Data processing method and device, equipment, medium and program product
By using prediction models integrating N classifiers selected based on semi-supervised learning and greedy algorithms in the credit approval scenario, the noise problem caused by instability in the credit data quality is solved and the data processing accuracy of credit approval is improved.
Patent Information
- Application Number
- CN202411704926.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-26
- Publication Date
- 2025-03-11
AI Technical Summary
In the credit approval scenario, the quality of credit data is unstable, which affects the accuracy of machine learning processing results, especially in the case of noise in the data.
A data processing method is adopted to obtain the data to be identified, including data that does not carry the tag and/or data of noise tags, and make predictions based on a preset prediction model, and obtain the prediction results of the data as the tag. This prediction model integrates N classifiers, which are selected by semi-supervised learning combined with greedy algorithms, and N is a positive integer greater than 2.
Through this method, the robustness of label prediction is enhanced, the interference of noise labels is effectively reduced, and the accuracy of credit approval data processing is improved.
Smart Images
Figure CN119598348A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of artificial intelligence, and in particular to a data processing method, apparatus, device, medium and program product. Background Art
[0002] With the development of technology in the credit approval process of banks, the application of machine learning to process credit application data is increasing. Although machine learning can process a large amount of credit application data, the accuracy of machine learning processing results is limited by the quality of the data set. In the credit approval scenario, the quality of credit data is not stable. Using machine learning methods to predict such noisy data will lead to the existence of results. Therefore, before conducting machine learning credit prediction, it is particularly important to correct the data with potential noise risks. Summary of the invention
[0003] In view of the above problems, the present disclosure provides a data processing method, apparatus, device, medium and program product for improving data processing.
[0004] According to a first aspect of the present disclosure, a data processing method is provided, comprising: obtaining data to be identified, the data to be identified including data without labels and / or data with noise labels; and predicting the data to be identified based on a preset prediction model, and obtaining a prediction result of the data to be identified as a label of the data to be identified, wherein the preset prediction model integrates N classifiers, and the classifiers are selected by a greedy algorithm in a semi-supervised learning process, wherein N is a positive integer greater than 2.
[0005] According to an embodiment of the present disclosure, the method for establishing the preset prediction model includes: obtaining a first training data set carrying real labels and a second training data set carrying noise labels; based on the first training data set and the second training data set, using semi-supervised learning to perform greater than or equal to two rounds of training rounds until a preset cutoff condition is met, to obtain L classifiers, where L is a positive integer greater than or equal to N; using a greedy algorithm to select the N classifiers from the L classifiers; and establishing a meta-classifier based on the N classifiers to form the prediction model.
[0006] According to an embodiment of the present disclosure, based on the first training data set and the second training data set, semi-supervised learning is used to perform greater than or equal to two rounds of training until a preset cutoff condition is met, including: for the first round of training, a first round classifier is obtained based on the first training data set and a model value corresponding to the first round classifier.
[0007] According to an embodiment of the present disclosure, the adopting a greedy algorithm to select the N classifiers from the L classifiers includes: selecting the first round classifier as one of the N classifiers.
[0008] According to an embodiment of the present disclosure, after the first round of classifiers and the model value corresponding to the first round of classifiers are obtained through training based on the first training data set, the first training data set and the second training data set are used to perform a number of training rounds greater than or equal to two rounds using semi-supervised learning until a preset cutoff condition is met, and further includes: for training other than the first round, repeatedly performing the following confidence prediction operations, training data screening operations, and semi-supervised training operations until the preset cutoff condition is met: the confidence prediction operation: predicting the second training data set based on the classifier generated in the previous round to obtain a confidence set of labels in the second training data set, wherein the classifier generated in the previous round includes the classifier of the first round and the classifier of the non-first round; the training data screening operation: selecting the confidence in the confidence set that meets the preset confidence screening rules and the corresponding training data as the third training data set; the semi-supervised training operation: training based on the third training data set to obtain the classifier generated in the current round and the model value corresponding to the classifier generated in the current round.
[0009] According to an embodiment of the present disclosure, the method of selecting the N classifiers from the L classifiers using a greedy algorithm also includes: comparing a model value corresponding to the classifier generated in the current round with a highest model value corresponding to a classifier generated in a historical round; and retaining the classifier generated in the current round as one of the N classifiers when the model value corresponding to the classifier generated in the current round is greater than the highest model value corresponding to the classifier generated in the historical round.
