User loss risk identification method and device based on neural network, and electronic equipment

Abstract

The invention discloses a user loss risk identification method and device based on a neural network and electronic equipment, and relates to the field of artificial intelligence or other related technical fields, and the method comprises the steps: obtaining a financial data set of a target user at each time point; preprocessing the financial data set at each time point to obtain a user feature set of the target user at each time point; the user features in the user feature set of each time point are input to a loss risk identification model one by one, a loss risk identification result of the target user is output, the loss risk identification result at least comprises whether the target user has a loss risk, and the loss risk identification model calculates the loss probability of the target user to identify the loss risk of the target user. And after comparing the churn probability with a preset churn probability threshold, outputting a churn risk identification result of the target user. According to the method and the device, the technical problem of relatively low accuracy of the identification result obtained by identifying the loss risk of the user based on a rule matching mode in the related technology is solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence or other related technical fields, and in particular to a method and device for identifying user churn risk based on a neural network, and an electronic device. Background Art

[0002] At present, the financial industry is facing fierce market competition. Maintaining customer loyalty is crucial to the business development and profit growth of financial institutions. Based on this, the identification of customer churn risk has become a key link for financial institutions to optimize customer relationship management, improve service quality and economic benefits. However, traditional methods are unable to cope with the complex and changing financial environment, especially in processing massive customer data and capturing the dynamic changes in customer behavior.

[0003] In related technologies, the risk of customer churn is mainly identified based on pre-created matching rules. The rule-based identification method relies on the experience of industry experts and the analysis of historical data, and identifies the risk of customer churn by formulating a series of decision rules. This method is effective in simple situations, but in the face of a rapidly changing market environment and a diverse customer base, its prediction effect is often limited by the static nature and limitations of the rules, resulting in low accuracy of the identification results.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a method and device for identifying user churn risk based on a neural network, and an electronic device, so as to at least solve the technical problem in the related art that the accuracy of the identification result obtained by identifying the user churn risk based on a rule matching method is low.

[0006] According to one aspect of an embodiment of the present invention, a method for identifying user churn risk based on a neural network is provided, comprising: obtaining a financial data set of a target user at each time point, wherein the financial data set includes N financial data, and the financial data includes at least one of the following: transaction data, asset data, and interactive behavior data, where N is a positive integer; preprocessing the financial data set at each time point to obtain a user feature set of the target user at each time point; inputting the user features in the user feature set at each time point into a churn risk identification model one by one, and outputting a churn risk identification result of the target user, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-constructed based on an attention mechanism and a deep interest network, and the second churn risk identification sub-model is a model pre-constructed based on a long short-term memory network, and the churn risk identification model calculates the churn probability of the target user, and compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

[0007] Optionally, the user features in the user feature set at each time point are input one by one into the churn risk identification model, and the step of outputting the churn risk identification result of the target user includes: inputting the user features in the user feature set at each time point one by one into the first churn risk identification sub-model, and outputting the initial churn probability of the target user at each time point and the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model; inputting the initial churn probability and the user weighted feature vector set of the target user at each time point into the second churn risk identification sub-model, and outputting the churn probability of the target user; generating the churn risk identification result of the target user based on the churn probability of the target user, and outputting the churn risk identification result of the target user based on the output layer of the churn risk identification model, wherein, when the churn probability of the target user is greater than or equal to a preset churn probability threshold, it is determined that the target user has a churn risk.

[0008] Optionally, the user features in the user feature set at each time point are input one by one into the first churn risk identification sub-model, and the step of outputting the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model includes: for the user feature set at each time point, the user features in the user feature set are input one by one into the first churn risk identification sub-model, and each of the user features is encoded through the encoding layer of the first churn risk identification sub-model to obtain a user feature vector set; the attention weight feature vector of the user feature is calculated through the attention layer of the first churn risk identification sub-model; vector operations are performed on each user feature vector in the user feature vector set and the attention weight feature vector through the first churn risk identification sub-model to obtain a user weighted feature vector set, and the user weighted feature vector set is output by the first churn risk identification sub-model.

[0009] Optionally, after the first churn risk identification sub-model outputs the user weighted feature vector set, it also includes: inputting the user weighted feature vector set into the sum pooling layer of the first churn risk identification sub-model, and outputting the user feature code; obtaining the associated feature code and the target churn type feature code received by the input layer of the first churn risk identification sub-model, wherein the associated feature code is obtained based on the user portrait of the target user, and the target churn type feature code is obtained based on the historical churn users; based on the feature splicing layer of the first churn risk identification sub-model, the user feature code, the associated feature code and the target churn type feature code are feature spliced ​​to obtain a spliced ​​feature; inputting the spliced ​​feature into the multi-layer neural network of the first churn risk identification sub-model, and outputting the initial churn probability of the target user at each time point.

[0010] Optionally, the step of extracting the associated feature code includes: obtaining a user portrait of the target user pre-constructed in a financial institution; extracting associated features associated with user churn from the user portrait, wherein the associated features include at least one of the following: user attribute features, transaction environment features; encoding the associated features to obtain the associated feature code.

[0011] Optionally, the step of extracting the target churn type feature code includes: obtaining the financial data of the historical churn users within a historical time period; performing cluster analysis on the financial data of the historical churn users and the target user to obtain clustering results; extracting common features of the target user and the historical churn users from the clustering results, and determining the target churn type features of the target user based on the common features and the churn types of the historical churn users; encoding the target churn type features to obtain the target churn type feature code.

[0012] Optionally, the step of preprocessing the financial data at each time point to obtain a user feature set of the target user at each time point includes: performing data cleaning, standardization and normalization on the financial data at each time point to obtain the processed financial data; performing feature extraction on the processed financial data to obtain a user feature set of the target user at each time point, wherein the feature extraction includes time point feature extraction and time period feature extraction, the time point feature extraction represents the extraction of the behavioral features of the target user at a single time point, and the time period feature extraction represents the extraction of the behavioral feature data of the target user within a predetermined time period.

