A derivative feature extraction method, device, computer equipment and storage medium
Through the grouping induction algorithm and RFM model combined with feature cross-border network, an attention mechanism is introduced, which solves the problem of low effectiveness of traditional data mining methods in financial business scenarios, and achieves efficient and meaningful feature extraction and interpretation.
Patent Information
- Application Number
- CN202411145445.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-20
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-20
AI Technical Summary
Traditional data mining methods have the problem of low data mining effectiveness in financial business scenarios, and cannot fully consider the business meaning of the characteristics, which makes the resulting characteristics difficult to explain and meaningless.
A grouping induction algorithm is used to mine historical ticket information, call the trained RFM model for data classification, and cross feature through feature crossing network, and introduce attention mechanisms to screen out features with potential value.
It improves the effectiveness of data mining, the generated features are business explanatory and practical, and can more effectively mine meaningful information in massive data.
Smart Images

Figure CN118656685B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of artificial intelligence data mining and data analysis, and in particular to a derivative feature extraction method, device, computer equipment and storage medium. Background Art
[0002] In the era of big data, in order to obtain the required information from massive data, it is necessary to conduct data mining. Data mining is a hot topic in the research of artificial intelligence and database. The so-called data mining refers to the non-trivial process of revealing implicit, previously unknown and potentially valuable information from a large amount of data in a database.
[0003] At present, in the financial business scenario, data mining methods are mainly implemented through manual methods, that is, algorithm engineers derive features according to certain algorithms (such as neural network algorithms), or use meaningless feature derivation methods to randomly combine data. This method does not fully consider the business meaning of the features and does not derive them in combination with the business meaning. Although a large number of features can be generated, these features cannot be directly explained in the business, and it is difficult to generate a large number of meaningful features. It can be seen that traditional data mining methods have the problem of low data mining effectiveness. Summary of the invention
[0004] The purpose of the embodiments of the present application is to provide a derivative feature extraction method, apparatus, computer equipment and storage medium to solve the problem of low data mining effectiveness in traditional data mining methods.
[0005] In order to solve the above technical problems, the present application embodiment provides a derivative feature extraction method, which adopts the following technical solution:
[0006] Receiving a feature extraction request carrying a target user sent by a user terminal;
[0007] Reading a system database, and obtaining historical ticket information corresponding to the target user in the system database;
[0008] Performing feature mining operations on the historical bill information according to a grouping induction algorithm to obtain mining feature information;
[0009] Calling a trained RFM model, and performing a data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data, wherein the frequency data of the RFM model includes input frequency data and output frequency data;
[0010] Performing a feature cross operation on the feature classification data according to a feature cross network to obtain derived feature information, wherein an attention mechanism is introduced into each layer of the feature cross network;
[0011] The derived feature information is output to the user terminal.
[0012] Furthermore, after the step of reading the system database and obtaining the historical ticket information corresponding to the target user in the system database, the method further includes the following steps:
[0013] Performing data cleaning operation on the historical ticket information to obtain cleaned ticket data;
[0014] Performing data integration operation on the cleaning ticket data to obtain an integrated data table;
[0015] Performing a redundancy detection operation on the integrated data table to obtain a redundancy detection result;
[0016] Performing a data elimination operation on the historical bill information according to the redundant detection result to obtain pre-processed bill data;
[0017] The step of performing a feature mining operation on the historical bill information according to the grouping induction algorithm to obtain the mined feature information specifically includes the following steps:
[0018] A feature mining operation is performed on the pre-processed bill data according to a grouping induction algorithm to obtain the mined feature information.
[0019] Furthermore, before the step of calling the trained RFM model and performing data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data, the following steps are also included:
[0020] Obtaining historical ticket information of blacklisted users, and using the historical ticket information of the blacklisted users as training data;
[0021] Calling the initial RFM model of the original threshold, and inputting the training data into the initial RFM model to perform a preliminary feature mining operation to obtain preliminary mining feature information;
[0022] Performing statistical analysis on the preliminary mined feature information to obtain feature threshold information;
[0023] The original threshold is optimized according to the feature threshold information to obtain the trained RFM model.
[0024] Furthermore, after the step of receiving the feature extraction request carrying the target user sent by the user terminal, the following steps are also included:
[0025] Acquire life cycle information corresponding to the target user;
[0026] The step of calling the trained RFM model and performing data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data specifically includes the following steps:
[0027] A trained lifecycle RFM model corresponding to the lifecycle information is called, and a data classification operation is performed on the mined feature information according to the trained lifecycle RFM model to obtain the feature classification data.
[0028] Furthermore, before the step of performing a feature cross operation on the feature classification data according to the feature cross network to obtain the derived feature information, the following steps are also included:
[0029] Performing differentiation processing on the feature classification data to obtain differentiated classification data;
[0030] The step of performing a feature cross operation on the feature classification data according to the feature cross network to obtain derived feature information specifically includes the following steps:
[0031] A feature crossover operation is performed on the differentiated classification data according to a feature crossover network to obtain derived feature information.
