Customer service method and system based on machine learning technology

Abstract

The invention discloses a customer service method and system based on a machine learning technology. The method comprises the following steps: constructing a customer group portrait of a target object; taking the customer group portrait and a purchased financial product corresponding to the customer group portrait as inputs of a neural collaborative filtering model based on deep learning, taking a label of the purchased financial product as an output of the neural collaborative filtering model based on deep learning, and training the neural collaborative filtering model based on deep learning; obtaining target data used for describing the target object, and determining a target financial product corresponding to the target data by using the target model; and determining a risk level corresponding to the target financial product by using the deep learning model, and sending the target financial product and the risk level to the target object. The technical problem that the operation efficiency is low due to the fact that village and town banks cannot recommend corresponding financial products for different groups served by the village and town banks based on the artificial intelligence technology is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a customer service method and system based on machine learning technology. Background Art

[0002] When serving different customer groups, township banks traditionally use manual judgment and experience accumulation to recommend financial products. This model is mainly based on the bank's account manager or risk assessor's understanding of the customer's basic information, financial status and credit history to decide which type of financial product is suitable for a certain customer group. However, this method has the following problems: 1. Limited information processing capabilities: The ability to manually process large amounts of customer data is limited, making it difficult to accurately and quickly analyze the characteristics and needs of customer groups, thus affecting the accuracy of product matching. 2. High subjectivity in decision-making: The personal experience and subjective judgment of account managers play an important role in product recommendations, which may lead to inconsistency and uncertainty in recommendation results, affecting customer satisfaction and the bank's reputation. 3. Inefficient service: Manual processing procedures are usually cumbersome and require a lot of time and human resources, resulting in slow service response and difficulty in meeting rapidly changing customer needs.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The present application provides a customer service method and system based on machine learning technology to at least solve the technical problem of low operating efficiency caused by the inability of township banks to recommend corresponding financial products for different groups they serve based on artificial intelligence technology.

[0005] According to one aspect of the present application, a customer service method based on machine learning technology is provided, including: constructing a customer profile of a target object, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: rural areas, farmers and rural populations; using the customer profile and the purchased financial products corresponding to the customer profile as inputs of a neural collaborative filtering model based on deep learning, and using the labels of the purchased financial products as outputs of the neural collaborative filtering model based on deep learning, training the neural collaborative filtering model based on deep learning, and during the model training process, using an optimizer to update the model parameters, calculating the loss function through a back propagation algorithm, and obtaining a trained target model when the loss function meets a preset convergence condition; obtaining target data for describing the target object, and determining the target financial product corresponding to the target data using the target model; determining the risk level corresponding to the target financial product using the deep learning model, and sending the target financial product and the risk level to the target object.

[0006] Optionally, constructing a customer profile of the target object includes: step S21, obtaining first information related to the first object from an internal database of a financial institution, wherein the first information includes: account information, transaction flow data, and credit record information, the account information includes: account opening time, account type, and account balance, the transaction flow data includes: transaction time, amount, counterparty, and transaction type, and the credit record information includes: loan amount, loan term, repayment method, repayment record, and overdue situation; step S22, obtaining basic enterprise information of the first object from a small and micro enterprise registration department, wherein the basic enterprise information includes: registration information, equity structure information, and enterprise change record information, wherein the registration information includes: registered capital, registered address, business scope, and establishment time; obtaining credit rating data of the first object from a third-party credit rating agency; obtaining store operation data of the first object having online business from an e-commerce platform; Step S23, extracting features from the first information, basic enterprise information, credit rating data, and store operation data to obtain multiple first features corresponding to the first object; Step S24, traversing each of the multiple first features, and calculating the number of data points in a neighborhood centered on the first target feature and with the first neighborhood radius as the radius; if the number of data points in the neighborhood is greater than or equal to the first minimum number of points, marking the first target feature as a core point; Step S25, creating a first empty cluster set, wherein all data points in the first empty cluster set are marked as unvisited; traversing the data points in the first empty cluster set, adding the first data point to the first cluster set, and marking it as visited, wherein the first data point is Core point; for all unvisited first target data points in the neighborhood of the first data point, if the first target data point is a core point, add it to the first cluster set and mark it as visited; if the first target data point is a boundary point, add it to the first cluster set and mark it as visited; if the first target data point is a noise point, mark it; step S26, repeat step S25 until all data points in the neighborhood of the first data point have been visited; step S27, repeat steps S24 to S26 until all core points in the multiple first features have been added to the cluster set or marked as noise points, and multiple first target cluster sets are obtained, and the multiple first target cluster sets are determined as multiple first portraits.

[0007] Optionally, before traversing each first feature of multiple first features, the method also includes: determining a first neighborhood radius based on financial indicators, operating indicators, and capital demand indicators, wherein the financial indicators include: asset size, revenue status information, profit level information; the operating indicators include: industry category, years of operation; the capital demand indicators include: loan amount demand information, capital turnover cycle information; and determining a first minimum point number based on credit indicators, wherein the credit indicators include: credit score, overdue record.

[0008] Optionally, constructing a customer profile of the target object includes: step S41, obtaining second information related to the second object from the internal database of the financial institution, wherein the second information includes: account information, deposit information, account information includes: account opening time, account type, account balance, and deposit information includes: deposit type, deposit amount, and deposit period; step S42, obtaining third information of the second object, wherein the third information includes: land contract information, agricultural subsidy information, agricultural product output and quality data, land contract information includes: land area, contract period, land use, and agricultural subsidy information includes: subsidy type, subsidy amount, and issuance time; step S43, performing feature extraction on the second information and the third information to obtain multiple second features corresponding to the second object; step S44, traversing each second feature of the multiple second features, and calculating the number of data points in a neighborhood with the second target feature as the center and the second neighborhood radius as the radius; if the number of data points in the neighborhood is greater than or equal to the second minimum number of points, marking the second target feature as a core. center point; step S45, create a second empty cluster set, wherein all data points in the second empty cluster set are marked as unvisited; traverse the data points in the second empty cluster set, add the second data point to the second cluster set, and mark it as visited, wherein the second data point is a core point; for all unvisited second target data points in the neighborhood of the second data point, if the second target data point is a core point, add it to the second cluster set and mark it as visited; if the second target data point is a boundary point, add it to the second cluster set and mark it as visited; if the second target data point is a noise point, mark it; step S46, repeat step S45 until all data points in the neighborhood of the second data point have been visited or marked as noise points; step S47, repeat steps S44 to S46 until all core points in multiple second features have been added to the cluster set or marked as noise points, and multiple second target cluster sets are obtained, and the multiple second target cluster sets are determined as multiple second portraits.

[0009] Optionally, before traversing each of the multiple second features, the method also includes: determining a second neighborhood radius based on agricultural resource utilization characteristics, wherein the agricultural resource utilization characteristics include: irrigation and land quality characteristics, agricultural equipment characteristics; determining a second minimum number of points based on agricultural production characteristics and economic income characteristics, wherein the agricultural production characteristics include: planting or breeding scale characteristics, crop or breeding variety type characteristics, and the economic income characteristics include: annual income level.