[0013] According to another aspect of an embodiment of the present invention, a user churn risk identification device based on a neural network is also provided, comprising: an acquisition unit, configured to acquire a financial data set of a target user at each time point, wherein the financial data set includes N financial data, and the financial data includes at least one of the following: transaction data, asset data, and interactive behavior data, where N is a positive integer; a processing unit, configured to pre-process the financial data set at each time point to obtain a user feature set of the target user at each time point; an output unit, configured to input user features in the user feature set at each time point into a churn risk identification model one by one, and output a churn risk identification result of the target user, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-constructed based on an attention mechanism and a deep interest network, and the second churn risk identification sub-model is a model pre-constructed based on a long short-term memory network, and the churn risk identification model calculates the churn probability of the target user, and compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

[0014] Optionally, the output unit includes: a first output sub-unit, used to input the user features in the user feature set at each time point into the first churn risk identification sub-model one by one, and output the initial churn probability of the target user at each time point and the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model; a second output sub-unit, used to input the initial churn probability and the user weighted feature vector set of the target user at each time point into the second churn risk identification sub-model, and output the churn probability of the target user; a third output sub-unit, used to generate a churn risk identification result of the target user based on the churn probability of the target user, and output the churn risk identification result of the target user based on the output layer of the churn risk identification model, wherein, when the churn probability of the target user is greater than or equal to a preset churn probability threshold, it is determined that the target user has a churn risk.

[0015] Optionally, the first output sub-unit includes: a first encoding module, for inputting the user features in the user feature set at each time point into the first churn risk identification sub-model one by one, encoding each of the user features through the encoding layer of the first churn risk identification sub-model to obtain a user feature vector set; a first calculation module, for calculating the attention weight feature vector of the user feature through the attention layer of the first churn risk identification sub-model; a first output module, for performing vector operations on each user feature vector in the user feature vector set and the attention weight feature vector through the first churn risk identification sub-model to obtain a user weighted feature vector set, and outputting the user weighted feature vector set by the first churn risk identification sub-model.

[0016] Optionally, the first output sub-unit includes: a second output module, used to input the user weighted feature vector set into the sum pooling layer of the first churn risk identification sub-model, and output the user feature code; a first acquisition module, used to obtain the associated feature code and the target churn type feature code received by the input layer of the first churn risk identification sub-model, wherein the associated feature code is obtained based on the user portrait of the target user, and the target churn type feature code is obtained based on the historical churn users; a first splicing module, used to perform feature splicing on the user feature code, the associated feature code and the target churn type feature code based on the feature splicing layer of the first churn risk identification sub-model to obtain a spliced ​​feature; a third output module, used to input the spliced ​​feature into the multi-layer neural network of the first churn risk identification sub-model, and output the initial churn probability of the target user at each time point.

[0017] Optionally, the neural network-based user churn risk identification device also includes: a first acquisition subunit, used to obtain a user portrait of the target user pre-constructed in a financial institution; a first extraction subunit, used to extract associated features associated with user churn from the user portrait, wherein the associated features include at least one of the following: user attribute features, transaction environment features; a first encoding subunit, used to encode the associated features to obtain the associated feature code.

[0018] Optionally, the neural network-based user churn risk identification device also includes: a second acquisition subunit, used to acquire the financial data of the historical churn users within a historical time period; a first analysis subunit, used to perform cluster analysis on the financial data of the historical churn users and the target users to obtain clustering results; a second extraction subunit, used to extract common features of the target user and the historical churn users from the clustering results, and determine the target churn type features of the target user based on the common features and the churn types of the historical churn users; a second encoding subunit, used to encode the target churn type features to obtain the target churn type feature encoding.

[0019] Optionally, the processing unit includes: a first processing sub-unit, used to perform data cleaning, standardization and normalization on the financial data at each time point to obtain the processed financial data; a first extraction sub-module, used to perform feature extraction on the processed financial data to obtain a user feature set of the target user at each time point, wherein the feature extraction includes time point feature extraction and time period feature extraction, the time point feature extraction represents the extraction of the behavioral features of the target user at a single time point, and the time period feature extraction represents the extraction of the behavioral feature data of the target user within a predetermined time period.

[0020] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is also provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the above-mentioned neural network-based user churn risk identification methods.

[0021] According to another aspect of an embodiment of the present invention, there is also provided an electronic device, comprising one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the above-mentioned neural network-based user churn risk identification methods.

[0022] In the present application, the following steps are performed: obtaining a financial data set of a target user at each time point, wherein the financial data set contains N financial data, and the financial data includes at least one of the following: transaction data, asset data, and interaction behavior data, where N is a positive integer; then preprocessing the financial data set at each time point to obtain a user feature set of the target user at each time point; finally, inputting the user features in the user feature set at each time point into a churn risk identification model one by one, and outputting a churn risk identification result of the target user, wherein the churn risk identification result at least includes: whether the target user has a churn risk; the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model; the first churn risk identification sub-model is a model pre-constructed based on an attention mechanism and a deep interest network; the second churn risk identification sub-model is a model pre-constructed based on a long short-term memory network; the churn risk identification model calculates the churn probability of the target user, and compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

[0023] In this application, the financial data of the target user at each time point is analyzed through a pre-built churn risk identification model. The churn risk identification model is composed of a first churn risk identification sub-model and a second churn risk identification sub-model. The deep interest network based on the first churn risk identification sub-model mines the risk features related to churn risk identification in the financial data, and the long short-term memory network based on the second churn risk identification sub-model learns the changing trend of user behavior, and then calculates the user churn probability based on the risk features and the changing trend of user behavior. The churn risk of the target user is accurately identified through the churn probability, achieving the purpose of quantitatively identifying the churn risk and achieving the technical effect of improving the accuracy of the user churn risk identification results. This solves the technical problem in the related art that the accuracy of the identification results obtained by identifying the churn risk of users based on the rule matching method is low. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0025] Figure 1 A hardware structure block diagram of a computer terminal (or mobile device) for implementing a user churn risk identification method based on a neural network is shown;

[0026] Figure 2 is a flow chart of an optional method for identifying user churn risk based on a neural network according to an embodiment of the present invention;

[0027] Figure 3is a schematic diagram of an optional neural network-based user churn risk identification process according to an embodiment of the present invention;

[0028] Figure 4 is a schematic diagram of an optional first churn risk identification sub-model according to an embodiment of the present invention;

[0029] Figure 5 is a schematic diagram of an optional neural network-based user churn risk identification device according to an embodiment of the present invention;

[0030] Figure 6 It is a hardware structure block diagram of an electronic device (or mobile device) that optionally executes a user churn risk identification method based on a neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention 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 interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] It should be noted that the user churn risk identification method and device based on neural network in the present application can be used in the field of artificial intelligence when the churn risk of target users is identified based on neural network, and can also be used in any field other than the field of artificial intelligence when the churn risk of target users is identified based on neural network. The application does not limit the application field of the user churn risk identification method and device based on neural network.