[0032] Furthermore, the step of performing a feature cross operation on the differentiated classification data according to the feature cross network to obtain derived feature information specifically includes the following steps:
[0033] The differentiated classification data of the same category are multiplied according to the Hadamard product to obtain the Hadamard product feature;
[0034] The Hadamard product feature is used as an input of the feature crossover network to perform a feature crossover operation to obtain the derived feature information.
[0035] In order to solve the above technical problems, the embodiment of the present application further provides a derivative feature extraction device, which adopts the following technical solution:
[0036] A request acquisition module, used to receive a feature extraction request carrying a target user sent by a user terminal;
[0037] A data acquisition module, used for reading a system database and acquiring historical ticket information corresponding to the target user from the system database;
[0038] A feature mining module, used to perform feature mining operations on the historical bill information according to a grouping induction algorithm to obtain mining feature information;
[0039] A data classification module is used to call the trained RFM model and perform data classification operations on the mined feature information according to the trained RFM model to obtain feature classification data;
[0040] A feature crossover module, used for performing a feature crossover operation on the feature classification data according to a feature crossover network to obtain derived feature information, wherein an attention mechanism is introduced into each layer of the feature crossover network;
[0041] A feature output module is used to output the derived feature information to the user terminal.
[0042] Furthermore, the device further comprises: a data cleaning module, a data integration module, a redundancy detection module and a data elimination module, and the feature mining module further comprises: a feature mining submodule, wherein:
[0043] The data cleaning module is used to perform data cleaning operations on the historical bill information to obtain cleaned bill data;
[0044] The data integration module is used to perform data integration operations on the cleaning bill data to obtain an integrated data table;
[0045] The redundancy detection module is used to perform a redundancy detection operation on the integrated data table to obtain a redundancy detection result;
[0046] The data elimination module is used to perform a data elimination operation on the historical bill information according to the redundant detection result to obtain pre-processed bill data;
[0047] The feature mining submodule is used to perform feature mining operations on the pre-processed bill data according to a grouping induction algorithm to obtain the mined feature information.
[0048] Furthermore, the device also includes:
[0049] A training data acquisition module, used to acquire historical ticket information of blacklisted users and use the historical ticket information of the blacklisted users as training data;
[0050] A preliminary feature mining module is used to call the initial RFM model of the original threshold, and input the training data into the initial RFM model to perform a preliminary feature mining operation to obtain preliminary mining feature information;
[0051] A statistical analysis module, used to perform statistical analysis on the preliminary mined feature information to obtain feature threshold information;
[0052] The optimization module is used to optimize the original threshold according to the feature threshold information to obtain the trained RFM model.
[0053] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:
[0054] The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the derivative feature extraction method as described above when executing the computer-readable instructions.
[0055] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:
[0056] The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps of the derivative feature extraction method described above are implemented.
[0057] The present application provides a derivative feature extraction method, including: receiving a feature extraction request carrying a target user sent by a user terminal; reading a system database, and obtaining historical ticket information corresponding to the target user in the system database; performing a feature mining operation on the historical ticket information according to a grouping induction algorithm to obtain mined feature information; calling a trained RFM model, and performing a data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data, wherein the frequency data of the RFM model includes input frequency data and output frequency data; performing a feature cross operation on the feature classification data according to a feature cross network to obtain derivative feature information; and outputting the derivative feature information to the user terminal. Compared with the prior art, the present application mines preliminary feature information of the ticket information through a grouping induction algorithm, and then performs data classification on the preliminary feature information according to the RFM model to obtain classified feature data, and finally performs feature crossover according to a feature crossover network and screens the crossover features according to an attention mechanism, thereby mining data information with higher validity. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0059] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;
[0060] Figure 2 This is a flowchart of the implementation of the derivative feature extraction method provided in Example 1 of the present application;
[0061] Figure 3 is a schematic diagram of the structure of a derivative feature extraction device provided in Example 2 of the present application;
[0062] Figure 4 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION
[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.
[0064] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0065] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.
[0066] like Figure 1 As shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.
[0067] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.
[0068] Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III), MP4 (Moving Picture Experts Group Audio Layer IV), laptop computers, desktop computers, etc.
[0069] The server 105 may be a server that provides various services, such as a background server that provides support for web pages displayed on the terminal devices 101 , 102 , and 103 .
[0070] It should be noted that the derivative feature extraction method provided in the embodiment of the present application is generally executed by a server / terminal device, and accordingly, the derivative feature extraction device is generally arranged in the server / terminal device.
[0071] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.
[0072] Continue to refer Figure 2 , showing a flow chart of an embodiment of a derivative feature extraction method according to the present application. The derivative feature extraction method comprises: step S201, step S202, step S203, step S204, step S205 and step S206.
[0073] In step S201, a feature extraction request carrying a target user sent by a user terminal is received.
[0074] In the embodiments of the present application, the user terminal refers to a terminal device used to execute the image processing method for preventing document abuse provided by the present application. The user terminal can be a mobile terminal such as a mobile phone, a smart phone, a laptop computer, a digital broadcast receiver, a PDA (personal digital assistant), a PAD (tablet computer), a PMP (portable multimedia player), a navigation device, etc., as well as a fixed terminal such as a digital TV, a desktop computer, etc. It should be understood that the examples of user terminals here are only for the convenience of understanding and are not used to limit the present application.