[0010] According to an embodiment of the present disclosure, the meta-classifier is established based on the N classifiers to form the prediction model, including: assigning weights to corresponding classifiers based on the model value; and a meta-classifier is established based on the weights and the N classifiers, and the weights are adjusted through a preset validation set to obtain the prediction model.
[0011] According to the embodiments of the present disclosure, the preset cutoff conditions include: the number of training iterations reaches a preset number of rounds, or the confidence that meets the preset confidence screening rules is less than a preset number; the preset confidence screening rules include: if it is greater than a preset confidence threshold, it passes the screening.
[0012] The second aspect of the present disclosure provides a data processing device, including: a data acquisition module, used to acquire data to be identified, the data to be identified including data without labels and / or data with noise labels; and a label prediction module, used to predict the data to be identified based on a preset prediction model, and obtain the prediction result of the data to be identified as the label of the data to be identified, wherein the preset prediction model integrates N classifiers, and the classifiers are selected based on a greedy algorithm in a semi-supervised learning process, wherein N is a positive integer greater than 2.
[0013] According to an embodiment of the present disclosure, the device also includes: a training module, the training module includes: a training data set acquisition unit, a training unit, a classifier selection unit and a prediction model establishment module, the training data set acquisition unit is used to acquire a first training data set carrying a real label and a second training data set carrying a noise label; the training unit is used to perform a number of training rounds greater than or equal to two rounds based on the first training data set and the second training data set using semi-supervised learning until a preset cutoff condition is met, to obtain L classifiers, wherein L is a positive integer greater than or equal to N; the classifier selection unit is used to select the N classifiers from the L classifiers using a greedy algorithm; and the prediction model establishment module is used to establish a meta-classifier based on the N classifiers and form the prediction model.
[0014] According to an embodiment of the present disclosure, the training unit is specifically used for training a first round of training, and obtaining a first round of classifiers and a model value corresponding to the first round of classifiers based on the first training data set.
[0015] According to an embodiment of the present disclosure, the classifier selection unit is specifically configured to select the first-round classifier as one of the N classifiers.
[0016] According to an embodiment of the present disclosure, after the first round of classifiers and the model value corresponding to the first round of classifiers are obtained through training based on the first training data set, the training unit is further specifically used to repeatedly perform the following confidence prediction operations, training data screening operations and semi-supervised training operations for training other than the first round, until the preset cutoff condition is reached: the confidence prediction operation: predicting the second training data set based on the classifier generated in the previous round to obtain a confidence set of labels in the second training data set, wherein the classifier generated in the previous round includes the classifier of the first round and the classifiers other than the first round; the training data screening operation: selecting the confidences and corresponding training data in the confidence set that meet the preset confidence screening rules as the third training data set; the semi-supervised training operation: training based on the third training data set to obtain the classifier generated in the current round and the model value corresponding to the classifier generated in the current round.
[0017] According to an embodiment of the present disclosure, the classifier selection unit is further specifically used to compare the model value corresponding to the classifier generated in the current round with the highest model value corresponding to the classifier generated in the historical round; and when the model value corresponding to the classifier generated in the current round is greater than the highest model value corresponding to the classifier generated in the historical round, retain the classifier generated in the current round as one of the N classifiers.
[0018] According to an embodiment of the present disclosure, the prediction model establishment module is specifically used to assign weights to corresponding classifiers based on the model value; and a meta-classifier established based on the weights and the N classifiers completes weight adjustment through a preset validation set to obtain the prediction model.
[0019] According to the embodiments of the present disclosure, the preset cutoff conditions include: the number of training iterations reaches a preset number of rounds, or the confidence that meets the preset confidence screening rules is less than a preset number; the preset confidence screening rules include: if it is greater than a preset confidence threshold, it passes the screening.
[0020] A third aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.
[0021] The fourth aspect of the present disclosure further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the above computer program or instructions are executed by a processor.
[0022] The fifth aspect of the present disclosure further provides a computer program product, including a computer program or instructions, which implement the steps of the above method when the above computer program or instructions are executed by a processor.