[0034] It should be noted that the collected information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of relevant data are in compliance with relevant laws, regulations and standards, necessary confidentiality measures are taken, and public order and good customs are not violated, and corresponding operation entrances are provided for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation entrances for users to choose to agree or refuse the results of automated decision-making; if the user chooses to refuse, the expert decision-making process will be entered.

[0035] The following embodiments of the present invention can be applied to various user churn risk identification systems / applications / devices based on neural networks. The present invention measures the importance of user features through a deep interest network and an attention mechanism, mines features related to churn risk, and identifies the changing trend of users within a certain period of time through a long short-term memory network, thereby predicting the probability of user churn, identifying the churn risk of target users based on the churn probability, achieving quantitative identification of churn risk, and improving the accuracy of churn risk identification results.

[0036] The present invention is described in detail below in conjunction with various embodiments.

[0037] Embodiment 1

[0038] According to an embodiment of the present invention, an embodiment of a method for identifying user churn risk based on a neural network 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.

[0039] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 The hardware structure block diagram of a computer terminal (or mobile device) for implementing a user churn risk identification method based on a neural network is shown. Figure 1As shown, the computer terminal 10 (or mobile device) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (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 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0040] It should be noted that the one or more processors 102 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 10 (or mobile device). As described 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).

[0041] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the user churn risk identification method based on a neural network in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the user churn risk identification method based on a neural network is realized. The memory 104 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 examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 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.

[0042] The transmission device 106 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 10. In one example, the transmission device 106 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 device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

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

[0044] Under the above operating environment, this application provides Figure 2 The user churn risk identification method based on neural network is shown, and the implementation body of this method is a user churn risk identification system based on neural network.

[0045] Figure 2 is a flowchart of an optional method for identifying user churn risk based on a neural network according to an embodiment of the present invention. Figure 2 As shown, the method comprises the following steps:

[0046] Step S201, obtaining a set of financial data of a target user at each time point.

[0047] It should be noted that the target users may be users registered with financial institutions. The risk of churn of users of financial institutions should be identified regularly to determine whether the target users are at risk of churn, and timely maintenance should be carried out to prevent user churn and improve user stickiness and loyalty.

[0048] It should be noted that financial data is the basis for identifying user churn risks. The financial data set of the target user at each time point is obtained. The financial data at each time point refers to taking a certain time point as an observation point and obtaining the accumulated financial data of the target user at that time point. Based on the financial data at each time point, the churn probability of the target user at each time point can be predicted. The churn probability has a direct impact on identifying user churn risks, thereby quickly and accurately identifying user churn risks.

[0049] It should be noted that the financial data set contains multiple financial data of the target users. The target users' transaction data, such as transaction type, transaction amount, number of profit and loss transactions divided by business scenarios, asset data, including the types of assets held by users, asset value, etc., and interactive behavior data, such as the frequency of users logging into financial applications and the frequency of users handling specific financial services, can be used to collect data from multiple dimensions and construct financial data.

[0050] Step S202, pre-processing the financial data set at each time point to obtain the user feature set of the target user at each time point.

[0051] It should be noted that after obtaining the financial data set of the target user at each time point, the financial data set at each time point needs to be preprocessed, including data cleaning, feature extraction and other operations, so as to obtain the user feature set of the target user at each time point.

[0052] Optionally, the step of preprocessing the financial data at each time point to obtain a user feature set of the target user at each time point includes: performing data cleaning, standardization and normalization on the financial data at each time point to obtain processed financial data; performing feature extraction on the processed financial data to obtain a user feature set of the target user at each time point, wherein feature extraction includes time point feature extraction and time period feature extraction, time point feature extraction means extracting the behavioral features of the target user at a single time point, and time period feature extraction means extracting the behavioral feature data of the target user within a predetermined time period.

[0053] It should be noted that preprocessing financial data sets can improve data quality and validity. Data cleaning is the first step in the preprocessing process, which aims to eliminate noise and inconsistency in the data and improve data quality. Specifically, it includes missing value processing, outlier processing and data consistency check. It is indeed only processing to perform statistical analysis on missing values ​​in transaction data and determine the filling strategy of missing values. If the proportion of missing values ​​is small, you can use adjacent values ​​or average values ​​to fill; if the proportion is large, you may need to consider deleting records with missing values ​​or using machine learning methods to predict missing values. Outlier detection can identify outliers in transaction amounts or account balances through statistical methods. The processing strategy for outliers should be based on the specific scenario and the nature of the outliers, and can be selected to be removed or corrected. Data consistency check is used to ensure the consistency of data formats, such as unifying the format of timestamps, checking and correcting data type errors (for example, a field that should be numeric is mistakenly identified as text). Data cleaning can significantly improve data quality, reduce prediction bias caused by data errors or inconsistencies, and provide a solid foundation for subsequent feature engineering and model analysis.

[0054] In addition to data cleaning, preprocessing also includes standardization and normalization, which aims to ensure that different types of financial data are compared on the same scale to avoid inappropriate impact of certain features on model prediction results due to large numerical ranges. Standardization can convert numerical features into standard normal scores with a mean of 0 and a standard deviation of 1, which is suitable for scenarios with large differences between features, such as transaction amounts and account balances. Normalization can scale the numerical range of features to between [0,1], which is suitable for scenarios where the numerical ranges of features are not much different but the scale needs to be consistent.

[0055] Feature extraction is the core step in preprocessing, which aims to extract the most valuable information for identifying customer churn risk from raw financial data. Feature extraction is divided into two aspects: point-in-time feature extraction and period-time feature extraction. Point-in-time feature extraction is to extract the user's immediate behavior features such as transaction amount, transaction type, account balance, and real-time interactive behavior features such as user login frequency and active status at a single time point. Period-time feature extraction is to extract behavioral feature data within a predetermined time period, such as the average number of transactions of the target in the past month, the volatility of the transaction amount, or the trend of user activity changes within a certain period. Through the comprehensive extraction of point-in-time features and period-time features, not only the immediate state of user behavior is captured, but also the historical trend and periodic law of user behavior are reflected, which provides a more comprehensive data perspective for model learning and helps to improve the recognition rate of potential churn customers.

[0056] Step S203: input the user features in the user feature set at each time point into the churn risk identification model one by one, and output the churn risk identification result of the target user.

