction behavior information, and client behavior data.

[0147] Those skilled in the art will understand that Figure 8 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones (such as Android phones, iOS phones, etc.), tablets, PDAs, mobile Internet devices (MIDs), PADs, and other terminal devices. Figure 8 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 8 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 8 The different configurations shown.

[0148] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0149] Example 4

[0150] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the potential customer prediction method provided in Embodiment 1.

[0151] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals. This application also provides a computer program product, which, when executed on a data processing device, is suitable for performing steps of a method for predicting potential customers. The sequence numbers of the embodiments in this application are merely descriptive and do not represent the superiority or inferiority of the embodiments. In the above embodiments of this application, the descriptions of each embodiment have their own emphasis; parts not described in detail in a certain embodiment can be referred to the relevant descriptions of other embodiments.

[0152] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0153] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0154] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0155] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0156] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting potential customers, characterized in that, include: Customers are divided into two categories based on transaction information within the first time period: the first category consists of customers who have not traded the target financial product within the first time period, and the second category consists of customers who have traded the target financial product within the first time period. Obtain the first transaction information of the first type of customer and the second transaction information of the second type of customer from the transaction information; First feature information is obtained by extracting features from the first transaction information, and second feature information is obtained by extracting features from the second transaction information. The first feature information is input into the first prediction model to predict potential customers in the first type of customers. The second feature information is input into the second prediction model to predict potential customers in the second type of customers. The product information of the target financial product is pushed to the potential customers in the first type of customers and the potential customers in the second type of customers. The first prediction model is a model obtained by training a preset model with a first sample set. The second prediction model is a model obtained by training the preset model with a second sample set. The times when historical customers in the first sample set and the second sample set purchased financial products are different. Before inputting the first feature information into the first prediction model to predict potential customers in the first type of customer, and inputting the second feature information into the second prediction model to predict potential customers in the second type of customer, the method further includes: Collect historical transaction information of the aforementioned historical customers from financial institutions; The first sample set and the second sample set are constructed based on the historical transaction information.

2. The method according to claim 1, characterized in that, First feature information is obtained by extracting features from the first transaction information, including: The first type of variables in the first transaction information are statistically analyzed according to different time dimensions to obtain the first type of data, wherein the first type of variables represents variables represented by a single numerical value; An unsupervised binning algorithm is used to calculate the second type of variables in the first transaction information to obtain the second type of data, wherein the second type of variables represents variables represented by values ​​within a range; Calculations are performed on multiple variables in the first transaction information to obtain a third type of data, wherein the multiple variables include: the first type of variables and the second type of variables; The first feature information corresponding to the first transaction information is determined based on the first type of data, the second type of data, and the third type of data.

3. The method according to claim 1, characterized in that, Before inputting the first feature information into the first prediction model to predict potential customers in the first type of customer, and inputting the second feature information into the second prediction model to predict potential customers in the second type of customer, the method further includes: The first sample set is subjected to feature extraction to obtain the third feature information, and the second sample set is subjected to feature extraction to obtain the fourth feature information; The third feature information is preprocessed to obtain the processed third feature information, and the fourth feature information is preprocessed to obtain the processed fourth feature information. The preset model is trained using the processed third feature information to obtain the first prediction model, and the preset model is trained using the processed fourth feature information to obtain the second prediction model.

4. The method according to claim 1, characterized in that, Constructing the first sample set and the second sample set based on the historical transaction information includes: A first positive sample is determined based on the historical transaction information of customers belonging to the third category, and a first negative sample is determined based on the historical transaction information of customers belonging to the fourth category. A first sample set is constructed based on the first positive sample and the first negative sample. The third category of customers are those who have not traded the target financial product in the second time period but have traded the target financial product in the third time period. The fourth category of customers are those who have not traded the target financial product in the second time period but have not traded the target financial product in the third time period. The third time period is later than the second time period. The ratio between the number of the first positive sample and the number of the first negative sample is a first preset ratio. A second positive sample is determined based on the historical transaction information of customers belonging to the fifth category, and a second negative sample is determined based on the historical transaction information of customers belonging to the sixth category. A second sample set is constructed based on the second positive sample and the second negative sample. The fifth category of customers are those who have traded the target financial product during the second time period and during the third time period. The sixth category of customers are those who have traded the target financial product during the second time period and have not traded the target financial product during the third time period. The ratio between the number of the second positive sample and the number of the second negative sample is a second preset ratio.

5. The method according to claim 3, characterized in that, The first prediction model is obtained by training the preset model using the processed third feature information, including: The processed third feature information is divided into a first training set and a first test set; The first training set is used to train N first preset models to obtain N third models and the output results of the N first preset models. The first test set is used to test the N third models respectively to obtain the output results of the N third models, and a second test set is constructed based on the output results of the N third models; A second training set is constructed based on the output results of the N first preset models and the first training set; The second preset model is trained using the second training set, and the trained second preset model is optimized using the second test set to obtain the first prediction model.

6. The method according to claim 3, characterized in that, The third feature information is preprocessed to obtain the processed third feature information, including: The missing information in the third feature information is determined, and the missing information in the third feature information is filled in according to the data type of the missing information to obtain the fifth feature information. The data type of the missing information includes at least one of the following: transaction behavior type, time type, and numerical type. Delete the preset string in the fifth feature information to obtain the sixth feature information, wherein the preset string includes at least: the string corresponding to the duplicate information and the string corresponding to the semantically unrelated information; Based on the target index value, key features are selected from the sixth feature information to obtain the processed third feature information. The calculation method of the target index value includes at least: evidence weight, information value, and random forest algorithm.

7. The method according to claim 1, characterized in that, The first transaction information for the first type of customer includes at least the following information: basic customer information, asset information, liability information, transaction behavior information, and client behavior data.

8. A device for predicting potential customers, characterized in that, include: The segmentation unit is used to segment customers based on transaction information within a first time period to obtain a first type of customer and a second type of customer. The first type of customer is a customer who has not traded the target financial product within the first time period, and the second type of customer is a customer who has traded the target financial product within the first time period. The acquisition unit is used to acquire the first transaction information of the first type of customer and the second transaction information of the second type of customer from the transaction information; The first extraction unit is used to extract features from the first transaction information to obtain first feature information, and to extract features from the second transaction information to obtain second feature information; The push unit is used to input the first feature information into a first prediction model to predict potential customers in the first type of customers, input the second feature information into a second prediction model to predict potential customers in the second type of customers, and push product information of the target financial product to potential customers in the first type of customers and potential customers in the second type of customers. The first prediction model is a model obtained by training a preset model with a first sample set, and the second prediction model is a model obtained by training the preset model with a second sample set. The times when historical customers in the first sample set and the second sample set purchased financial products are different. The device further includes: a collection unit, configured to collect historical transaction information of the historical customers in the financial institution before inputting the first feature information into the first prediction model to predict potential customers in the first type of customers and inputting the second feature information into the second prediction model to predict potential customers in the second type of customers; and a construction unit, configured to construct the first sample set and the second sample set based on the historical transaction information.

9. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, performs the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 7.

Images