[0075] In step S202, the system database is read to obtain historical ticket information corresponding to the target user in the system database.
[0076] In the embodiment of the present application, the system database is mainly used to store the historical transaction ticket information of each user. Specifically, the transaction ticket information includes transaction amount, transaction object, transaction time, transaction subject and other information.
[0077] In step S203, a feature mining operation is performed on the historical bill information according to the grouping induction algorithm to obtain mined feature information.
[0078] In the embodiment of the present application, the Group Method of Data Handling (GMDH) is a recursive algorithm that achieves data modeling and prediction by gradually building and improving the complexity of the model. In GMDH, the data is divided into different groups, each of which contains a portion of the input variables. Then, by performing step-by-step feature selection and model building for each group, starting from a simple model, the complexity of the model is gradually increased until a certain prediction performance is achieved. This step-by-step modeling process can effectively mine the features in the data and automatically select the best feature combination and model structure. Specifically, the implementation process of the group induction algorithm is as follows:
[0079] 1. Data preparation
[0080] First, collect transaction data within a period of time, including but not limited to transaction timestamps, transaction amounts, transaction accounts, etc. For example:
[0081] Trading time R: 2023-04-01 10:00:00, 2023-04-02 15:30:00, ...
[0082] Transaction amount M: 100, 100, 250, ...
[0083] Trading accounts: A001, A001, B002, ...
[0084] 2. Feature construction
[0085] Based on the original data, construct features that may affect the transaction pattern. For the transaction time R, you can extract the hour, date, whether it is a weekday, season, etc.; the transaction frequency F can be obtained by calculating the number of transactions per account per unit time; the transaction amount M can be used directly, or you can construct statistical features such as the average transaction amount, the standard deviation of the transaction amount, etc.
[0086] 3. Apply GMDH algorithm
[0087] At the beginning of the GMDH algorithm, the original features or constructed features are combined from all possible simple functions (such as linear, polynomial), and then the best feature combination is selected according to a certain criterion (such as minimizing the prediction error). This process is recursive, and each iteration further builds a more complex model based on the best feature combination selected in the previous round.
[0088] The specific steps are as follows:
[0089] Initialization: Set the initial feature pool, which may include the original R, F, M and their simple derivative features.
[0090] Model generation: Combine the features in the feature pool to generate a series of candidate models, such as R (hour) * F (number of transactions per hour), the average value of M, etc.
[0091] Evaluation and selection: Use historical data to train these models and evaluate their performance using methods such as cross-validation. Keep the model with the best performance.
[0092] Feature optimization: Based on the optimal model, remove features that contribute less to the model, or add new features and repeat the above process again until the preset stopping conditions are met (such as reaching the predetermined model complexity or the prediction accuracy is no longer significantly improved).
[0093] In practical applications, suppose that preliminary analysis shows that the peak trading period is from 8 to 10 p.m. on weekdays, and the average transaction amount of account A001 during this period has changed significantly compared with the previous day. GMDH may find in the iterative process that "(today is a weekday)*(the current time belongs to the evening peak period)*(the average change rate of the transaction amount of account A001)" is a strong predictor for predicting the trading behavior of the account at a specific time point.
[0094] In step S204, the trained RFM model is called, and data classification operations are performed on the mined feature information according to the trained RFM model to obtain feature classification data, wherein the frequency data of the RFM model includes input frequency data and output frequency data.
[0095] In the embodiment of the present application, the RFM model is an important tool and means for measuring customer value and customer's ability to create benefits, and is widely used in the analysis mode of customer relationship management (CRM). The model describes the value status of customers through three key indicators: the most recent consumption (recency), consumption frequency (frequency) and consumption amount (monetary), specifically:
[0096] Recent Purchase (R): refers to the time of the customer's last purchase. In theory, customers with more recent purchases are usually more valuable and more likely to respond to immediate goods or services;
[0097] Consumption frequency (F): refers to the number of times a customer purchases within a specified period of time. Customers who purchase frequently tend to have higher satisfaction and loyalty;
[0098] Consumption amount (M): refers to the total amount or average amount of consumption by a customer in the recent period. This reflects the customer's contribution to the company.
[0099] In the embodiments of the present application, the present application has made appropriate improvements to the RFM model. Specifically, taking into account the characteristics of account transactions for illegal fund-raising activities, such as the obvious division of labor in accounts: the funds in the collection account are "dispersedly transferred in and concentratedly transferred out", and sometimes huge amounts of funds are deposited; the funds in the transfer account are "quickly in and out, and frequently collected and paid", and the book balance is basically zero or very small; the funds in the refund account are "concentratedly transferred in and dispersedly transferred out", and the transaction time is often relatively concentrated, etc. When the RFM model is used in the field of financial risks, the RFM model is improved (Reinforce-RFM) to change the indicator of "consumption frequency" from undirected to directed, that is, from a simple high / low transaction frequency to a high / low transaction transfer frequency and a high / low transfer-in frequency, so as to capture the account category and transaction activity cycle of abnormal transaction accounts.