[0057] It should be noted that the embodiment of the present invention uses a pre-built churn risk identification model to identify and analyze user characteristics to obtain churn risk identification results for target users. Based on the churn risk identification results, potential churn customers can be maintained in a timely manner to improve customer stickiness and ensure customer stability. The churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model. The output result of the first churn risk identification sub-model is the input data of the second churn risk identification sub-model. The first churn risk identification sub-model can calculate the churn probability of the target user at each time point. The second churn risk identification sub-model calculates the total churn probability of the target user based on the churn probability at each time point and the behavior change trend of the target user over a period of time, and determines whether the target user has a churn risk based on the churn probability, thereby achieving the purpose of quantitative identification of churn risk and improving the accuracy of churn risk results.

[0058] Optionally, the step of inputting user features in the user feature set at each time point into the churn risk identification model one by one, and outputting the churn risk identification result of the target user includes: inputting user features in the user feature set at each time point into the first churn risk identification sub-model one by one, and outputting the initial churn probability of the target user at each time point and the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model; inputting the initial churn probability of the target user at each time point and the user weighted feature vector set into the second churn risk identification sub-model, and outputting the churn probability of the target user; generating the churn risk identification result of the target user based on the churn probability of the target user, and outputting the churn risk identification result of the target user based on the output layer of the churn risk identification model, wherein, when the churn probability of the target user is greater than or equal to a preset churn probability threshold, it is determined that the target user has a churn risk.

[0059] The above-mentioned first churn risk identification sub-model is a pre-built model based on the attention mechanism and the deep interest network. The attention network can be used to mine the importance of each user feature to the churn risk identification, configure a weighted feature vector for the user feature, and quantify the weighted feature vector through a multi-layer neural network to calculate the churn probability of the target user at each time point. The weighted feature vector and the churn probability at each time point are used as the output of the first churn risk identification sub-model.

[0060] The second churn risk identification sub-model is a model pre-built based on the long short-term memory network. The output of the first churn risk identification sub-model is used as the input of the second churn risk identification sub-model. The second churn risk identification sub-model analyzes and quantifies the behavior change trend of the target user based on the weighted feature vector of the target user at each time point and the historical information stored in the model, and outputs the total churn probability of the target user in combination with the churn probability of the user at each time point.

[0061] The churn risk identification model is based on the total churn probability output by the second churn risk identification sub-model, compares it with the preset churn probability threshold, determines the churn risk of the target user according to the comparison result, generates the churn risk identification result of the target user, and outputs the churn risk identification result of the target user.

[0062] Optionally, the user features in the user feature set at each time point are input one by one into the first churn risk identification sub-model, and the step of outputting the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model includes: for the user feature set at each time point, the user features in the user feature set are input one by one into the first churn risk identification sub-model, and each user feature is encoded through the encoding layer of the first churn risk identification sub-model to obtain a user feature vector set; the attention weight feature vector of the user feature is calculated through the attention layer of the first churn risk identification sub-model; vector operations are performed on each user feature vector in the user feature vector set and the attention weight feature vector through the first churn risk identification sub-model to obtain a user weighted feature vector set, and the user weighted feature vector set is output by the first churn risk identification sub-model.

[0063] It should be noted that when analyzing the user feature set at each time point based on the first churn risk identification sub-model, for each time point, the user features in the user feature set are input into the first churn risk identification sub-model one by one, and the encoding layer and attention layer of the model process the user features at the same time. The encoding layer encodes each user feature and converts the original user features (such as transaction amount, number of logins, account balance, etc.) into a feature vector of fixed length, so as to facilitate the subsequent neural network to identify and process. The attention layer uses the attention mechanism to assign different weights to different user feature vectors, so that the model pays more attention to those features that are most relevant to the churn risk during the data processing process. Specifically, the attention layer compares the user feature vector set with the features of the target churn type (known churn customer behavior patterns), calculates the "attention" or relevance of each user feature vector to the target churn type, generates an attention weight feature vector, multiplies the attention weight feature vector with each user feature vector to obtain a user weighted feature vector, and then obtains a user weighted feature vector set corresponding to each user feature vector. The user weighted feature vector set will be used as one of the output data of the first churn risk identification sub-model, and will be subsequently input into the second churn risk identification sub-model.

[0064] Optionally, after the first churn risk identification sub-model outputs the user weighted feature vector set, it also includes: inputting the user weighted feature vector set into the sum pooling layer of the first churn risk identification sub-model, and outputting the user feature code; obtaining the associated feature code and the target churn type feature code received by the input layer of the first churn risk identification sub-model, wherein the associated feature code is obtained based on the user portrait of the target user, and the target churn type feature code is obtained based on the historical churn user; based on the feature splicing layer of the first churn risk identification sub-model, the user feature code, the associated feature code and the target churn type feature code are feature spliced ​​to obtain the spliced ​​feature; inputting the spliced ​​feature into the multi-layer neural network of the first churn risk identification sub-model, and outputting the initial churn probability of the target user at each time point.

[0065] It should be noted that after the user weighted feature vector set is output through the first churn risk identification sub-model, the weighted feature vector set is used to combine the features extracted from the user portrait of the target user and the features extracted from the behavior pattern of the historical churn users to generate the initial churn probability of the target user at each time point. This process includes summing and pooling, feature concatenation, and quantitative calculation of multi-layer neural networks to achieve accurate churn risk assessment. Specifically, the user weighted feature vector set is input into the summing and pooling layer of the first churn risk identification sub-model. The function of the summing and pooling layer is to perform weighted summation on all vectors in the user feature vector set to generate a user feature code that integrates all behavioral features of the user at that time point. This encoding process effectively aggregates the user feature information weighted by the attention mechanism. While generating the user feature code, two additional sets of codes need to be obtained, the associated feature code and the target churn type feature code. The associated feature code is obtained based on the user portrait of the target user, and contains the user's basic information, preferences, historical transaction records and other feature information, aiming to fully reflect the user's behavior pattern and possible churn risk factors. The target churn type feature code is generated based on the data set of historical churn users. By analyzing the common features of churn users, the features most relevant to churn risk are extracted. Based on the feature splicing layer of the first churn risk identification sub-model, the user feature code, associated feature code, and target churn type feature code are spliced ​​to effectively integrate the feature information of different dimensions of the user, and the spliced ​​features are input into the multi-layer neural network of the first churn risk identification sub-model. The multi-layer neural network can perform deep learning on the spliced ​​features, quantify the spliced ​​features into user churn probabilities, and finally output the initial churn probabilities of the target users at each time point.