[0023] In the embodiments of the present disclosure, in order to solve the technical problem of low data quality of the approval data set, the present application predicts and identifies data labels through a pre-set prediction model, which integrates multiple different base classifiers, and each base classifier is obtained through semi-supervised learning combined with a greedy algorithm. The embodiments of the present disclosure adopt the idea of combining greedy thinking and ensemble learning thinking with semi-supervised learning, and at the same time treats the noise label data as unlabeled data for processing, which enhances the robustness of label prediction while effectively reducing the interference of noise labels. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The above contents and other purposes, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:
[0025] Figure 1 The application scenario diagram of the data processing method according to the embodiment of the present disclosure is schematically shown;
[0026] Figure 2 A flowchart schematically shows a data processing method according to an embodiment of the present disclosure;
[0027] Figure 3 A flow chart schematically shows a model building method according to an embodiment of the present disclosure;
[0028] Figure 4 A flow chart of a model building method according to an embodiment of the present disclosure is schematically shown;
[0029] Figure 5 A structural block diagram schematically shows a data processing device according to an embodiment of the present disclosure; and
[0030] Figure 6 A block diagram of an electronic device suitable for implementing a data processing method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION
[0031] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.
[0032] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.
[0033] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.
[0034] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).
[0035] Before the embodiments of the present disclosure are disclosed in detail, the key technical terms in the embodiments of the present disclosure are explained one by one, as shown below:
[0036] Greedy algorithm: also known as greedy algorithm, refers to an algorithm that takes the best or optimal choice in each step when solving a problem, so that the result is the best or optimal algorithm.
[0037] Semi-supervised learning: is a machine learning method that lies between supervised learning and unsupervised learning. Its goal is to use unlabeled data to obtain a better performing model. Semi-supervised learning combines a small amount of labeled data with a large amount of unlabeled data for learning. This method is particularly suitable for situations where labeled data is scarce, and can effectively use unlabeled data to improve the generalization ability and prediction accuracy of the model.
[0038] Ensemble learning: Ensemble learning solves a single prediction problem by building several models. Its working principle is to generate multiple classifiers / models, each of which learns and makes predictions independently. These predictions are finally combined into a combined prediction. The accuracy and generalization of ensemble learning can be better than the previous single-classification prediction. Ensemble learning can generally be divided into: bagging, boosting and stacking.
[0039] Meta-classifier: A meta-classifier is a classifier that provides additional data processing. Ensemble learning combines different classifiers into a meta-classifier, which can generally be used in bagging, stacking and other methods in ensemble learning.
[0040] In the technical solution of the present disclosure, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all 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 the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.
[0041] In the scenario of using personal information for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure provide users with corresponding operation portals for users to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating a person's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs, and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge and skills, and have reached a certain level of professionalism.
[0042] In today's financial field, banks, as core institutions of credit business, play an irreplaceable role in promoting economic development and maintaining financial stability. In order to cope with the increasingly complex and changing market environment and meet customer needs, the banking industry is actively increasing investment and application of new technologies. By introducing advanced technologies such as machine learning and big data analysis, business processes are optimized, service efficiency is improved, and risk prevention and control are strengthened.
[0043] Taking the credit approval scenario as an example, although the traditional manual loan review ensures the security of credit business to a certain extent, it also has obvious limitations. These limitations are mainly reflected in the cumbersome approval process, low efficiency, strong subjectivity and difficulty in dealing with complex credit risks.
[0044] To overcome these limitations, banks have begun to actively explore and apply new technologies, especially machine learning, to optimize the credit approval process and improve the accuracy of risk assessment. Machine learning uses algorithms to analyze and learn large amounts of data, automatically discovering patterns and regularities in the data, and providing more objective, accurate, and efficient decision support for credit approval.
[0045] In predicting bank credit approval, the important role of machine learning is mainly reflected in the following aspects:
[0046] Improve approval efficiency: Machine learning can automatically process large amounts of credit application data and quickly give approval results, greatly reducing the workload and time cost of manual review; Reduce subjectivity: Machine learning is data-driven, reducing the impact of human factors on approval results, making approval results more objective and fair; Improve risk assessment accuracy: Machine learning algorithms can analyze data in multiple dimensions, including applicants’ credit records, financial status, behavioral patterns, etc., to more accurately assess applicants’ default risks.