[0100] In an embodiment of the present application, the feature classification data includes core corporate accounts, transit fund accounts, private consumption accounts, and R, F, and M data features corresponding to the three types of accounts.
[0101] In step S205, a feature cross operation is performed on the feature classification data according to the feature cross network to obtain derived feature information, wherein an attention mechanism is introduced in each layer of the feature cross network.
[0102] In an embodiment of the present application, a feature cross network is mainly used to learn the complex interactive relationship between features in feature classification data. The feature cross network (Dual Attention Cross Network, DACN) is a feature cross network combined with an attention mechanism. This attention mechanism is different from the traditional feature cross network. DACN introduces an attention mechanism in the cross operation of each layer. This means that when performing feature crossover, the importance between different features will be learned. When deriving the Nth layer, the features of the N-1 layer will be filtered to remove features whose importance is lower than the threshold. Avoid the shortcomings of the traditional feature cross network that crosses all features and eliminate redundant information.
[0103] In the embodiment of the present application, the introduction of the attention mechanism can make the network pay more attention to and emphasize the important features in the input data. The attention mechanisms are the spatial attention mechanism and the channel attention mechanism to weight the input features, thereby improving the network's ability to represent and model the input data. Specifically:
[0104] (1) Spatial attention mechanism: The spatial attention mechanism focuses on the importance of different positions in the input data. In DACN, the spatial attention mechanism can help the network dynamically learn the relationship between features at different positions, thereby better capturing the spatial information in the input data;
[0105] (2) Channel attention mechanism: The channel attention mechanism focuses on the importance of different feature channels in the input data. Through the channel attention mechanism, DACN can adaptively learn the relationship between different feature channels, thereby better capturing the complex interactive relationship between features in the input data.
[0106] In the embodiment of the present application, each layer of the feature cross network consists of two parts: a cross operation and a residual connection. The cross operation is used to learn the high-order interaction between features, and the residual connection ensures that there will be no gradient disappearance or gradient explosion problems during the deep training of the model. The feature cross network DCN is used to derive the basic features of each category. This type of combination is sequential, generally R, F, M alone, or a combination of R and F, M, to derive basic features of each category such as R, F, M, RF, RM, etc., and a large number of cross features are derived through random cross combinations between different basic features, such as the R feature "last transaction time", the M feature "amount is greater than 100", and the RF feature "number of transactions in the past month is greater than 5 times". At the same time, the basic features, a layer of cross features and the newly generated cross features can also be randomly cross-combined to derive more and deeper features. More potential customer features can be excavated through the control of the derived level.
[0107] In the embodiment of the present application, an attention mechanism is introduced into the feature intersection network to screen out feature information with more potential value, wherein:
[0108] 1. Attention Mechanism Design
[0109] 1.1 Defining the attention function
[0110] For each pair of feature crosses, define an attention score calculation function Attention(·), which accepts cross features as input and outputs a scalar value representing the importance of the cross feature. It includes the following structure:
[0111] Input layer: receives the feature cross vector mined previously.
[0112] Hidden layer: Processed through a fully connected layer (FC) and an activation function (such as ReLU) to extract a high-level representation of feature interactions.
[0113] Output layer: A single neuron layer that uses a Sigmoid activation function to ensure that the output value is between 0 and 1, which serves as the attention weight for the cross-features.
[0114] 1.2 Calculating Attention Weights
[0115] For each feature intersection, apply the above attention function Attention() to calculate its attention weight. This process is dynamic and depends on the specific content of the intersection feature and the model training status.
[0116] 2. Weighted fusion features
[0117] The vectors of all feature crosses are multiplied by their corresponding attention weights, and then weighted summed to generate a comprehensive feature representation vector. This step realizes the dynamic adjustment of the importance of feature crosses and strengthens the cross features that are more valuable to the prediction target.
[0118] 3. Model training and optimization
[0119] The fused feature vector is input into the downstream task model to optimize the objective function (such as cross entropy loss).
[0120] During the training process, backpropagation is used to update not only the parameters of the task model, but also the parameters in the attention mechanism to optimize the attention weights of feature crosses.
[0121] In step S206, the derived feature information is output to the user terminal.
[0122] In an embodiment of the present application, a derivative feature extraction method is provided, including: receiving a feature extraction request carrying a target user sent by a user terminal; reading a system database, and obtaining historical ticket information corresponding to the target user in the system database; performing a feature mining operation on the historical ticket information according to a grouping induction algorithm to obtain mined feature information; calling a trained RFM model, and performing a data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data, wherein the frequency data of the RFM model includes input frequency data and output frequency data; performing a feature cross operation on the feature classification data according to a feature cross network to obtain derived feature information; and outputting the derived feature information to the user terminal. Compared with the prior art, the present application mines preliminary feature information of the ticket information through a grouping induction algorithm, and then performs data classification on the preliminary feature information according to the RFM model to obtain classified feature data, and finally performs feature crossover according to a feature crossover network and screens the crossover features according to an attention mechanism, thereby mining data information with higher validity.