[0066] Optionally, the step of extracting the associated feature code includes: obtaining a user portrait of the target user pre-built in the financial institution; extracting associated features associated with user churn from the user portrait, wherein the associated features include at least one of the following: user attribute features, transaction environment features; encoding the associated features to obtain the associated feature code.

[0067] It should be noted that in order to more comprehensively assess the risk of user churn, related features closely related to the churn risk can be extracted based on the user portraits pre-built by financial institutions, including: user attribute features, transaction environment features. User attribute features can be user age, user occupation, etc. Transaction environment features can include: transaction time, transaction channel and other features. The related features are then encoded to obtain the related feature codes, and the related feature codes are input into the first churn risk identification sub-model, which can enrich the feature vector and improve the accuracy of the output results.

[0068] Optionally, the step of extracting the target churn type feature code includes: obtaining financial data of historical churn users within a historical time period; performing cluster analysis on the financial data of historical churn users and target users to obtain clustering results; extracting common features of the target user and the historical churn users from the clustering results, and determining the target churn type features of the target user based on the common features and the churn types of historical churn users; encoding the target churn type features to obtain the target churn type feature code.

[0069] It should be noted that the target churn type characteristics can be used as the standard for calculating the churn probability. They are determined based on the common characteristics of historical churn users. Specifically, the user data of churn behaviors in a specified historical time period are extracted from the historical data warehouse of the financial institution. These data usually include the user's transaction records, account balance changes, product usage, customer service interaction records, etc., covering the user's behavior trajectory in the financial institution. Cluster analysis is performed on the financial data of historical churn users and target users, and the common characteristics of churned users are extracted through clustering. For example: a sudden decrease in the user's high-frequency transactions, and the observation that the user's transaction frequency has dropped significantly within a certain period of time may indicate a decrease in interest in the financial institution's services; large-scale capital flows, abnormal large-scale capital inflows or outflows in user accounts, indicate that financial services may no longer meet their needs or there are better alternatives. Reduced service interactions, and the reduction in customer service interaction records between users and financial institutions may indicate that users no longer trust or use the feedback channels of financial institutions when encountering problems. Based on the extracted common features, combined with the churn types of historical churn users (such as switching to competitors, service dissatisfaction, change in demand, etc.) and the user characteristics of the target users, we can obtain the churn type characteristics that the target users may face, thereby obtaining the target churn type characteristics. Finally, the target churn type characteristics are encoded to obtain the target churn type feature encoding that can be recognized by the model.

[0070] Through the above steps, a financial data set of the target user at each time point is obtained, wherein the financial data set contains N financial data, and the financial data includes at least one of the following: transaction data, asset data, and interactive behavior data, and N is a positive integer. Then, the financial data set at each time point is preprocessed to obtain a user feature set of the target user at each time point. Finally, the user features in the user feature set at each time point are input one by one into the churn risk identification model, and a churn risk identification result of the target user is output, wherein the churn risk identification result at least includes: whether the target user has a churn risk. The churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model. The first churn risk identification sub-model is a model pre-constructed based on an attention mechanism and a deep interest network, and the second churn risk identification sub-model is a model pre-constructed based on a long short-term memory network. The churn risk identification model calculates the churn probability of the target user, and compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

[0071] In this embodiment, the financial data of the target user at each time point is analyzed through a pre-constructed churn risk identification model, which is composed of a first churn risk identification sub-model and a second churn risk identification sub-model. The deep interest network based on the first churn risk identification sub-model mines the risk features related to churn risk identification in the financial data, and the long short-term memory network based on the second churn risk identification sub-model learns the changing trend of user behavior, and then calculates the user churn probability based on the risk features and the changing trend of user behavior, and accurately identifies the churn risk of the target user through the churn probability, thereby achieving the purpose of quantitatively identifying the churn risk and achieving the technical effect of improving the accuracy of the user churn risk identification results. This solves the technical problem in the related art that the accuracy of the identification results obtained by identifying the churn risk of users based on the rule matching method is low.

[0072] Another optional specific implementation is described in detail below.

[0073] Figure 3 is a schematic diagram of an optional neural network-based user churn risk identification process according to an embodiment of the present invention, such as Figure 3 As shown in the figure, the user churn risk identification process based on neural network includes:

[0074] Step 1: Data collection;

[0075] Collect financial data of target users from multiple dimensions, including transaction details, asset data, and interaction behavior data between users and offline entities of financial institutions and financial applications.

[0076] Step 2: data processing;

[0077] Preprocess the data to obtain high-quality data and data that can be recognized by the model, including: data cleaning and feature extraction.

[0078] Step 3: Based on the deep interest network of the churn risk identification model (corresponding to the first churn risk identification sub-model mentioned above), the initial churn probability at a single time point is output;

[0079] The deep interest network calculates the initial churn probability based on the user feature set at each time point to obtain the initial churn probability at each time point. At the same time, the attention weight feature vector is calculated based on the user feature based on the attention layer. The user weighted feature vector is calculated based on the user feature vector and the attention weight vector to obtain the user weighted feature vector set. Finally, the initial churn probability at each time point ( Figure 3 The churn probability at time point 1, the churn probability at time point 2, and the churn probability at time point N are used as an example) and the user weighted feature vector set ( Figure 3In the figure, weighted vector set 1, weighted vector set 2, ..., weighted vector set N) are used as the output results of the deep interest network.

[0080] Step 4: Output the target user churn probability based on the long short-term memory network of the churn risk identification model (corresponding to the second churn risk identification sub-model mentioned above);

[0081] The output results of the deep interest network are used as the input of the long short-term memory network, and the total churn probability of the target users is calculated through the long short-term memory network.

[0082] Step 5: Output the target user churn risk based on the churn risk identification model.

[0083] The churn risk identification model compares the churn probability output by the long short-term memory network with the preset churn probability threshold, and outputs the churn risk identification result for the target user based on the comparison result.

[0084] The first churn risk identification sub-model is a neural network model built based on a deep interest network. Figure 4 is a schematic diagram of an optional first churn risk identification sub-model according to an embodiment of the present invention, such as Figure 4 As shown, when the first churn risk identification sub-model calculates the initial churn probability of the target user at each time point, the encoding layer and the local attention module (corresponding to the above-mentioned attention layer) process the user's historical behavior characteristics respectively, output the user feature vector based on the encoding layer, output the attention weight feature vector based on the local attention module, and then multiply the user feature vector with the attention weight feature vector to obtain the user weighted feature vector, and then obtain the user weighted feature vector set at each time point, which will be used as one of the output results of the first churn risk identification sub-model.