[0047] However, in practical applications, the accuracy of machine learning prediction results is also limited by the quality of the data set. In bank credit approval scenarios, there are often only a small number of clean labeled data sets (i.e., sample data with known credit approval results), while a large amount of data has noisy labels (i.e., some borrowers cannot repay on time, or the labels are inaccurate due to missing sample data). This data class imbalance and noise interference problem will have an adverse effect on the training of machine learning models, causing the model to have deviations or errors in prediction. However, previous work has not focused on the identification of credit approval result labels, resulting in erroneous loan approval labels being used as input for various support services. Manual identification is a costly and time-consuming task.
[0048] In order to solve the technical problems existing in the prior art, an embodiment of the present invention provides a data processing method, which includes: obtaining data to be identified, wherein the data to be identified includes data without labels and / or data with noise labels; and predicting the data to be identified based on a preset prediction model, and obtaining a prediction result of the data to be identified as a label of the data to be identified, wherein the preset prediction model integrates N classifiers, and the classifiers are selected by a greedy algorithm in a semi-supervised learning process, wherein N is a positive integer greater than 2.
[0049] In the embodiments of the present disclosure, in order to solve the technical problem of low data quality of the approval data set, the present application predicts and identifies data labels through a pre-set prediction model, which integrates multiple different base classifiers, and each base classifier is obtained through semi-supervised learning combined with a greedy algorithm. The embodiments of the present disclosure adopt the idea of combining greedy thinking and ensemble learning thinking with semi-supervised learning, and at the same time treats the noise label data as unlabeled data for processing, which enhances the robustness of label prediction while effectively reducing the interference of noise labels.
[0050] Figure 1 The application scenario diagram of the data processing method according to the embodiment of the present disclosure is schematically shown.
[0051] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used to provide a medium for a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.
[0052] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).
[0053] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.
[0054] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.
[0055] It should be noted that the data processing method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the data processing device provided in the embodiment of the present disclosure can generally be set in the server 105. The data processing method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the data processing device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.
[0056] It should be understood that Figure 1The 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 the implementation requirements.
[0057] The following will be based on Figure 1 The scene described by Figure 2~Figure 4 The data processing method of the disclosed embodiment is described in detail.
[0058] Figure 2 The flowchart of the data processing method according to the embodiment of the present disclosure is schematically shown.
[0059] like Figure 2 As shown, the data processing method of this embodiment includes operations S210 to S220, and the data processing method can be executed by the server 105.
[0060] In operation S210 , data to be identified is acquired, where the data to be identified includes data without a label and / or data with a noise label.
[0061] In an embodiment of the present disclosure, before obtaining the user's information, the user's consent or authorization may be obtained. For example, before operation S220, a request to obtain the user's information may be issued to the user. When the user agrees or authorizes that the user's information may be obtained, the operation S210 is performed.
[0062] The data to be identified is the user's data, and the user can be a corporate user or a natural person user.
[0063] Specifically, in the field of credit approval, ideally, machine learning obtains clean data of users, and the machine learning model directly implements credit prediction on users through the clean data and calculates user credit data. However, in the specific implementation scenario, the acquired user data has more or less some missing or noise, which greatly affects the authenticity of the data. Therefore, before making credit predictions for users, it is necessary to identify the authenticity of the content of these user data, that is, by performing authenticity recognition on the user data (that is, the above-mentioned data to be identified), which is specifically manifested in the form of labeling. The label can be a binary label, that is, true or false / passed or rejected, or a label that implements confidence on the label, where the confidence is in the range of 0~1. In the embodiments of the present disclosure, there are many data sources for user data, and some data may be data that has been labeled. The authenticity of these data labels also needs to be further judged. Therefore, the above-mentioned data to be identified includes data without labels and / or data with noise labels.
[0064] In a typical scenario, such as a credit approval scenario, the data to be identified is the approval feature, which is the feature of the input data, including basic information, occupational income, financial status, credit history, asset status, etc. The approval label is the real classification of credit approval (approved or rejected), which is specifically represented by the manually marked label in the clean data set. The approval label in the data set is 1 for approval and 0 for rejection.