[0010] Optionally, the target financial product corresponding to the target data is determined using the target model, including: converting various features in the target data according to the processing method of the training set to generate a customer feature vector in the same format as the training data; inputting the customer feature vector and the feature vectors of all the financial products to be recommended in the financial institution into the target model respectively, the target model performs nonlinear transformation on the input features through a multi-layer neural network, learns the potential relationship between the customer feature vector and the financial products to be recommended, outputs the recommendation score of each financial product to be recommended, and determines the financial products to be recommended with the top n recommendation scores as the target financial products, where n is a positive integer.

[0011] Optionally, the deep learning model is obtained by training through the following method: initialize the random forest model, set the model parameters according to the number of target objects in the area where the financial institution is located and the number of feature dimensions in the customer portrait, wherein the model parameters include: the number of trees, the maximum depth, the minimum number of sample splits, and the minimum number of leaf node samples; for each decision tree, use the bootstrap sampling method to extract samples with replacement from all features in the customer portrait to construct multiple different training subsets; use multiple different training subsets to train the random forest model, each decision tree is independently trained on its own training subset, and when the node is split, the optimal split attribute is selected according to the information gain; when the evaluation index of the model parameters meets the preset conditions, a deep learning model is obtained.

[0012] According to another aspect of the present application, a customer service system based on machine learning technology is also provided, including: a construction module, used to construct a customer portrait of a target object, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: rural areas, farmers and rural populations; a training module, used to use the customer portrait and the purchased financial products corresponding to the customer portrait as inputs of a neural collaborative filtering model based on deep learning, and the labels of the purchased financial products as outputs of the neural collaborative filtering model based on deep learning, to train the neural collaborative filtering model based on deep learning, and during the model training process, use an optimizer to update the model parameters, calculate the loss function through a back propagation algorithm, and obtain a trained target model when the loss function meets a preset convergence condition; a first determination module, used to obtain target data for describing the target object, and use the target model to determine the target financial product corresponding to the target data; a second determination module, used to use the deep learning model to determine the risk level corresponding to the target financial product, and send the target financial product and the risk level to the target object.

[0013] According to another aspect of the present application, a non-volatile storage medium is also provided, the storage medium includes a stored program, wherein when the program runs, the device where the storage medium is located is controlled to execute the above customer service method based on machine learning technology.

[0014] According to another aspect of the present application, an electronic device is provided, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the above customer service method based on machine learning technology is executed when the program is running.

[0015] According to yet another aspect of the present application, a computer program is provided, wherein when the computer program is executed by a processor, the above customer service method based on machine learning technology is implemented.

[0016] According to another aspect of the present application, a computer program product is provided, which includes a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above customer service method based on machine learning technology is implemented.

[0017] In the present application, a customer profile of a target object is constructed, wherein the target objects include: a first object and a second object, wherein the first object includes: small and micro enterprises, and the second object includes: rural areas, farmers and rural populations; the customer profile and the purchased financial products corresponding to the customer profile are respectively used as inputs of a neural collaborative filtering model based on deep learning, and the labels of the purchased financial products are used as outputs of the neural collaborative filtering model based on deep learning, and the neural collaborative filtering model based on deep learning is trained. During the model training process, an optimizer is used to update the model parameters, and a loss function is calculated by a back propagation algorithm. When the loss function meets a preset convergence condition, a trained target model is obtained; target data used to describe the target object is obtained, and a target financial product corresponding to the target data is determined using the target model; a risk level corresponding to the target financial product is determined using a deep learning model, and the target financial product and the risk level are sent to the target object, thereby achieving the purpose of recommending corresponding financial products to different groups served by the bank based on artificial intelligence technology, thereby achieving the technical effect of improving operating efficiency, and further solving the technical problem of low operating efficiency caused by the inability of township banks to recommend corresponding financial products to different groups served by the bank based on artificial intelligence technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0019] Figure 1 is a flow chart of a customer service method based on machine learning technology according to an embodiment of the present application;

[0020] Figure 2 is a structural diagram of a customer service system based on machine learning technology according to an embodiment of the present application;

[0021] Figure 3 It is a hardware structure block diagram of a computer terminal of a customer service method based on machine learning technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present application.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] According to an embodiment of the present application, a method embodiment of a customer service method based on machine learning technology is provided. It should be noted that the steps shown in the flowchart of 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 can be executed in an order different from that shown here.

[0025] Figure 1 is a flowchart of a customer service method based on machine learning technology according to an embodiment of the present application, such as Figure 1 As shown, the method comprises the following steps:

[0026] Step S102, constructing a customer profile of the target object, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: the rural population.

[0027] Specifically, for small and micro enterprises, we obtain their financial data (assets, income, profits, liabilities, etc.), operating data (number of employees, years of operation, market share, etc.), credit data (credit rating, loan history, overdue records, etc.) and industry data (industry development trends, competition) from multiple data sources. We clean the data, remove outliers and missing values, and then standardize the data to unify the dimensions.

[0028] The DBSCAN (density-based clustering) algorithm is used to cluster the data of small and micro enterprises in combination with the neighborhood radius and the first minimum number of points parameters. For example, the neighborhood radius is set to 50 units and the first minimum number of points is set to 10. The core points, boundary points and noise points are analyzed to form several clusters. For each cluster, the commonalities and differences of small and micro enterprises in financial, operational, credit and industry characteristics are analyzed to build a customer profile. For example, one cluster may contain small and micro enterprises with good financial conditions, while another cluster may contain enterprises with higher credit risks.

[0029] For the rural population, collect information on age, gender, family structure, education, planting and breeding information, economic income, financial needs and policy dependence. Perform data cleaning and standardization to ensure data quality. Use the same DBSCAN algorithm (or select other clustering algorithms based on data characteristics) to cluster the rural population data with appropriate neighborhood radius and minimum number of points. For example, set the neighborhood radius to 30 units and the minimum number of points to 5 to analyze various types of planting or breeding professional households, part-time farmers and low-income farmers. Analyze the basic characteristics, agricultural production, economic income, financial needs and policy dependence of the rural population in each cluster to build a customer profile.

[0030] Step S104, taking the customer portrait and the purchased financial products corresponding to the customer portrait as the input of the neural collaborative filtering model based on deep learning, and taking the labels of the purchased financial products as the output of the neural collaborative filtering model based on deep learning, training the neural collaborative filtering model based on deep learning, and during the model training process, using the optimizer to update the model parameters, and calculating the loss function through the back propagation algorithm, and obtaining the target model that has completed the training when the loss function meets the preset convergence conditions.

[0031] The customer profiles of small and micro enterprises and rural groups and the labels of the financial products they have purchased are used as input data. The financial product labels reflect the target objects' needs and preferences for financial products. A neural collaborative filtering model is created using a deep learning framework. The input layer of the model receives customer profiles and financial product labels, the hidden layer contains multiple neurons, and the output layer predicts financial products. The input data and output labels are input into the model for training, and the model parameters are updated using the back propagation algorithm and optimizer until the loss function converges. For example, the loss function can be mean square error or cross entropy loss, and the preset convergence condition may be that the change in the loss function is less than 0.001.