[0123] In some optional implementations of the embodiments of the present application, after step S202, the following steps are further included:
[0124] Performing data cleaning operation on the historical ticket information to obtain cleaned ticket data;
[0125] Performing data integration operation on the cleaning ticket data to obtain an integrated data table;
[0126] Performing a redundancy detection operation on the integrated data table to obtain a redundancy detection result;
[0127] Performing a data elimination operation on the historical bill information according to the redundant detection result to obtain pre-processed bill data;
[0128] The above step S203 specifically includes the following steps:
[0129] A feature mining operation is performed on the pre-processed bill data according to a grouping induction algorithm to obtain the mined feature information.
[0130] In an embodiment of the present application, the historical ticket data obtained contains a large amount of data information of different categories and quantities, and there is duplicate data. The collected historical ticket data needs to be preprocessed. Specifically, the historical ticket data is integrated to combine data from multiple data sources into a coherent data table, and data redundancy detection is performed based on the data table to obtain redundant data information.
[0131] In the embodiment of the present application, the redundancy detection may be implemented as follows:
[0132] 1) Analyze the relationship between data tables:
[0133] ① Check how each table is connected and whether there are redundant key values and data columns;
[0134] ② Eliminate duplicate data by deleting these columns, modifying the table structure, or merging tables;
[0135] 2) Retrieve duplicate records:
[0136] ① Use SQL statements to query all duplicate data items in the table and mark invalid records;
[0137] ② After marking, you can delete them manually or write SQL commands to automatically delete them;
[0138] 3. Database design analysis:
[0139] ① Use tools such as ER diagram (entity-relationship diagram) and data dictionary to analyze database design and find the existence of redundant data;
[0140] 4. Database query analysis:
[0141] ① By querying the database, analyzing the relationship between data tables, and identifying whether there is redundant data;
[0142] 5. Database statistical analysis:
[0143] ① Count the data in the statistical table, analyze the data distribution, and identify whether there is redundant data;
[0144] ②If a data value appears multiple times in multiple data tables, there may be redundant data;
[0145] 6. Database index analysis:
[0146] ① By analyzing the indexes in the database, identify whether there are duplicate indexes and redundant indexes. Duplicate indexes and redundant indexes will cause database performance to degrade and increase the complexity of data management;
[0147] 7. Use deduplication technology:
[0148] In spreadsheet software such as Excel, you can use the "conditional formatting" function to mark duplicate rows, and then delete redundant data based on the marked content;
[0149] In the database, you can directly use deduplication commands similar to the distinct function.
[0150] It should be understood that the examples of redundant detection operations here are only for ease of understanding and are not intended to limit the present application.
[0151] In some optional implementations of the embodiments of the present application, before step S204, the following steps are also included:
[0152] Obtain the historical ticket information of the blacklisted users and use the historical ticket information of the blacklisted users as training data;
[0153] Call the initial RFM model of the original threshold, and input the training data into the initial RFM model for preliminary feature mining operations to obtain preliminary mining feature information;
[0154] Perform statistical analysis on the initially mined feature information to obtain feature threshold information;
[0155] The original threshold is optimized according to the feature threshold information to obtain the trained RFM model.
[0156] In the embodiment of the present application, the transaction information of a batch of blacklisted enterprise accounts is obtained, including information such as transaction amount, transaction object, transaction time, etc. These data can be obtained through channels such as blacklist lists provided by public security agencies and internal data of banking systems; the GMDH (group data processing method) neural network is used for preliminary feature mining and classification, and the fund transaction data of blacklisted enterprise accounts are input into the GMDH neural network, so that the model can automatically learn the characteristics and patterns of the data, and perform preliminary feature mining and classification; according to the output results of the GMDH neural network, the characteristics of fund transactions can be preliminarily divided. According to the distribution of transaction time, the frequency of transaction, the size of transaction amount, and the transaction category, the classification is carried out; by performing statistical analysis on the divided features, the initial threshold of the feature can be determined. For example, the threshold of the abnormal transaction time period can be determined according to the distribution of transaction time, the threshold of the abnormal transaction frequency can be determined according to the frequency of transaction, and the threshold of the abnormal transaction amount can be determined according to the size of the transaction amount. These initial thresholds can be used as the basis for subsequent monitoring and screening to help identify potential abnormal transaction behaviors.
[0157] In some optional implementations of the embodiments of the present application, the feature cross network can automatically create interaction terms between features and generate new derived features, thereby capturing the complex relationships between variables and improving the expressiveness and predictive performance of the model. However, these derived features are often more abstract and difficult to understand intuitively than the original features, which poses a challenge to the interpretability of the model. Therefore, on the basis of screening the features using the feature cross network, before step S206, an additional step is added: SHAP (SHapley Additive exPlanations) value to calculate the interpretability of the features. The detailed steps are as follows:
[0158] 1) Model training
[0159] Data preparation: Datasets of the aforementioned original features and derived features generated by feature cross-network.