[0085] After obtaining the user weighted feature vector set, the user weighted feature vector set is input into the summation pooling layer of the first churn risk identification sub-model, and all vectors in the user feature vector set are weighted summed to generate user feature coding, and then feature splicing is performed based on the associated feature coding and the target churn type feature coding to obtain spliced ​​features. Finally, the spliced ​​features can be deeply learned through a multi-layer neural network, and the spliced ​​features are quantified into user churn probabilities, and finally the estimated results, i.e., the initial churn probabilities of the target users at each time point, are output.

[0086] In an embodiment of the present invention, based on measuring the importance of user features through a deep interest network and an attention mechanism, features related to churn risk are mined, and the changing trend of users within a certain time period is identified through a long short-term memory network, so as to predict the user churn probability, identify the churn risk of target users based on the churn probability, realize quantitative identification of churn risk, and improve the accuracy of churn risk identification results.

[0087] The following is a detailed description in conjunction with another embodiment.

[0088] Embodiment 2

[0089] The user churn risk identification method based on a neural network provided in this embodiment includes multiple implementation units, each implementation unit corresponds to each implementation step in the above-mentioned embodiment 1. Its specific implementation method and beneficial effects can refer to the above-mentioned method embodiment, which will not be repeated here.

[0090] Figure 5 is a schematic diagram of an optional neural network-based user churn risk identification device according to an embodiment of the present invention, such as Figure 5 As shown, the user churn risk identification device based on neural network may include: an acquisition unit 51, a processing unit 52, and an output unit 53, wherein:

[0091] An acquisition unit 51 is used to acquire a financial data set of a target user at each time point, wherein the financial data set includes N pieces of financial data, and the financial data includes at least one of the following: transaction data, asset data, and interaction behavior data, and N is a positive integer;

[0092] A processing unit 52, used to pre-process the financial data set at each time point to obtain a user feature set of the target user at each time point;

[0093] The output unit 53 is used to input the user features in the user feature set at each time point into the churn risk identification model one by one, and output the churn risk identification result of the target user, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-built based on the attention mechanism and the deep interest network, and the second churn risk identification sub-model is a model pre-built based on the long short-term memory network. The churn risk identification model calculates the churn probability of the target user, compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

[0094] The above-mentioned user churn risk identification device based on neural network obtains the financial data set of the target user at each time point through the acquisition unit 51, wherein the financial data set contains N financial data, and the financial data includes at least one of the following: transaction data, asset data, and interactive behavior data, and N is a positive integer; the financial data set at each time point is pre-processed by the processing unit 52 to obtain the user feature set of the target user at each time point; the user features in the user feature set at each time point are input into the churn risk identification model one by one through the output unit 53, and the churn risk identification result of the target user is output, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-constructed based on the attention mechanism and the deep interest network, and the second churn risk identification sub-model is a model pre-constructed based on the long short-term memory network, and the churn risk identification model calculates the churn probability of the target user, compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

[0095] In this embodiment, the financial data of the target user at each time point is analyzed through a pre-constructed churn risk identification model, which is composed of a first churn risk identification sub-model and a second churn risk identification sub-model. The deep interest network based on the first churn risk identification sub-model mines the risk features related to churn risk identification in the financial data, and the long short-term memory network based on the second churn risk identification sub-model learns the changing trend of user behavior, and then calculates the user churn probability based on the risk features and the changing trend of user behavior, and accurately identifies the churn risk of the target user through the churn probability, thereby achieving the purpose of quantitatively identifying the churn risk and achieving the technical effect of improving the accuracy of the user churn risk identification results. This solves the technical problem in the related art that the accuracy of the identification results obtained by identifying the churn risk of users based on the rule matching method is low.

[0096] Optionally, the output unit 53 includes: a first output sub-unit, used to input the user features in the user feature set at each time point into the first churn risk identification sub-model one by one, and output the initial churn probability of the target user at each time point and the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model; a second output sub-unit, used to input the initial churn probability of the target user at each time point and the user weighted feature vector set into the second churn risk identification sub-model, and output the churn probability of the target user; a third output sub-unit, used to generate a churn risk identification result of the target user based on the churn probability of the target user, and output the churn risk identification result of the target user based on the output layer of the churn risk identification model, wherein, when the churn probability of the target user is greater than or equal to a preset churn probability threshold, it is determined that the target user has a churn risk.

[0097] Optionally, the first output sub-unit includes: a first encoding module, for inputting the user features in the user feature set at each time point into the first churn risk identification sub-model one by one, encoding each user feature through the encoding layer of the first churn risk identification sub-model to obtain a user feature vector set; a first calculation module, for calculating the attention weight feature vector of the user feature through the attention layer of the first churn risk identification sub-model; a first output module, for performing vector operations on each user feature vector in the user feature vector set and the attention weight feature vector through the first churn risk identification sub-model to obtain a user weighted feature vector set, and outputting the user weighted feature vector set by the first churn risk identification sub-model.

[0098] Optionally, the first output sub-unit includes: a second output module, used to input the user weighted feature vector set into the sum pooling layer of the first churn risk identification sub-model, and output the user feature code; a first acquisition module, used to obtain the associated feature code and the target churn type feature code received by the input layer of the first churn risk identification sub-model, wherein the associated feature code is obtained based on the user portrait of the target user, and the target churn type feature code is obtained based on the historical churn user; a first splicing module, used to perform feature splicing on the user feature code, the associated feature code and the target churn type feature code based on the feature splicing layer of the first churn risk identification sub-model to obtain a spliced ​​feature; a third output module, used to input the spliced ​​feature into the multi-layer neural network of the first churn risk identification sub-model, and output the initial churn probability of the target user at each time point.

[0099] Optionally, the user churn risk identification device based on neural network also includes: a first acquisition subunit, used to obtain a user portrait of the target user pre-built in a financial institution; a first extraction subunit, used to extract associated features associated with user churn from the user portrait, wherein the associated features include at least one of the following: user attribute features, transaction environment features; a first encoding subunit, used to encode the associated features to obtain the associated feature codes.

[0100] Optionally, the user churn risk identification device based on neural network also includes: a second acquisition subunit, used to acquire the financial data of historical churn users within a historical time period; a first analysis subunit, used to perform cluster analysis on the financial data of historical churn users and target users to obtain clustering results; a second extraction subunit, used to extract common features of target users and historical churn users from the clustering results, and determine target churn type features of the target users based on the common features and the churn types of historical churn users; a second encoding subunit, used to encode the target churn type features to obtain the target churn type feature encoding.