[0032] In step 104, the target model may be loaded into the memory, for example, the raw data of the target model may be loaded from the non-volatile memory into the volatile memory, so that the processor runs the target model. The raw data of the target model refers to unprocessed data, usually including parameters and structural data of the target model, and the structural data may be a calculation relationship based on the parameters, such as a forward propagation calculation relationship between intermediate layers and neurons. Specifically, the structural data may include code related to the structure of the target model, such as code for performing related calculations between intermediate layers and neurons.

[0033] In one embodiment, an area for loading the target model can be divided in the memory, which may include a structure data storage area and a parameter storage area. The structure data storage area is used to store structure-related codes, and the parameters referenced by it can point to the address of a specific parameter in the parameter storage area through a pointer. During the training process of the target model, it may be necessary to frequently update the parameters, and the parameter values ​​in the parameter storage area can be updated.

[0034] Step S106, obtaining target data for describing the target object, and using the target model to determine the target financial product corresponding to the target data.

[0035] The financial institution first collects a series of detailed information about a specific farmer (i.e., the target object), which constitutes the target data. For example, the farmer is 45 years old, has a good credit score, owns 100 mu of land, and mainly grows rice; has an annual income of 150,000 yuan, has 80% complete irrigation facilities, and receives an annual subsidy of 3,000 yuan in agricultural subsidies; and has account information at the financial institution, including the account opening date of 2015, the account type of a comprehensive account, and the recent average account balance of 50,000 yuan.

[0036] In order to input the collected data into the model, we need to preprocess the data. For example, we standardize the numerical features such as the farmer's age, credit score, land area, and annual income to ensure that the numerical range of these features is consistent and avoid the impact of dimensional differences between features. For categorical features, such as the main types of crops, we use one-hot encoding to convert them into numerical representations to facilitate model recognition and processing.

[0037] The preprocessed target data is input into the previously trained target model. This model is a random forest model, which is used to predict the most suitable financial product type and risk level based on the comprehensive characteristics of farmers.

[0038] The model integrates the prediction results of multiple decision trees and outputs a list of financial products that the farmer may be interested in or suitable for. For example, the model predicts that the farmer may be interested in the "agricultural planting loan" product and recommends the "agricultural insurance" product.

[0039] Step S108, using the deep learning model to determine the risk level corresponding to the target financial product, and sending the target financial product and risk level to the target object.

[0040] Using the deep learning risk assessment model, the target financial product and target object data are input, and the corresponding risk level of the product is output. The model can be evaluated based on the target object's historical credit record, industry risk, and macroeconomic indicators. The recommended financial products and their risk levels are sent to the target object based on the target object's contact information, such as phone, email, or SMS. At the same time, detailed financial product introductions and risk warnings are provided to promote informed decision-making.

[0041] According to the above steps, a customer profile of the target object is constructed, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: the rural groups; the customer profile and the purchased financial products corresponding to the customer profile are respectively used as inputs of the neural collaborative filtering model based on deep learning, and the labels of the purchased financial products are used as outputs of the neural collaborative filtering model based on deep learning, and the neural collaborative filtering model based on deep learning is trained. During the model training process, the optimizer is used to update the model parameters, and the loss function is calculated by the back propagation algorithm. When the loss function meets the preset convergence conditions, the trained target model is obtained; target data for describing the target object is obtained, and the target financial product corresponding to the target data is determined by the target model; the risk level corresponding to the target financial product is determined by the deep learning model, and the target financial product and the risk level are sent to the target object, so as to achieve the purpose of recommending corresponding financial products for different groups served by it based on artificial intelligence technology, thereby realizing the technical effect of improving operating efficiency.

[0042] The following Figure 1 The steps shown are exemplary and explanatory.

[0043] According to some optional embodiments of the present application, constructing a customer profile of the target object can be achieved by the following method: Step S21, obtaining first information related to the first object from the internal database of the financial institution, wherein the first information includes: account information, transaction flow data, and credit record information; the account information includes: account opening time, account type, and account balance; the transaction flow data includes: transaction time, amount, counterparty, and transaction type; the credit record information includes: loan amount, loan term, repayment method, repayment record, and overdue situation; Step S22, obtaining the basic enterprise information of the first object from the small and micro enterprise registration department, wherein the basic enterprise information includes: registration information, equity structure information, and enterprise change record information, wherein the registration information includes: registered capital, registered address, business scope, and establishment time; obtaining the credit rating data of the first object from a third-party credit rating agency; obtaining the store operation data of the first object with online business from the e-commerce platform; Step S23, performing feature extraction on the first information, basic enterprise information, credit rating data, and store operation data to obtain multiple first features corresponding to the first object; Step S24, traversing each of the multiple first features, and calculating The number of data points in a neighborhood with the first target feature as the center and the first neighborhood radius as the radius; if the number of data points in the neighborhood is greater than or equal to the first minimum number of points, the first target feature is marked as a core point; step S25, creating a first empty cluster set, wherein all data points in the first empty cluster set are marked as unvisited; traversing the data points in the first empty cluster set, adding the first data point to the first cluster set, and marking it as visited, wherein the first data point is a core point; for all unvisited first target data points in the neighborhood of the first data point, if the first target data point is Core points, add it to the first cluster set and mark it as visited. If the first target data point is a boundary point, add it to the first cluster set and mark it as visited. If the first target data point is a noise point, mark it; step S26, repeat step S25 until all data points in the neighborhood of the first data point have been visited; step S27, repeat steps S24 to S26 until all core points in the multiple first features have been added to the cluster set or marked as noise points, and multiple first target cluster sets are obtained, and the multiple first target cluster sets are determined as multiple first portraits.

[0044] For example, in step S21, the account information of small and micro enterprises is obtained from the internal database of the township bank (such as the account opening time is 2020, the account type is a checking account, and the account balance is 50,000 yuan), transaction flow data (such as February 1, 2023, the transaction amount is 2,000 yuan, the counterparty is ABC Company, and the transaction type is payment for goods), and credit record information (such as a loan of 100,000 yuan in 2021, a term of 2 years, monthly repayment, good repayment record, and no overdue situation).

[0045] Step S22, obtain the registration information of small and micro enterprises from the Industrial and Commercial Administration Bureau (e.g., registered capital of RMB 1 million, registered address in a township industrial zone, business scope including agricultural product processing, and establishment date in 2019). Obtain credit rating data of small and micro enterprises from the National Enterprise Credit Information Publicity System or professional credit rating agencies, such as AAA. Obtain store operation data of small and micro enterprises from e-commerce platforms, such as monthly sales, customer reviews, online time, etc.

[0046] Step S23, extracting features from the collected data to obtain multiple features of small and micro enterprises, such as financial health, business scale, credit rating, online business activity, etc.

[0047] Step S24, traverse all the extracted first features, take "financial health status" as an example, set the neighborhood radius to 10 units, and the first minimum number of points to 15. If there are more than 15 data points in the neighborhood of the financial health status feature value of a small and micro enterprise, it will be marked as a core point.