[0160] Model selection: Select XGBoost as the ensemble model.
[0161] Training and validation set split: Splitting the dataset into training and validation sets (and sometimes test sets) ensures that the model can perform well on unseen data and avoids overfitting.
[0162] Model training: Train the selected model on a training set containing the original features and the newly generated derivative features. Adjust the model’s hyperparameters through cross-validation to optimize its performance.
[0163] Model evaluation and optimization: Use the validation set to evaluate model performance based on evaluation indicators (such as accuracy, AUC-ROC, precision, recall, etc.).
[0164] 2) SHAP value calculation
[0165] Calculate SHAP value: Use the trained model and SHAP tool to calculate the marginal contribution of each feature (including original features and derived features) to the prediction result. This is usually done through the following steps:
[0166] Select multiple blacklisted enterprise samples, and for each sample, calculate how its predicted value is accumulated by the contribution of all features.
[0167] The SHAP value assigns a contribution score to each feature, which reflects the relative importance of the feature in the prediction of the current sample, taking into account the interactions between features.
[0168] 3) Interpretation of results:
[0169] Summary Plot: Displays the average SHAP value of all features, helping to identify which features contribute most to the overall model predictions.
[0170] Force Plot: Provides in-depth analysis of a single prediction instance, visually demonstrating how each feature drives the model to make a specific prediction decision.
[0171] Dependence Plot: Explore the relationship between two features and how they jointly affect the model output.
[0172] Decision optimization and explanation: Based on the results of SHAP analysis, the most influential features can be identified and feature selection can be optimized. At the same time, SHAP values provide clear explainability for decisions, which helps to communicate the model decision process with non-technical team members.
[0173] Through the above steps, model retraining combined with SHAP value calculation not only improves the model's predictive ability, but also enhances the transparency and explainability of model decisions, which is crucial to improving the model's trust and application value.
[0174] In some optional implementations of the embodiments of the present application, after step S201, the following steps are further included:
[0175] Acquire life cycle information corresponding to the target user;
[0176] The above step S204 specifically includes the following steps:
[0177] A trained lifecycle RFM model corresponding to the lifecycle information is called, and a data classification operation is performed on the mined feature information according to the trained lifecycle RFM model to obtain the feature classification data.
[0178] In an embodiment of the present application, in order to achieve a more accurate evaluation, the system pre-sets RFM models according to users in different life cycle stages, such as new customers, growing customers, mature customers, and dormant customers, etc. Using corresponding RFM models for different life cycle stages can effectively improve the accuracy of data classification operations.
[0179] In some optional implementations of the embodiments of the present application, before step S205, the following steps are also included:
[0180] Performing differentiation processing on the feature classification data to obtain differentiated classification data;
[0181] Step S205 specifically includes the following steps:
[0182] A feature crossover operation is performed on the differentiated classification data according to a feature crossover network to obtain derived feature information.
[0183] In an embodiment of the present application, before performing feature crossover through a feature crossover network, the features are grouped homogeneously, and are differentiated during crossover, which helps to improve the performance of the model.
[0184] In some optional implementations of the embodiments of the present application, the step of performing a feature cross operation on the differentiated classification data according to the feature cross network to obtain derived feature information specifically includes the following steps:
[0185] The differentiated classification data of the same category are multiplied according to the Hadamard product to obtain the Hadamard product feature;
[0186] The Hadamard product feature is used as the input of the feature cross network to perform feature cross operation and obtain the derived feature information.
[0187] In the embodiment of the present application, the feature crossover according to the Hadamard Product is mainly to multiply the elements of two or more feature matrices (or vectors) at corresponding positions. The Hadamard Product requires that the dimensions of the two feature matrices or vectors are the same so as to perform the element-by-element multiplication operation, specifically:
[0188] 1. Determine the feature matrix or vector: First, you need to determine the two feature matrices or vectors to be crossed. These features can be numeric, Boolean, or any type that can be combined by multiplying corresponding positions.
[0189] 2. Ensure the same dimensions: Before performing the Hadamard product, you must ensure that the dimensions of the two feature matrices or vectors are exactly the same. If the dimensions are different, the Hadamard product operation cannot be performed.
[0190] 3. Perform the Hadamard product: multiply the elements of the two eigenmatrices or vectors at corresponding positions. The result will be a new matrix or vector with the same dimensions as the original matrix or vector. As an example, suppose there are two eigenvectors A and B:
[0191] A=[a1,a2,a3]
[0192] B=[b1,b2,b3]
[0193] Their Hadamard product is:
[0194] A×B=[a1b1,a2b2,a3*b3];
[0195] If A and B are matrices, then the Hadamard product is the matrix multiplication of corresponding elements.
[0196] Using crossed features: After completing the Hadamard product, you can use the resulting new features as input to your machine learning model.
[0197] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.
[0198] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.
[0199] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).
[0200] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps. Example
[0201] Further references Figure 3 , as a response to the above Figure 2 The present application provides an embodiment of a derivative feature extraction device, and the device embodiment is similar to Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.