[0101] Optionally, the processing unit 52 includes: a first processing sub-unit, used to perform data cleaning, standardization and normalization processing on the financial data at each time point to obtain processed financial data; a first extraction sub-module, used to perform feature extraction on the processed financial data to obtain a user feature set of the target user at each time point, wherein feature extraction includes time point feature extraction and time period feature extraction, time point feature extraction means extracting the behavioral features of the target user at a single time point, and time period feature extraction means extracting the behavioral feature data of the target user within a predetermined time period.

[0102] It should be noted that the acquisition unit 51, the processing unit 52, and the output unit 53 correspond to steps S201 to S203 in the first embodiment. The examples and application scenarios implemented by the above units and the corresponding steps are the same, but are not limited to the contents disclosed in the first embodiment. It should be noted that the above modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules or units can also be part of the device and can be run in the computer terminal 10 provided in the first embodiment.

[0103] The present invention is described below in conjunction with another optional embodiment.

[0104] Embodiment 3

[0105] An embodiment of the present invention may also provide an electronic device, Figure 6is a hardware structure block diagram of an electronic device (or mobile device) that optionally executes a user churn risk identification method based on a neural network according to an embodiment of the present invention, such as Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (only one is shown) processor 602, memory 604, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0106] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned method. The memory 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 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal 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 combinations thereof.

[0107] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain a financial data set of the target user at each time point, wherein the financial data set contains N financial data, and the financial data includes at least one of the following: transaction data, asset data, and interactive behavior data, and N is a positive integer; pre-process the financial data set at each time point to obtain a user feature set of the target user at each time point; input the user features in the user feature set at each time point into the churn risk identification model one by one, and output the churn risk identification result of the target user, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-built based on the attention mechanism and the deep interest network, and the second churn risk identification sub-model is a model pre-built based on the long short-term memory network, and the churn risk identification model calculates the churn probability of the target user, compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

[0108] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: input the user features in the user feature set at each time point into the churn risk identification model one by one, and the step of outputting the churn risk identification result of the target user includes: inputting the user features in the user feature set at each time point into the first churn risk identification sub-model one by one, and outputting the initial churn probability of the target user at each time point and the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model; inputting the initial churn probability of the target user at each time point and the user weighted feature vector set into the second churn risk identification sub-model, and outputting the churn probability of the target user; generating the churn risk identification result of the target user based on the churn probability of the target user, and outputting the churn risk identification result of the target user based on the output layer of the churn risk identification model, wherein, when the churn probability of the target user is greater than or equal to a preset churn probability threshold, it is determined that the target user has a churn risk.

[0109] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: input the user features in the user feature set at each time point one by one into the first churn risk identification sub-model, and the step of outputting the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model includes: for the user feature set at each time point, input the user features in the user feature set one by one into the first churn risk identification sub-model, encode each user feature through the encoding layer of the first churn risk identification sub-model to obtain a user feature vector set; calculate the attention weight feature vector of the user feature through the attention layer of the first churn risk identification sub-model; perform vector operation on each user feature vector in the user feature vector set and the attention weight feature vector through the first churn risk identification sub-model to obtain a user weighted feature vector set, and output the user weighted feature vector set by the first churn risk identification sub-model.

[0110] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: after the first churn risk identification sub-model outputs the user weighted feature vector set, it also includes: inputting the user weighted feature vector set into the sum pooling layer of the first churn risk identification sub-model, and outputting the user feature code; obtaining the associated feature code and the target churn type feature code received by the input layer of the first churn risk identification sub-model, wherein the associated feature code is obtained based on the user portrait of the target user, and the target churn type feature code is obtained based on the historical churn user; based on the feature splicing layer of the first churn risk identification sub-model, the user feature code, the associated feature code and the target churn type feature code are feature spliced ​​to obtain the spliced ​​feature; the spliced ​​feature is input into the multi-layer neural network of the first churn risk identification sub-model, and the initial churn probability of the target user at each time point is output.

[0111] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: the step of extracting the associated feature code includes: obtaining the user portrait of the target user pre-built in the financial institution; extracting the associated features associated with user churn from the user portrait, wherein the associated features include at least one of the following: user attribute features, transaction environment features; encoding the associated features to obtain the associated feature code.

[0112] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: the step of extracting the target churn type feature code includes: obtaining the financial data of historical churn users within a historical time period; performing cluster analysis on the financial data of historical churn users and target users to obtain clustering results; extracting the common features of the target user and the historical churn users from the clustering results, and determining the target churn type features of the target user based on the common features and the churn types of the historical churn users; encoding the target churn type features to obtain the target churn type feature code.

[0113] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: pre-processing the financial data at each time point to obtain the user feature set of the target user at each time point, the step comprising: performing data cleaning, standardization and normalization processing on the financial data at each time point to obtain processed financial data; performing feature extraction on the processed financial data to obtain the user feature set of the target user at each time point, wherein the feature extraction includes time point feature extraction and time period feature extraction, the time point feature extraction means extracting the behavioral features of the target user at a single time point, and the time period feature extraction means extracting the behavioral feature data of the target user within a predetermined time period.

[0114] By adopting the embodiment of the present invention, a method for identifying user churn risk based on a neural network is provided. The financial data of the target user at each time point is analyzed by a pre-constructed churn risk identification model. The churn risk identification model is composed of a first churn risk identification sub-model and a second churn risk identification sub-model. The risk features related to churn risk identification in the financial data are mined based on the deep interest network of the first churn risk identification sub-model, and the changing trend of user behavior is learned based on the long short-term memory network of the second churn risk identification sub-model. Then, the user churn probability is calculated based on the risk features and the changing trend of user behavior. The churn risk of the target user is accurately identified through the churn probability, so as to achieve the purpose of quantitatively identifying the churn risk and obtain the technical effect of improving the accuracy of the user churn risk identification results. This solves the technical problem in the related art that the accuracy of the identification results obtained by identifying the churn risk of the user based on the rule matching method is low.

[0115] It can be understood by those skilled in the art that Figure 6 The structure shown is for illustration only, and the electronic device may also be a smart phone, a tablet computer, a PDA, a mobile Internet device (MID), a PAD or other terminal device. Figure 6 The structure of the electronic device is not limited. Figure 6 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Figure 6 Different configurations are shown.

[0116] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0117] The present invention is described below in conjunction with another optional embodiment.