[0048] Step S25-26, create a first empty cluster set, and mark all small and micro enterprise data points as unvisited. Traverse the data points in the first empty cluster set, find the first core point (such as a small and micro enterprise with good financial health), add it to the first cluster set, and mark it as visited. Check all unvisited data points in the neighborhood of the core point. If there are other small and micro enterprise data points (i.e., the first target data point) in the neighborhood and they also have high financial health, add them to the same cluster; if the financial health of these data points is lower than that of the core point, but they have an intersection with the neighborhood of the core point, add them to the cluster as boundary points; if the data point is far away from the core point and exists alone, it is marked as a noise point. All data points in the neighborhood are marked as visited, and the above execution steps are repeated until all data points in the neighborhood have been visited.

[0049] Step S27, repeat steps S24 to S26 until all core points are added to the corresponding clusters or marked as noise points, forming multiple first target cluster sets. Analyze the characteristics of each cluster set, such as financial health, industry type, credit rating, etc., to form different customer group portraits of small and micro enterprises. Each cluster set is used as part of the first portrait for subsequent financial product recommendations and risk assessment.

[0050] Preferably, before traversing each first feature of the multiple first features, the following steps can also be performed: determine the first neighborhood radius based on financial indicators, operating indicators, and capital demand indicators, wherein the financial indicators include: asset size, revenue status information, profit level information; the operating indicators include: industry category, years of operation; the capital demand indicators include: loan amount demand information, capital turnover cycle information; determine the first minimum point number based on the credit indicators, wherein the credit indicators include: credit score, overdue record.

[0051] In density-based clustering algorithms, the first neighborhood radius (Eps) and the first minimum number of points (MinPts) are two key parameters that define the concepts of "density" and "neighborhood" in the data space, thereby determining which data points can be considered to belong to the same cluster.

[0052] The role of the first neighborhood radius (Eps): 1. Define the neighborhood: The Eps parameter defines the size of the neighborhood centered on a data point. If another data point is located within a sphere with Eps as the radius and the current point as the center, then these two points are considered to be points in the neighborhood. 2. Density threshold: Eps and MinPts together define the "density" of data points. If a point has at least MinPts other points within the Eps radius around it, then this point is considered to be part of the "high-density" area and can be regarded as a core point. 3. Cluster formation: Smaller Eps values ​​mean that closer data points are considered neighbors, thus forming smaller and more compact clusters; larger Eps values ​​may form larger and looser clusters. The choice of Eps value depends on the distribution of the data and the clustering target.

[0053] The role of the first minimum number of points (MinPts): 1. Density measurement: The MinPts parameter is used to measure the "neighborhood density" around a data point. If there are at least MinPts other points within the Eps radius around a point, then this point is considered a core point, indicating that it is in a "high-density" area. 2. Cluster identification: The larger the MinPts value, the more points the identified clusters will contain, and the clusters will be more strict, only containing truly high-density areas. Conversely, a smaller MinPts value may identify more and smaller clusters, and may even mistake noise points for part of a cluster. 3. Filtering of noise points: When there are not enough points within the Eps radius around a data point (less than MinPts), then these points are considered noise points and will not be added to any cluster. Setting the MinPts parameter helps filter out isolated points or noise data.

[0054] Assuming there is a set of data on small and micro enterprises, we first analyze the asset size, revenue and profit, and find that the asset size is concentrated between 2 million and 8 million, the revenue is between 300,000 and 1.5 million, and the profit is between 30,000 and 100,000, and the distribution is relatively even. At the same time, it is observed that most small and micro enterprises have been in operation for 3 to 8 years, and the industry categories involve agriculture, manufacturing and services, with varying degrees of competition. In terms of capital demand, the loan amount demand is generally between 200,000 and 1 million, and the capital turnover cycle is mostly 1 month to 3 months. Based on these analyses, the first neighborhood radius can be set to 80 units to include the fluctuation range of the above financial and operating indicators.

[0055] For credit scores, it is found that most small and micro enterprises have a credit score of 700 to 800 points and have few overdue records, which indicates that the overall credit quality of enterprises is high. Considering that there are many enterprises in the high credit score range, in order to more finely identify small and micro enterprises with good credit quality, the first minimum number of points can be set to a higher value, such as 25 points, to ensure that there are enough high-credit enterprise samples in the cluster.

[0056] Through the above embodiments, township banks can systematically build customer profiles of small and micro enterprises. Based on these profiles, banks can better understand the characteristics and needs of different small and micro enterprises, and thus provide more accurate financial products and services while conducting effective risk management.

[0057] In some optional embodiments of the present application, constructing a customer profile of the target object can be achieved by the following method: step S41, obtaining second information related to the second object from the internal database of the financial institution, wherein the second information includes: account information, deposit information, account information includes: account opening time, account type, account balance, deposit information includes: deposit type, deposit amount, deposit period; step S42, obtaining third information of the second object, wherein the third information includes: land contract information, agricultural subsidy information, agricultural product output and quality data, land contract information includes: land area, contract period, land use, agricultural subsidy information includes: subsidy type, subsidy amount, issuance time; step S43, extracting features from the second information and the third information to obtain multiple second features corresponding to the second object; step S44, traversing each second feature of the multiple second features, and calculating the number of data points in a neighborhood centered on the second target feature and with the second neighborhood radius as the radius; if the number of data points in the neighborhood is greater than or equal to the second minimum number of points, the first The second target feature is marked as a core point; step S45, creating a second empty cluster set, wherein all data points in the second empty cluster set are marked as unvisited; traversing the data points in the second empty cluster set, adding the second data point to the second cluster set, and marking it as visited, wherein the second data point is a core point; for all unvisited second target data points in the neighborhood of the second data point, if the second target data point is a core point, add it to the second cluster set and mark it as visited; if the second target data point is a boundary point, add it to the second cluster set and mark it as visited; if the second target data point is a noise point, mark it; step S46, repeating step S45 until all data points in the neighborhood of the second data point have been visited or marked as noise points; step S47, repeating steps S44 to S46 until all core points in multiple second features have been added to cluster sets or marked as noise points, obtaining multiple second target cluster sets, and determining the multiple second target cluster sets as multiple second portraits.

[0058] For example, in step S41, the second information related to the "Three Rural Areas" group is obtained from the financial institution; the account information of the "Three Rural Areas" group is obtained from the internal database of the township bank, for example, the account opening time of a certain farmer is 2018, the account type is a savings account, and the account balance is 30,000 yuan. Deposit information is obtained, including deposit type (such as time deposit), deposit amount (such as 50,000 yuan), deposit period (such as one year), etc. This information is helpful to understand the fund management and savings habits of the "Three Rural Areas" group.

[0059] Step S42, obtain agricultural-related data of the three rural groups from external sources. Obtain land contract information from the agricultural department. For example, a farmer has contracted 20 mu of land for a period of 10 years, and the land is used to grow rice. Obtain agricultural subsidy information, including subsidy type (such as grain planting subsidy), subsidy amount (such as 1,000 yuan per year), issuance time (such as October each year), etc., which is helpful for analyzing the policy dependence of the three rural groups. Obtain agricultural product output and quality data from the agricultural monitoring system, such as a farmer's annual rice output of 4,000 kilograms, and the quality grade is excellent.