[0202] like Figure 3 As shown, the derivative feature extraction device 200 of the embodiment of the present application includes:
[0203] The request acquisition module 210 is used to receive a feature extraction request carrying a target user sent by a user terminal;
[0204] The data acquisition module 220 is used to read the system database and obtain the historical ticket information corresponding to the target user in the system database;
[0205] The feature mining module 230 is used to perform feature mining operations on the historical bill information according to the grouping induction algorithm to obtain mining feature information;
[0206] The data classification module 240 is used to call the trained RFM model and perform data classification operations on the mined feature information according to the trained RFM model to obtain feature classification data;
[0207] A feature crossover module 250 is used to perform a feature crossover operation on the feature classification data according to a feature crossover network to obtain derived feature information, wherein an attention mechanism is introduced into each layer of the feature crossover network;
[0208] The feature output module 260 is used to output the derived feature information to the user terminal.
[0209] In an embodiment of the present application, a derivative feature extraction device 200 is provided, including: a request acquisition module 210, for receiving a feature extraction request carrying a target user sent by a user terminal; a data acquisition module 220, for reading a system database, and acquiring historical ticket information corresponding to the target user in the system database; a feature mining module 230, for performing a feature mining operation on the historical ticket information according to a grouping induction algorithm to obtain mining feature information; a data classification module 240, for calling a trained RFM model, and performing a data classification operation on the mining feature information according to the trained RFM model to obtain feature classification data; a feature cross module 250, for performing a feature cross operation on the feature classification data according to a feature cross network to obtain derivative feature information, wherein an attention mechanism is introduced in each layer of the feature cross network; a feature output module 260, for outputting the derivative feature information to the user terminal. Compared with the prior art, the present application mines preliminary feature information of ticket information through a grouping induction algorithm, and then performs data classification on the preliminary feature information according to the RFM model to obtain classified feature data, and finally performs feature cross according to the feature cross network and screens the cross features according to the attention mechanism, thereby mining data information with higher validity.
[0210] In some optional implementations of the embodiments of the present application, the apparatus 200 further includes:
[0211] A training data acquisition module is used to obtain historical ticket information of blacklisted users and use the historical ticket information of blacklisted users as training data;
[0212] The preliminary feature mining module is used to call the initial RFM model of the original threshold, and input the training data into the initial RFM model to perform preliminary feature mining operations to obtain preliminary mining feature information;
[0213] The statistical analysis module is used to perform statistical analysis on the preliminary mined feature information to obtain feature threshold information;
[0214] The optimization module is used to optimize the original threshold according to the feature threshold information to obtain a trained RFM model.
[0215] To solve the above technical problems, the present application also provides a computer device. Figure 4, Figure 4 This is a basic structural block diagram of a computer device according to an embodiment of the present application.
[0216] The computer device 300 includes a memory 310, a processor 320, and a network interface 330 that are interconnected and communicated through a system bus. It should be noted that the figure only shows a computer device 300 having components 310-330, but it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable gate arrays (Field-Programmable Gate Array, FPGA), digital processors (Digital Signal Processor, DSP), embedded devices, etc.
[0217] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.
[0218] The memory 310 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 310 can be an internal storage unit of the computer device 300, such as a hard disk or memory of the computer device 300. In other embodiments, the memory 310 can also be an external storage device of the computer device 300, such as a plug-in hard disk equipped on the computer device 300, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card, etc. Of course, the memory 310 can also include both the internal storage unit of the computer device 300 and its external storage device. In the embodiment of the present application, the memory 310 is generally used to store the operating system and various application software installed on the computer device 300, such as computer-readable instructions for the derived feature extraction method, etc. In addition, the memory 310 can also be used to temporarily store various data that have been output or are to be output.
[0219] The processor 320 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 320 is generally used to control the overall operation of the computer device 300. In the embodiment of the present application, the processor 320 is used to run the computer-readable instructions stored in the memory 310 or process data, such as computer-readable instructions for running the derivative feature extraction method.
[0220] The network interface 330 may include a wireless network interface or a wired network interface. The network interface 330 is generally used to establish a communication connection between the computer device 300 and other electronic devices.
[0221] The computer device provided in the present application mines preliminary feature information of the bill information through a grouping induction algorithm, and then classifies the preliminary feature information according to the RFM model to obtain classified feature data, and finally performs feature cross-linking according to the feature cross-linking network and screens the cross-linked features according to the attention mechanism, thereby mining data information with higher effectiveness.
[0222] The present application also provides another embodiment, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to enable the at least one processor to perform the steps of the derivative feature extraction method as described above.
[0223] The computer-readable storage medium provided in the present application mines preliminary feature information of the bill information through a grouping induction algorithm, and then classifies the preliminary feature information according to the RFM model to obtain classified feature data, and finally performs feature crossover according to a feature crossover network and screens the crossover features according to an attention mechanism, thereby mining data information with higher effectiveness.
[0224] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.
[0225] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.