[0065] In operation S220, the data to be identified is predicted based on a preset prediction model, and a prediction result of the data to be identified is obtained as a label of the data to be identified, wherein the preset prediction model integrates N classifiers, and the classifiers are selected based on a greedy algorithm in a semi-supervised learning process, wherein N is a positive integer greater than 2.
[0066] In the embodiments of the present disclosure, a corresponding operation entry may be provided for the user to choose to agree or reject the automated decision result. That is, before data processing / decision making is performed on the user information, an instruction for agreeing or rejecting the processing / decision making input by the user through the corresponding operation entry may be obtained. If the user agrees to perform the processing / decision making, data processing / decision making is performed on the user information, i.e., step S220 is executed. If the user rejects the processing / decision making, the expert decision making process is entered.
[0067] Specifically, the prediction model predicts the label of the user's data to be identified. The embodiment of the present disclosure aims to provide a targeted processing method for the label of the credit data to be identified, and solves the problems of existing data class imbalance and data class labels being noise labels through an integrated semi-supervised learning algorithm based on greedy selection. To this end, the prediction model adopted is based on the establishment of a greedy semi-supervised algorithm for ensemble learning, and the scheme can also be called (Ensemblegreedy semi-supervised classifier, referred to as EGSSC). Among them, the prediction model is a learning model (learning model) obtained after the supervised classifier is trained. It is a mapping from input to output: f:X→Y, and the mapping f is obtained by learning {(x1, y1), (x2, y2), ..., (xd, yd)} in the training set (in the embodiment of the present disclosure, x is the approval feature and y is the approval label), so as to use f to predict its accurate label yn according to the noise label data feature xn. The model functions generated by training different supervised classifiers are the same, but the information stored in the model is very different. In the specific implementation, the model is essentially an object composed of the data structure used by the supervised classifier, the manually input hyperparameters, and the weights obtained through training. Subsequently, the data labels identified by the prediction model can be directly used in the credit measurement model as input data to realize the credit prediction of users.
[0068] It should be noted that the common semi-supervised method will iteratively call the supervised classifier, and in each iteration, select some data from the unlabeled data set and add it to the training set. The semi-supervised method used in the embodiment of the present disclosure will also iteratively call the supervised classifier for training. The scale of the training set will be expanded in each iteration, and data can be iteratively selected from the noise label data set and added to the training set.
[0069] In the embodiments of the present disclosure, in order to solve the technical problem of low data quality of the approval data set, the present application predicts and identifies data labels through a pre-set prediction model, which integrates multiple different base classifiers, and each base classifier is obtained through semi-supervised learning combined with a greedy algorithm. The embodiments of the present disclosure adopt the idea of combining greedy thinking and ensemble learning thinking with semi-supervised learning, and at the same time treats the noise label data as unlabeled data for processing, which enhances the robustness of label prediction while effectively reducing the interference of noise labels.
[0070] The following is a detailed description of the method for establishing the model in the embodiments of the present disclosure, as shown below:
[0071] Figure 3 The flowchart of the model building method according to the embodiment of the present disclosure is schematically shown.
[0072] like Figure 3 As shown, the data processing method of this embodiment includes operations S310 to S340, as shown below:
[0073] In operation S310 , a first training data set carrying a true label and a second training data set carrying a noise label are obtained.
[0074] Among them, the true label is the real historical data, and the noise label includes the real historical data, false data and some missing data.
[0075] In operation S320, based on the first training data set and the second training data set, semi-supervised learning is used to perform two or more training rounds until a preset cutoff condition is met, and L classifiers are obtained, where L is a positive integer greater than or equal to N.
[0076] During the semi-supervised learning process, since new training data sets need to be generated iteratively, at least two rounds of training are performed.
[0077] According to an embodiment of the present disclosure, based on the first training data set and the second training data set, semi-supervised learning is used to perform greater than or equal to two rounds of training until a preset cutoff condition is met, including: for the first round of training, a first round classifier is obtained based on the first training data set and a model value corresponding to the first round classifier.
[0078] Among them, the model value is the indicator value for evaluating model performance, such as the commonly used AUC value as an indicator value for evaluating performance models.
[0079] Specifically, the classifier established in the first round is also called the basic model, which is the model obtained by the semi-supervised algorithm using the supervised classifier and the training set T before the iteration process begins. At this time, the training set T is in the initial state, that is, the training set only uses the clean data in the first training data set and is not modified by any new training data in the semi-supervised algorithm. In addition, the performance of the basic model will serve as the judgment basis for the greedy selection of the semi-supervised algorithm in the iteration.