[0060] Step S43, feature extraction performs feature extraction on the collected account information, deposit information, land contract information, agricultural subsidy information, and agricultural product output and quality data to obtain multiple second features of the three rural groups, such as account activity, savings preference, land size, agricultural product market competitiveness, etc.

[0061] Step S44, traverse all second features, take deposit information as an example, set the second neighborhood radius to 50 units, and the second minimum number of points to 10. If there are more than 10 data points in the neighborhood of the deposit information feature value of a farmer, it will be marked as a core point.

[0062] Step S45, create a second empty cluster set, and mark all data points of the three rural groups as unvisited. Traverse the data points in the second empty cluster set, find the first core point (such as farmers with stable deposit habits), add it to the second cluster set, and mark it as visited. Check all unvisited data points in the neighborhood of the core point. If it is a core point or a boundary point, include it in the same cluster and mark it as visited; if it is a noise point, mark it to indicate that it does not belong to any cluster.

[0063] Step S46, repeating step S45 until all data points in the neighborhood of the core point have been visited or marked as noise points.

[0064] Step S47, repeat steps S44 to S46 until all core points have been added to the cluster set or marked as noise points, and finally form multiple second target cluster sets, that is, customer portraits of the three rural groups.

[0065] For example, for deposit information, if it is found that there are a large number of farmers with a generally long deposit period (such as 1 year or more), and the deposit amounts of these farmers are relatively concentrated between 10,000 and 100,000, the second neighborhood radius can be set to 50 units. In this way, farmers with at least 10 points (the second minimum number of points is set to 10) in the neighborhood will be identified as core points, thereby forming a cluster of rural groups with similar deposit preferences.

[0066] Preferably, before traversing each of the multiple second features, the following steps can also be performed: determine the second neighborhood radius based on agricultural resource utilization characteristics, wherein the agricultural resource utilization characteristics include: irrigation and land quality characteristics, agricultural equipment characteristics; determine the second minimum number of points based on agricultural production characteristics and economic income characteristics, wherein the agricultural production characteristics include: planting or breeding scale characteristics, crop or breeding variety type characteristics, and the economic income characteristics include: annual income level.

[0067] Suppose that the farmer data of a certain township is analyzed and it is found that the irrigation facilities are complete and the land quality level is mainly concentrated in level II. In terms of agricultural equipment, most farmers have basic tractors and harvesters, which indicates that agricultural resource utilization is relatively efficient. Therefore, the second neighborhood radius is set to a smaller value, such as 50 units. For agricultural production characteristics, it is observed that the planting scale is mainly between 10 mu and 30 mu, and the types of crops planted are relatively single, mainly rice and wheat, which indicates that agricultural production has certain commonalities. Considering that the annual income level of farmers is mostly between 30,000 yuan and 80,000 yuan, and is relatively concentrated. Based on these characteristics, the second minimum number of points can be set to a higher value, such as 20 points, to ensure that the farmers included in the cluster have similar planting scales, crop types and income levels, which is convenient for financial institutions to design financial products that meet the needs of this group.

[0068] In some optional embodiments of the present application, the target financial product corresponding to the target data is determined by using the target model, which can be achieved by the following method: converting each feature in the target data according to the processing method of the training set to generate a customer feature vector in the same format as the training data; inputting the customer feature vector and the feature vectors of all the financial products to be recommended in the financial institution into the target model respectively, the target model performs nonlinear transformation on the input features through a multi-layer neural network, learns the potential relationship between the customer feature vector and the financial products to be recommended, outputs the recommendation score of each financial product to be recommended, and determines the financial products to be recommended with the top n recommendation scores as the target financial products, where n is a positive integer.

[0069] Assume that the target dataset contains the characteristics of the three rural groups, such as land area, contract period, irrigation facilities, main crop types, annual income, credit rating, etc. First, these features are standardized. For example, the land area is converted from mu to a standard unit, and the credit rating is converted from 1 to 10 to a continuous value from 0 to 1 to ensure that the feature range is consistent with the training set. The converted features are integrated into a vector. For example, for a certain farmer, the generated feature vector may be [50 (standardized land area), 5 (contract period), 0.8 (perfection of irrigation facilities), 1 (main crop type code), 0.6 (standardized annual income), 0.9 (standardized credit rating)]. This vector has the same format as the vector in the training set, which is convenient for model input.

[0070] The above-generated feature vectors of the rural population and the feature vectors of all financial products to be recommended in financial institutions (such as loan product type, interest rate, loan term, application threshold, etc.) are respectively input into the previously trained target model (multi-layer neural network). The multi-layer neural network learns the complex relationship between input features through layer-by-layer nonlinear transformation. Each layer of neurons adjusts the weights to perform a weighted sum of the input features, and then performs a nonlinear transformation through an activation function (such as ReLU, Sigmoid, etc.) to extract a more advanced feature representation. In the hidden layer of the model, the neural network learns the potential association between the customer feature vector and the financial product feature vector, for example, the relationship between a farmer's credit rating and the application threshold of a loan product, or the relationship between the land area and the financing amount. The output layer of the model performs the final transformation on the input and generates a recommendation score for each financial product, indicating the matching degree of the product with a specific farmer. The higher the score, the more suitable the product is for the farmer.

[0071] Sort the recommendation scores of all financial products output by the model from high to low. Determine the target product: Select the top n financial products with the highest recommendation scores as the target financial products. For example, if n=5, the top five financial products with the highest recommendation scores will be determined as the target financial products for the farmer.

[0072] In summary, assume that the model training set already contains a large amount of data on rural groups and financial products, and the multi-layer neural network model has been successfully trained. When recommending financial products to a specific farmer, first collect and preprocess the farmer's data, for example, the land area is 50 mu, the credit rating is 9, the annual income is 60,000, etc. These are converted into feature vector format, such as [0.5, 5, 0.8, 1, 0.6, 0.9], and input into the neural network model. At the same time, the feature vectors of all loan products provided by financial institutions are input into the model. The vector of each loan product may include product type code, interest rate level, loan term, application threshold and other information. After multiple nonlinear transformations, the model outputs the recommendation score for each loan product. For example, the matching score of the farmer's feature vector with the "low threshold agricultural loan" product is 0.9, the score with the "long-term low interest loan" product is 0.8, and the scores with several other loan products are 0.75, 0.7, 0.65, etc. When n is set to 5, the top five loan products with the highest scores are selected as target financial products, namely, “low-threshold agricultural loans”, “long-term low-interest loans” and other five products, to provide personalized financial product recommendations for the farmer.

[0073] As other optional embodiments of the present application, the deep learning model is obtained by training through the following method: initialize the random forest model, set the model parameters according to the number of target objects in the area where the financial institution is located and the number of feature dimensions in the customer portrait, wherein the model parameters include: the number of trees, the maximum depth, the minimum number of sample splits, and the minimum number of leaf node samples; for each decision tree, use the bootstrap sampling method to extract samples with replacement from all features in the customer portrait to construct multiple different training subsets; use multiple different training subsets to train the random forest model, each decision tree is independently trained on its own training subset, and when the node is split, the optimal split attribute is selected according to the information gain; when the evaluation index of the model parameters meets the preset conditions, a deep learning model is obtained.