Claims
1. A derivative feature extraction method, characterized in that: The steps include: Receiving a feature extraction request carrying a target user sent by a user terminal; Reading a system database, and obtaining historical ticket information corresponding to the target user in the system database; Performing feature mining operations on the historical bill information according to a grouping induction algorithm to obtain mining feature information; Calling a trained RFM model, and performing a data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data, wherein the frequency data of the RFM model includes input frequency data and output frequency data; Performing a feature cross operation on the feature classification data according to a feature cross network to obtain derived feature information, wherein an attention mechanism is introduced into each layer of the feature cross network; Outputting the derived feature information to the user terminal; After the step of reading the system database and obtaining the historical ticket information corresponding to the target user in the system database, the following steps are also included: Performing data cleaning operation on the historical ticket information to obtain cleaned ticket data; Performing data integration operation on the cleaning ticket data to obtain an integrated data table; Performing a redundancy detection operation on the integrated data table to obtain a redundancy detection result; Performing a data elimination operation on the historical bill information according to the redundant detection result to obtain pre-processed bill data; The step of performing a feature mining operation on the historical bill information according to the grouping induction algorithm to obtain the mined feature information specifically includes the following steps: A feature mining operation is performed on the pre-processed bill data according to a grouping induction algorithm to obtain the mined feature information.
2. The derivative feature extraction method according to claim 1, characterized in that: Before the step of calling the trained RFM model and performing data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data, the following steps are also included: Obtaining historical ticket information of blacklisted users, and using the historical ticket information of the blacklisted users as training data; Calling the initial RFM model of the original threshold, and inputting the training data into the initial RFM model to perform a preliminary feature mining operation to obtain preliminary mining feature information; Performing statistical analysis on the preliminary mined feature information to obtain feature threshold information; The original threshold is optimized according to the feature threshold information to obtain the trained RFM model.
3. The derivative feature extraction method according to claim 1, characterized in that: After the step of receiving the feature extraction request carrying the target user sent by the user terminal, the following steps are also included: Acquire life cycle information corresponding to the target user; The step of calling the trained RFM model and performing data classification operation on the mined feature information according to the trained RFM model to obtain feature classification data specifically includes the following steps: A trained lifecycle RFM model corresponding to the lifecycle information is called, and a data classification operation is performed on the mined feature information according to the trained lifecycle RFM model to obtain the feature classification data.
4. The derivative feature extraction method according to claim 1, characterized in that: Before the step of performing a feature cross operation on the feature classification data according to the feature cross network to obtain derived feature information, the following steps are also included: Performing differentiation processing on the feature classification data to obtain differentiated classification data; The step of performing a feature cross operation on the feature classification data according to the feature cross network to obtain derived feature information specifically includes the following steps: A feature crossover operation is performed on the differentiated classification data according to a feature crossover network to obtain the derived feature information.
5. The derivative feature extraction method according to claim 4, characterized in that: The step of performing a feature cross operation on the differentiated classification data according to the feature cross network to obtain derived feature information specifically includes the following steps: The differentiated classification data of the same category are multiplied according to the Hadamard product to obtain the Hadamard product feature; The Hadamard product feature is used as an input of the feature crossover network to perform a feature crossover operation to obtain the derived feature information.
6. A derivative feature extraction device, characterized in that: include: A request acquisition module, used to receive a feature extraction request carrying a target user sent by a user terminal; A data acquisition module, used for reading a system database and acquiring historical ticket information corresponding to the target user from the system database; A feature mining module, used to perform feature mining operations on the historical bill information according to a grouping induction algorithm to obtain mining feature information; A data classification module is used to call the trained RFM model and perform data classification operations on the mined feature information according to the trained RFM model to obtain feature classification data; A feature crossover module, used for performing a feature crossover operation on the feature classification data according to a feature crossover network to obtain derived feature information, wherein an attention mechanism is introduced into each layer of the feature crossover network; A feature output module, used to output the derived feature information to the user terminal; The device further comprises: a data cleaning module, a data integration module, a redundancy detection module and a data elimination module, and the feature mining module further comprises: a feature mining submodule, wherein: The data cleaning module is used to perform data cleaning operations on the historical bill information to obtain cleaned bill data; The data integration module is used to perform data integration operations on the cleaning bill data to obtain an integrated data table; The redundancy detection module is used to perform a redundancy detection operation on the integrated data table to obtain a redundancy detection result; The data elimination module is used to perform a data elimination operation on the historical bill information according to the redundant detection result to obtain pre-processed bill data; The feature mining submodule is used to perform feature mining operations on the pre-processed bill data according to a grouping induction algorithm to obtain the mined feature information.
7. A computer device, comprising a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the derivative feature extraction method according to any one of claims 1 to 5 when executing the computer-readable instructions.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the derivative feature extraction method according to any one of claims 1 to 5 are implemented.
Citation Information
Patent Citations
Multi-task learning-based credibility evaluation method for differential feature mining
CN109992703A
Sequence recommendation method based on self-attention network with deeper feature level
CN110083770A
Intention mining method, device and equipment and storage medium
CN111859984A
Commodity recommendation method based on neighbor users and comment information
CN112884551A
Intelligent feature automatic derivation method based on RFM and DFS algorithms
CN114385707A