[0118] Embodiment 4

[0119] The embodiment of the present invention further provides a computer-readable storage medium. Optionally, in the embodiment of the present invention, the computer-readable storage medium can be used to store the program code executed by the user churn risk identification method based on neural network provided in the first embodiment.

[0120] Optionally, in an embodiment of the present invention, the above-mentioned storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0121] The embodiment of the present invention also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the user churn risk identification method based on a neural network: obtaining a financial data set of a target user at each time point, wherein the financial data set contains N financial data, and the financial data includes at least one of the following: transaction data, asset data, and interactive behavior data, and N is a positive integer; preprocessing the financial data set at each time point to obtain a user feature set of the target user at each time point; inputting the user features in the user feature set at each time point into a churn risk identification model one by one, and outputting a churn risk identification result of the target user, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-constructed based on an attention mechanism and a deep interest network, and the second churn risk identification sub-model is a model pre-constructed based on a long short-term memory network, and the churn risk identification model calculates the churn probability of the target user, and after comparing the churn probability with a preset churn probability threshold, outputs the churn risk identification result of the target user.

[0122] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0123] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0124] 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.

[0125] 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.

[0126] In addition, each functional unit in each embodiment of the present invention 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.

[0127] 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 invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0128] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for identifying user churn risk based on a neural network, characterized in that: include: Obtain a financial data set of the target user at each time point, wherein the financial data set includes N pieces of financial data, and the financial data includes at least one of the following: transaction data, asset data, and interaction behavior data, where N is a positive integer; Preprocessing the financial data set at each time point to obtain a user feature set of the target user at each time point; The user features in the user feature set at each time point are input one by one into the churn risk identification model, and the churn risk identification result of the target user is output, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-constructed based on an attention mechanism and a deep interest network, and the second churn risk identification sub-model is a model pre-constructed based on a long short-term memory network, and the churn risk identification model calculates the churn probability of the target user, compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

2. The method according to claim 1, characterized in that The step of inputting the user features in the user feature set at each time point into the churn risk identification model one by one and outputting the churn risk identification result of the target user comprises: Inputting the user features in the user feature set at each time point into the first churn risk identification sub-model one by one, and outputting the initial churn probability of the target user at each time point and the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model; Inputting the initial churn probability of the target user at each time point and the user weighted feature vector set into the second churn risk identification sub-model, and outputting the churn probability of the target user; A churn risk identification result of the target user is generated based on the churn probability of the target user, and the churn risk identification result of the target user is output based on the output layer of the churn risk identification model, wherein when the churn probability of the target user is greater than or equal to a preset churn probability threshold, it is determined that the target user has a churn risk.

3. The method according to claim 2, characterized in that The steps of inputting the user features in the user feature set at each time point into the first churn risk identification sub-model one by one, and outputting the user weighted feature vector set of the target user at each time point through the first churn risk identification sub-model include: For a user feature set at each time point, inputting user features in the user feature set into the first churn risk identification sub-model one by one, encoding each user feature through the encoding layer of the first churn risk identification sub-model to obtain a user feature vector set; Calculate the attention weight feature vector of the user feature through the attention layer of the first churn risk identification sub-model; The first churn risk identification sub-model performs vector operations on each user feature vector in the user feature vector set and the attention weight feature vector to obtain a user weighted feature vector set, and the first churn risk identification sub-model outputs the user weighted feature vector set.

4. The method according to claim 3, characterized in that After the first churn risk identification sub-model outputs the user weighted feature vector set, the method further includes: Inputting the user weighted feature vector set into the summing pooling layer of the first churn risk identification sub-model, and outputting the user feature code; Obtaining the associated feature code and the target churn type feature code received by the input layer of the first churn risk identification sub-model, wherein the associated feature code is obtained based on the user profile of the target user, and the target churn type feature code is obtained based on historical churn users; Based on the feature concatenation layer of the first churn risk identification sub-model, the user feature code, the associated feature code and the target churn type feature code are concatenated to obtain a concatenated feature; The splicing features are input into the multi-layer neural network of the first churn risk identification sub-model, and the initial churn probability of the target user at each time point is output.

5. The method according to claim 4, characterized in that The step of extracting the associated feature code comprises: Obtaining a user profile of the target user pre-built in a financial institution; Extracting correlation features associated with user churn from the user portrait, wherein the correlation features include at least one of the following: user attribute features, transaction environment features; The associated features are encoded to obtain the associated feature codes.

6. The method according to claim 4, characterized in that The step of extracting the target loss type feature code includes: Obtaining financial data of the historical lost users within a historical time period; Performing cluster analysis on the financial data of the historical lost users and the target users to obtain clustering results; Extracting common features between the target user and the historical churned users from the clustering results, and determining a target churn type feature of the target user based on the common features and the churn type of the historical churned users; The target loss type feature is encoded to obtain the target loss type feature code.

7. The method according to claim 1, characterized in that The step of preprocessing the financial data at each time point to obtain a user feature set of the target user at each time point includes: Performing data cleaning, standardization and normalization on the financial data at each time point to obtain the processed financial data; Perform feature extraction on the processed financial data to obtain a set of user features of the target user at each time point, wherein the feature extraction includes time point feature extraction and time period feature extraction, the time point feature extraction means extracting the behavioral features of the target user at a single time point, and the time period feature extraction means extracting the behavioral feature data of the target user within a predetermined time period.

8. A user churn risk identification device based on a neural network, characterized in that: include: An acquisition unit, configured to acquire a financial data set of a target user at each time point, wherein the financial data set includes N pieces of financial data, and the financial data includes at least one of the following: transaction data, asset data, and interaction behavior data, where N is a positive integer; A processing unit, configured to pre-process the financial data set at each time point to obtain a user feature set of the target user at each time point; The output unit is used to input the user features in the user feature set at each time point into the churn risk identification model one by one, and output the churn risk identification result of the target user, wherein the churn risk identification result at least includes: whether the target user has a churn risk, the churn risk identification model includes a first churn risk identification sub-model and a second churn risk identification sub-model, the first churn risk identification sub-model is a model pre-constructed based on an attention mechanism and a deep interest network, and the second churn risk identification sub-model is a model pre-constructed based on a long short-term memory network, and the churn risk identification model calculates the churn probability of the target user, compares the churn probability with a preset churn probability threshold, and then outputs the churn risk identification result of the target user.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the user churn risk identification method based on a neural network as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the user churn risk identification method based on a neural network as described in any one of claims 1 to 7.

Images