[0074] Assume that there are 10,000 farmers in total, and the characteristic dimensions involved in the customer portrait include but are not limited to land area, crop type, irrigation facility status, annual income, credit score, etc., a total of 20 characteristic dimensions.

[0075] According to the data scale and feature dimension, the random forest model is initialized and the following parameters are set: 1. The number of trees: In order to ensure the stability and prediction accuracy of the model, 500 trees can be set to be generated. 2. Maximum depth: In order to prevent overfitting and taking into account the number of feature dimensions, the maximum depth of each tree is set to 30. This can capture the complex relationships of the data while avoiding excessive complexity of the model. 3. In order to ensure that the decision of each node is based on a certain number of samples, the minimum number of sample splits is set to 100, which means that at least 100 samples are required when making a split decision at each node of the decision tree. 4. Minimum number of leaf node samples: In order to ensure that the decision of the leaf node is representative, the minimum number of leaf node samples is set to 50, that is, each leaf node contains at least 50 samples of data.

[0076] For each tree, bootstrap sampling is used to sample with replacement from the 10,000 farmer data included in the customer profile to construct their own training subset. For example, the training subset of the first tree may contain 7,000 farmer data, but the actual sample may exceed 7,000 because bootstrap sampling allows repeated sampling.

[0077] Using multiple different training subsets constructed, each tree is trained independently on its own subset. In this process, each tree will learn based on a different feature subset to increase the diversity and generalization ability of the model. When splitting a node, the optimal split attribute is selected based on indicators such as information gain. For example, on a tree, if a node contains samples with a large information gain on the "annual income" feature, then this feature will be selected as the split attribute of the node.

[0078] Evaluation metrics such as accuracy, recall, and F1 value were set to ensure that the model’s performance met the preset conditions. In this case, the goal was for the model to achieve at least 90% accuracy in predicting the financial risk level of the rural population. After initial model training, the model was found to have an accuracy of 87%, close to but not reaching the 90% target. To optimize the model, the number of trees, maximum depth, minimum sample splits, and minimum leaf node sample number were adjusted: Number of trees: increased from 500 to 700 to further improve the stability and prediction accuracy of the model. Maximum depth: reduced from 30 to 25 to reduce the risk of overfitting and improve the generalization ability of the model. Minimum sample splits: reduced from 100 to 50 to ensure that the model’s decision on features is more flexible and sensitive. Minimum leaf node sample number: kept unchanged at 50, but the performance of the model can be further monitored and fine-tuned if necessary.

[0079] Through the above specific implementation steps, financial institutions can use the random forest model to effectively predict the risk level based on the characteristic data of the rural population, providing strong support for financial product design and risk management.

[0080] Figure 2 is a structural diagram of a customer service system based on machine learning technology according to an embodiment of the present application, such as Figure 2 As shown, the system includes:

[0081] The construction module 22 is used to construct a customer profile of the target object, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: the rural groups.

[0082] The training module 24 is used to use the customer profile and the purchased financial products corresponding to the customer profile as the input of the neural collaborative filtering model based on deep learning, and the label of the purchased financial product as the output of the neural collaborative filtering model based on deep learning, to train the neural collaborative filtering model based on deep learning, and during the model training process, use the optimizer to update the model parameters, calculate the loss function through the back propagation algorithm, and obtain the target model that has completed the training when the loss function meets the preset convergence conditions.

[0083] The first determination module 26 is used to obtain target data for describing the target object, and determine the target financial product corresponding to the target data by using the target model.

[0084] The second determination module 28 is used to determine the risk level corresponding to the target financial product by using the deep learning model, and send the target financial product and the risk level to the target object.

[0085] Optionally, constructing a customer profile of the target object includes: step S21, obtaining first information related to the first object from an internal database of a financial institution, wherein the first information includes: account information, transaction flow data, and credit record information, the account information includes: account opening time, account type, and account balance, the transaction flow data includes: transaction time, amount, counterparty, and transaction type, and the credit record information includes: loan amount, loan term, repayment method, repayment record, and overdue situation; step S22, obtaining basic enterprise information of the first object from a small and micro enterprise registration department, wherein the basic enterprise information includes: registration information, equity structure information, and enterprise change record information, wherein the registration information includes: registered capital, registered address, business scope, and establishment time; obtaining credit rating data of the first object from a third-party credit rating agency; obtaining store operation data of the first object having online business from an e-commerce platform; Step S23, extracting features from the first information, basic enterprise information, credit rating data, and store operation data to obtain multiple first features corresponding to the first object; Step S24, traversing each of the multiple first features, and calculating the number of data points in a neighborhood centered on the first target feature and with the first neighborhood radius as the radius; if the number of data points in the neighborhood is greater than or equal to the first minimum number of points, marking the first target feature as a core point; Step S25, creating a first empty cluster set, wherein all data points in the first empty cluster set are marked as unvisited; traversing the data points in the first empty cluster set, adding the first data point to the first cluster set, and marking it as visited, wherein the first data point is Core point; for all unvisited first target data points in the neighborhood of the first data point, if the first target data point is a core point, add it to the first cluster set and mark it as visited; if the first target data point is a boundary point, add it to the first cluster set and mark it as visited; if the first target data point is a noise point, mark it; step S26, repeat step S25 until all data points in the neighborhood of the first data point have been visited; step S27, repeat steps S24 to S26 until all core points in the multiple first features have been added to the cluster set or marked as noise points, and multiple first target cluster sets are obtained, and the multiple first target cluster sets are determined as multiple first portraits.

[0086] Optionally, before traversing each first feature of multiple first features, the method also includes: determining a first neighborhood radius based on financial indicators, operating indicators, and capital demand indicators, wherein the financial indicators include: asset size, revenue status information, profit level information; the operating indicators include: industry category, years of operation; the capital demand indicators include: loan amount demand information, capital turnover cycle information; and determining a first minimum point number based on credit indicators, wherein the credit indicators include: credit score, overdue record.

[0087] Optionally, constructing a customer profile of the target object includes: step S41, obtaining second information related to the second object from the internal database of the financial institution, wherein the second information includes: account information, deposit information, account information includes: account opening time, account type, account balance, and deposit information includes: deposit type, deposit amount, and deposit period; step S42, obtaining third information of the second object, wherein the third information includes: land contract information, agricultural subsidy information, agricultural product output and quality data, land contract information includes: land area, contract period, land use, and agricultural subsidy information includes: subsidy type, subsidy amount, and issuance time; step S43, performing feature extraction on the second information and the third information to obtain multiple second features corresponding to the second object; step S44, traversing each second feature of the multiple second features, and calculating the number of data points in a neighborhood with the second target feature as the center and the second neighborhood radius as the radius; if the number of data points in the neighborhood is greater than or equal to the second minimum number of points, marking the second target feature as a core. center point; step S45, create a second empty cluster set, wherein all data points in the second empty cluster set are marked as unvisited; traverse the data points in the second empty cluster set, add the second data point to the second cluster set, and mark it as visited, wherein the second data point is a core point; for all unvisited second target data points in the neighborhood of the second data point, if the second target data point is a core point, add it to the second cluster set and mark it as visited; if the second target data point is a boundary point, add it to the second cluster set and mark it as visited; if the second target data point is a noise point, mark it; step S46, repeat step S45 until all data points in the neighborhood of the second data point have been visited or marked as noise points; step S47, repeat steps S44 to S46 until all core points in multiple second features have been added to the cluster set or marked as noise points, and multiple second target cluster sets are obtained, and the multiple second target cluster sets are determined as multiple second portraits.