[0080] According to an embodiment of the present disclosure, after the first round of classifiers and the model value corresponding to the first round of classifiers are obtained through training based on the first training data set, the first training data set and the second training data set are used to perform a number of training rounds greater than or equal to two rounds using semi-supervised learning until a preset cutoff condition is met, and further includes: for training other than the first round, repeatedly performing the following confidence prediction operations, training data screening operations, and semi-supervised training operations until the preset cutoff condition is met: the confidence prediction operation: predicting the second training data set based on the classifier generated in the previous round to obtain a confidence set of labels in the second training data set, wherein the classifier generated in the previous round includes the classifier of the first round and the classifier of the non-first round; the training data screening operation: selecting the confidence in the confidence set that meets the preset confidence screening rules and the corresponding training data as the third training data set; the semi-supervised training operation: training based on the third training data set to obtain the classifier generated in the current round and the model value corresponding to the classifier generated in the current round.
[0081] According to an embodiment of the present disclosure, the preset cutoff condition includes: the number of training iterations reaches a preset number of rounds, or the confidence that meets the preset confidence screening rule is less than a preset number; the preset confidence screening rule includes: if it is greater than a preset confidence threshold, it passes the screening.
[0082] The preset medium condition for training is that the preset number of iterations or the amount of training data with higher confidence cannot support subsequent training. The preset confidence screening rule is less than the preset threshold, such as: the confidence threshold is set to 0.8, then the data with a confidence exceeding 0.8 is retained, and the data with a confidence less than 0.8 is discarded and does not participate in the subsequent generation of a new training set.
[0083] Specifically, after the basic model is trained, the subsequent generative model adopts a semi-supervised learning method, combining a small amount of labeled data and a large amount of unlabeled data for learning. In each round, the training data with higher confidence is selected as the training data for the next round to achieve iterative training.
[0084] In this process, for the newly generated third training set, the positive and negative samples of the third training data set can be balanced, that is, the data labeled with 0 and 1 are ensured to be half each, to avoid data skew.
[0085] In operation S330, a greedy algorithm is used to select the N classifiers from the L classifiers.
[0086] Specifically, the above N classifiers serve as a model set that forms a meta-classifier, which is an abstract set of models trained by N supervised classifiers. The models in the model set are all generated during the iteration process of the semi-supervised algorithm, and the models that meet the performance requirements will be included in the model set.
[0087] According to an embodiment of the present disclosure, the adopting a greedy algorithm to select the N classifiers from the L classifiers includes: selecting the first round classifier as one of the N classifiers.
[0088] According to an embodiment of the present disclosure, the method of selecting the N classifiers from the L classifiers using a greedy algorithm also includes: comparing a model value corresponding to the classifier generated in the current round with a highest model value corresponding to a classifier generated in a historical round; and retaining the classifier generated in the current round as one of the N classifiers when the model value corresponding to the classifier generated in the current round is greater than the highest model value corresponding to the classifier generated in the historical round.
[0089] Specifically, the above greedy algorithm mainly selects the classifier model whose model value is greater than the highest previous value among the model values obtained in history during the training iteration. Among them, the model value of the first basic model (i.e., the classifier trained in the first round) selected by the greedy algorithm is used as the benchmark value, which is used as the basis for starting subsequent selection.
[0090] It should be noted that in order to solve the class imbalance problem, the criteria for screening data, in addition to the confidence level meeting the requirements, must also ensure that the number of data with the class label passed and the number of data with the class label rejected are consistent in each new training set. This way, the size of the training set can be expanded while maintaining the class balance of the training set data.
[0091] In operation S340, a meta classifier is established based on the N classifiers to form the prediction model.
[0092] Specifically, a meta-classifier is a classifier that provides additional data processing, and is generally used in methods such as bagging and stacking in ensemble learning. The meta-classifier in the embodiments of the present disclosure may use the stacking method to take the output value of each model in the model set as its input value. The model generated by the meta-classifier contains the weight of each model, thereby achieving the purpose of integrating all models.