[0088] Optionally, before traversing each of the multiple second features, the method also includes: determining a second neighborhood radius based on agricultural resource utilization characteristics, wherein the agricultural resource utilization characteristics include: irrigation and land quality characteristics, agricultural equipment characteristics; determining a second minimum number of points based on agricultural production characteristics and economic income characteristics, wherein the agricultural production characteristics include: planting or breeding scale characteristics, crop or breeding variety type characteristics, and the economic income characteristics include: annual income level.

[0089] Optionally, the target financial product corresponding to the target data is determined using the target model, including: converting various features in the target data according to the processing method of the training set to generate a customer feature vector in the same format as the training data; inputting the customer feature vector and the feature vectors of all the financial products to be recommended in the financial institution into the target model respectively, the target model performs nonlinear transformation on the input features through a multi-layer neural network, learns the potential relationship between the customer feature vector and the financial products to be recommended, outputs the recommendation score of each financial product to be recommended, and determines the financial products to be recommended with the top n recommendation scores as the target financial products, where n is a positive integer.

[0090] Optionally, the deep learning model is obtained by training through the following method: initialize the random forest model, set the model parameters according to the number of target objects in the area where the financial institution is located and the number of feature dimensions in the customer portrait, wherein the model parameters include: the number of trees, the maximum depth, the minimum number of sample splits, and the minimum number of leaf node samples; for each decision tree, use the bootstrap sampling method to extract samples with replacement from all features in the customer portrait to construct multiple different training subsets; use multiple different training subsets to train the random forest model, each decision tree is independently trained on its own training subset, and when the node is split, the optimal split attribute is selected according to the information gain; when the evaluation index of the model parameters meets the preset conditions, a deep learning model is obtained.

[0091] It should be noted that the above Figure 2 Each module in the can be a program module (for example, a set of program instructions that implement a specific function) or a hardware module. For the latter, it can be expressed in the following forms, but is not limited to this: the expression form of each of the above modules is a processor, or the functions of each of the above modules are implemented by a processor.

[0092] It should be noted that Figure 2 The preferred implementation of the illustrated embodiment can be found in Figure 1 The relevant description of the illustrated embodiment will not be repeated here.

[0093] Figure 3 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a customer service method based on machine learning technology. Figure 3As shown, the computer terminal 30 may include one or more (302a, 302b, ..., 302n are used to illustrate) processors 302 (the processor 302 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 304 for storing data, and a transmission module 306 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 3 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 3 More or fewer components as shown, or with Figure 3 Different configurations are shown.

[0094] It should be noted that the one or more processors 302 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 30. As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0095] The memory 304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the customer service method based on machine learning technology in the embodiment of the present application. The processor 302 executes various functional applications and data processing by running the software programs and modules stored in the memory 304, that is, realizing the above-mentioned customer service method based on machine learning technology. The memory 304 may include a high-speed random access memory, and may also include a 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 304 may further include a memory remotely arranged relative to the processor 302, and these remote memories may be connected to the computer terminal 30 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0096] The transmission module 306 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the computer terminal 30. In one example, the transmission module 306 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission module 306 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0097] The display may be, for example, a touch screen liquid crystal display (LCD), which may enable the target object to interact with the target object interface of the computer terminal 30 .

[0098] It should be noted that, in some optional embodiments, the above Figure 3 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. It should be noted that Figure 3 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0099] It should be noted that Figure 3 The computer terminal shown is used to execute Figure 1 The customer service method based on machine learning technology shown in the figure, therefore the relevant explanations and instructions in the execution method of the above command are also applicable to the electronic device and will not be repeated here.

[0100] An embodiment of the present application also provides a non-volatile storage medium, which includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the above customer service method based on machine learning technology.

[0101] A program for executing the following functions on a non-volatile storage medium: constructing a customer profile of a target object, wherein the target object includes: a first object and a second object, wherein the first object includes: small and micro enterprises, and the second object includes: rural areas, farmers and rural populations; using the customer profile and the purchased financial products corresponding to the customer profile as inputs of a neural collaborative filtering model based on deep learning, and using the labels of the purchased financial products as outputs of the neural collaborative filtering model based on deep learning, training the neural collaborative filtering model based on deep learning, and during the model training process, using an optimizer to update the model parameters, calculating the loss function through a back propagation algorithm, and obtaining a trained target model when the loss function meets a preset convergence condition; obtaining target data for describing the target object, and determining a target financial product corresponding to the target data using the target model; determining a risk level corresponding to the target financial product using the deep learning model, and sending the target financial product and the risk level to the target object.

[0102] An embodiment of the present application also provides an electronic device, including: a memory and a processor, the processor being used to run a program stored in the memory, wherein the above customer service method based on machine learning technology is executed when the program is running.

[0103] The processor is used to run a program that performs the following functions: construct a customer profile of a target object, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: rural areas, farmers and rural populations; using the customer profile and the purchased financial products corresponding to the customer profile as inputs of a neural collaborative filtering model based on deep learning, and using the labels of the purchased financial products as outputs of the neural collaborative filtering model based on deep learning, training the neural collaborative filtering model based on deep learning, and during the model training process, using an optimizer to update the model parameters, calculating the loss function through a back propagation algorithm, and obtaining a trained target model when the loss function meets a preset convergence condition; obtaining target data for describing the target object, and using the target model to determine the target financial product corresponding to the target data; using the deep learning model to determine the risk level corresponding to the target financial product, and sending the target financial product and the risk level to the target object.

[0104] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0105] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0106] In the above-mentioned embodiments of the present application, the collected information is information and data authorized by the target object or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary protection measures, do not violate public order and good morals, and provide corresponding operation entrances for the target object to choose to authorize or refuse.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0108] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0109] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0110] If the integrated unit is implemented in the form of 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 the present application, or the part that contributes to the relevant technology or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0111] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A customer service method based on machine learning technology, characterized in that: include: Constructing a customer group portrait of the target object, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: rural groups; The customer group portrait and the purchased financial products corresponding to the customer group portrait are respectively used as inputs of the neural collaborative filtering model based on deep learning, and the labels of the purchased financial products are used as outputs of the neural collaborative filtering model based on deep learning, and the neural collaborative filtering model based on deep learning is trained. During the model training process, the model parameters are updated using an optimizer, and the loss function is calculated using a back propagation algorithm. When the loss function meets a preset convergence condition, a target model that has completed the training is obtained; Acquire target data for describing the target object, and determine a target financial product corresponding to the target data using the target model; A deep learning model is used to determine the risk level corresponding to the target financial product, and the target financial product and the risk level are sent to the target object.