[0093] According to an embodiment of the present disclosure, the meta-classifier is established based on the N classifiers to form the prediction model, including: assigning weights to corresponding classifiers based on the model value; and a meta-classifier is established based on the weights and the N classifiers, and the weights are adjusted through a preset validation set to obtain the prediction model.
[0094] Specifically, for the meta-classifier to the final prediction model, the model can be trained by the meta-classifier using the validation set and the model set. The weights of different models in the model set are obtained through training. The prediction value output by the final model is also the final prediction value of the entire semi-supervised method.
[0095] In the embodiments of the present disclosure, the EGSSC method uses the idea of ensemble learning. Unlike the traditional self-training method, which only retains the model generated in the last iteration after the algorithm runs, it combines multiple models generated during the training process to form a model set. After the iteration, the meta-classifier will use the output of all models in the model set as its input, and use the validation set to train the weights of each model to achieve the purpose of integrating the model. Because ensemble learning can reduce the overall error by averaging the variance of multiple single models, the ensemble learning method is usually more robust. In addition, the EGSSC method also uses the greedy idea in the selection of models in the model set. In the process of EGSSC iteration, each iteration will generate a model, but not every model will be included in the model set in the end. Only new models with improved performance will be included in the model set of EGSSC.
[0096] The following is a specific example of the EGSSC method, as shown below:
[0097] Figure 4 A flow chart of a model building method according to an embodiment of the present disclosure is schematically shown.
[0098] like Figure 4 As shown, the input is: clean label dataset CL and noisy label dataset NL; the output is: model set and final model.
[0099] Step 1: Use the clean label dataset to train a basic model, add it to the model set C, and calculate the AUC value of the model.
[0100] Step 2: Use the newly generated model to predict the noise label dataset NL, and filter out the data with confidence that meets the requirements and keep the number of the two labels consistent, and add them to the training set.
[0101] Step 3: Use the new training set to train a new model and test the AUC value of the new model. If the AUC value is higher than the previous value, update the AUC value and add the model to the model set C.
[0102] Step4: Iterate Step2 and Step3 according to the confidence list and the newly added data list confList and numList.
[0103] Step 5: After the iteration, the meta-classifier is used to integrate the model set to form the final prediction model.
[0104] The confList list stores the increasing confidence threshold conf, and the numList list stores the increasing number of iterations num. EGSSC improves the data access requirements for adding new training sets by increasing the confidence threshold conf and the number of iterations num.
[0105] In each iteration, the model accumulation function Accmodel() will learn the model on the training set T and apply the model to the noise label data set NL. It will generate the prediction result temp, and then filter out the data with a prediction probability higher than the confidence threshold conf to form temp'. Then randomly extract 1 / 2num of the data labeled "approved" from temp', and similarly randomly extract 1 / 2num of the data labeled "approved" from temp'. Then remove them from the NL data set, merge them into the training set and use them as training data for subsequent iterations, and use the predicted labels as their label values. The model accumulation function will continue to iterate until the amount of data that meets the conditions is insufficient or the maximum number of iterations is reached.
[0106] In the embodiments of the present disclosure, the idea of combining greedy thinking and ensemble learning with semi-supervised learning is adopted, and the noise label data is treated as unlabeled data for processing, which enhances the robustness of label prediction while effectively reducing the interference of noise labels. Accurate identification of loan approval result labels can provide more accurate data support for banking data analysis scenarios, provide more accurate and efficient decision support for credit approval business, and avoid the disadvantages of high cost and long time of manual identification.
[0107] Based on the above data processing method, the present disclosure also provides a data processing device. Figure 5 The device is described in detail.
[0108] Figure 5 The structure block diagram of a data processing device according to an embodiment of the present disclosure is schematically shown.
[0109] like Figure 5As shown, the data processing device 500 of this embodiment includes a data acquisition module 510 and a label prediction module 520 .
[0110] The data acquisition module 510 is used to acquire the data to be identified, and the data to be identified includes data without labels and / or data with noise labels. In one embodiment, the data acquisition module 510 can be used to perform the operation S210 described above, which will not be repeated here.
[0111] The label prediction module 520 is used to predict the data to be identified based on a preset prediction model, and obtain the prediction result of the data to be identified as the label of the data to be identified, wherein the preset prediction model integrates N classifiers, and the classifiers are selected based on the greedy algorithm in the process of semi-supervised learning, wherein N is a positive integer greater than 2. In one embodiment, the label prediction module 520 can be used to perform the operation S220 described above, which will not be repeated here.