2. The method according to claim 1, characterized in that Build a customer profile of the target audience, including: Step S21, obtaining first information related to the first object from an internal database of a financial institution, wherein the first information includes: account information, transaction flow data, and credit record information; the account information includes: account opening time, account type, and account balance; the transaction flow data includes: transaction time, amount, transaction counterparty, and transaction type; the credit record information includes: loan amount, loan term, repayment method, repayment record, and overdue situation; Step S22, obtaining basic enterprise information of the first object from the small and micro enterprise registration department, wherein the basic enterprise information includes: registration information, equity structure information, and enterprise change record information, wherein the registration information includes: registered capital, registered address, business scope, and establishment time; obtaining credit rating data of the first object from a third-party credit rating agency; and obtaining store operation data of the first object with online business from an e-commerce platform; Step S23, performing feature extraction on the first information, the enterprise basic information, the credit rating data, and the store operation data to obtain a plurality of first features corresponding to the first object; Step S24, traversing each first feature of the plurality of first features, calculating the number of data points in a neighborhood centered on the first target feature and having a first neighborhood radius as a radius; if the number of data points in the neighborhood is greater than or equal to a first minimum number of points, marking the first target feature as a core point; Step S25, creating a first empty cluster set, wherein all data points in the first empty cluster set are marked as unvisited; traversing the data points in the first empty cluster set, adding the first data point to the first cluster set, and marking it as visited, wherein the first data point is a core point; for all unvisited first target data points in the neighborhood of the first data point, if the first target data point is a core point, adding it to the first cluster set and marking it as visited; if the first target data point is a boundary point, adding it to the first cluster set and marking it as visited; if the first target data point is a noise point, marking it; Step S26, repeating step S25 until all data points in the neighborhood of the first data point have been visited; Step S27, repeating steps S24 to S26 until all core points in the first features are added to the cluster set or marked as noise points, obtaining multiple first target cluster sets, and determining the first target cluster sets as multiple first portraits.

3. The method according to claim 2, characterized in that Before traversing each of the first features, the method further includes: Determine the first neighborhood radius according to financial indicators, operating indicators, and capital demand indicators, wherein the financial indicators include: asset scale, revenue status information, and profit level information; the operating indicators include: industry category and years of operation; and the capital demand indicators include: loan amount demand information and capital turnover cycle information; The first minimum number of points is determined according to credit indicators, wherein the credit indicators include: credit score and overdue record.

4. The method according to claim 1, characterized in that: Build a customer profile of the target audience, including: Step S41, obtaining second information related to the second object from an internal database of a financial institution, wherein the second information includes: account information and deposit information, the account information includes: account opening time, account type, account balance, and the deposit information includes: deposit type, deposit amount, and deposit term; Step S42, obtaining third information of the second object, wherein the third information includes: land contract information, agricultural subsidy information, agricultural product output and quality data, the land contract information includes: land area, contract period, land use, the agricultural subsidy information includes: subsidy type, subsidy amount, and issuance time; Step S43, performing feature extraction on the second information and the third information to obtain a plurality of second features corresponding to the second object; Step S44, traversing each of the plurality of second features, calculating the number of data points in a neighborhood centered on the second target feature and having a second neighborhood radius as a radius; if the number of data points in the neighborhood is greater than or equal to the second minimum number of points, marking the second target feature as a core point; Step S45, creating a second empty cluster set, wherein all data points in the second empty cluster set are marked as unvisited; traversing the data points in the second empty cluster set, adding the second data point to the second cluster set, and marking it as visited, wherein the second data point is a core point; for all unvisited second target data points in the neighborhood of the second data point, if the second target data point is a core point, adding it to the second cluster set and marking it as visited, if the second target data point is a boundary point, adding it to the second cluster set and marking it as visited, and if the second target data point is a noise point, marking it; Step S46, repeating step S45 until all data points in the neighborhood of the second data point have been visited or marked as noise points; Step S47, repeating steps S44 to S46 until all core points in the plurality of second features have been added to the cluster set or marked as noise points, obtaining a plurality of second target cluster sets, and determining the plurality of second target cluster sets as a plurality of second portraits.

5. The method according to claim 4, characterized in that Before traversing each of the plurality of second features, the method further includes: Determining the second neighborhood radius according to agricultural resource utilization characteristics, wherein the agricultural resource utilization characteristics include: irrigation and land quality characteristics, agricultural equipment characteristics; The second minimum number of points is determined based on agricultural production characteristics and economic income characteristics, wherein the agricultural production characteristics include: planting or breeding scale characteristics, crop or breeding variety type characteristics, and the economic income characteristics include: annual income level.

6. The method according to claim 1, characterized in that Determining a target financial product corresponding to the target data using the target model includes: Convert each feature in the target data according to the processing method of the training set to generate a customer group feature vector in the same format as the training data; The customer group feature vector and the feature vectors of all the financial products to be recommended in the financial institution are respectively input into the target model, and the target model performs nonlinear transformation on the input features through a multi-layer neural network, learns the potential relationship between the customer group feature vector and the financial products to be recommended, outputs the recommendation score of each of the financial products to be recommended, and determines the financial products to be recommended with the top n recommendation scores as the target financial products, wherein n is a positive integer.

7. The method according to claim 1, characterized in that The deep learning model is obtained by training through the following method: Initialize the random forest model, and set model parameters according to the number of target objects in the area where the financial institution is located and the number of feature dimensions in the customer profile, wherein the model parameters include: the number of trees, the maximum depth, the minimum number of sample splits, and the minimum number of leaf node samples; For each decision tree, use the bootstrap sampling method to extract samples with replacement from all the features in the customer profile to construct multiple different training subsets; The random forest model is trained using multiple different training subsets. Each decision tree is trained independently on its own training subset, and when a node is split, the optimal split attribute is selected based on information gain. When the evaluation index of the model parameters meets the preset conditions, the deep learning model is obtained.

8. A customer service system based on machine learning technology, characterized in that: include: A construction module is used to construct a customer group portrait of a target object, wherein the target object includes: a first object and a second object, the first object includes: small and micro enterprises, and the second object includes: rural groups; A training module, used to use the customer group portrait and the purchased financial products corresponding to the customer group portrait as inputs of the neural collaborative filtering model based on deep learning, and the labels of the purchased financial products as outputs of the neural collaborative filtering model based on deep learning, to train the neural collaborative filtering model based on deep learning, and during the model training process, to update the model parameters using an optimizer, to calculate the loss function using a back propagation algorithm, and to obtain a target model that has completed the training when the loss function meets a preset convergence condition; A first determination module is used to obtain target data for describing the target object, and determine a target financial product corresponding to the target data using the target model; The second determination module is used to determine the risk level corresponding to the target financial product by using a deep learning model, and send the target financial product and the risk level to the target object.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute the customer service method based on machine learning technology as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the customer service method based on machine learning technology described in any one of claims 1 to 7 is implemented.

Images