[0112] In the embodiments of the present disclosure, in order to solve the technical problem of low data quality of the approval data set, the present application predicts and identifies data labels through a pre-set prediction model, which integrates multiple different base classifiers, and each base classifier is obtained through semi-supervised learning combined with a greedy algorithm. The embodiments of the present disclosure adopt the idea of combining greedy thinking and ensemble learning thinking with semi-supervised learning, and at the same time treats the noise label data as unlabeled data for processing, which enhances the robustness of label prediction while effectively reducing the interference of noise labels.
[0113] According to an embodiment of the present disclosure, the device also includes: a training module, the training module includes: a training data set acquisition unit, a training unit, a classifier selection unit and a prediction model establishment module, the training data set acquisition unit is used to acquire a first training data set carrying a real label and a second training data set carrying a noise label; the training unit is used to perform a number of training rounds greater than or equal to two rounds based on the first training data set and the second training data set using semi-supervised learning until a preset cutoff condition is met, to obtain L classifiers, wherein L is a positive integer greater than or equal to N; the classifier selection unit is used to select the N classifiers from the L classifiers using a greedy algorithm; and the prediction model establishment module is used to establish a meta-classifier based on the N classifiers and form the prediction model.
[0114] According to an embodiment of the present disclosure, the training unit is specifically used for training a first round of training, and obtaining a first round of classifiers and a model value corresponding to the first round of classifiers based on the first training data set.
[0115] According to an embodiment of the present disclosure, the classifier selection unit is specifically configured to select the first-round classifier as one of the N classifiers.
[0116] According to an embodiment of the present disclosure, after the first round of classifiers and the model value corresponding to the first round of classifiers are obtained through training based on the first training data set, the training unit is further specifically used to repeatedly perform the following confidence prediction operations, training data screening operations and semi-supervised training operations for training other than the first round, until the preset cutoff condition is reached: the confidence prediction operation: predicting the second training data set based on the classifier generated in the previous round to obtain a confidence set of labels in the second training data set, wherein the classifier generated in the previous round includes the classifier of the first round and the classifiers other than the first round; the training data screening operation: selecting the confidences and corresponding training data in the confidence set that meet the preset confidence screening rules as the third training data set; the semi-supervised training operation: training based on the third training data set to obtain the classifier generated in the current round and the model value corresponding to the classifier generated in the current round.
[0117] According to an embodiment of the present disclosure, the classifier selection unit is further specifically used to compare the model value corresponding to the classifier generated in the current round with the highest model value corresponding to the classifier generated in the historical round; and when the model value corresponding to the classifier generated in the current round is greater than the highest model value corresponding to the classifier generated in the historical round, retain the classifier generated in the current round as one of the N classifiers.
[0118] According to an embodiment of the present disclosure, the prediction model establishment module is specifically used to assign weights to corresponding classifiers based on the model value; and a meta-classifier established based on the weights and the N classifiers completes weight adjustment through a preset validation set to obtain the prediction model.
[0119] According to the embodiments of the present disclosure, the preset cutoff conditions include: the number of training iterations reaches a preset number of rounds, or the confidence that meets the preset confidence screening rules is less than a preset number; the preset confidence screening rules include: if it is greater than a preset confidence threshold, it passes the screening.
[0120] According to an embodiment of the present disclosure, any multiple modules in the data acquisition module 510 and the label prediction module 520 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the data acquisition module 510 and the label prediction module 520 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the data acquisition module 510 and the label prediction module 520 can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be performed.
[0121] Figure 6 A block diagram of an electronic device suitable for implementing a data processing method according to an embodiment of the present disclosure is schematically shown.
[0122] like Figure 6 As shown, the electronic device 600 according to an embodiment of the present disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 to a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include an onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.
[0123] In RAM 603, various programs and data required for the operation of electronic device 600 are stored. Processor 601, ROM 602 and RAM 603 are connected to each other via bus 604. Processor 601 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 602 and / or RAM 603. It should be noted that the program can also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in the one or more memories.
[0124] According to an embodiment of the present disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to the bus 604. The electronic device 600 may further include one or more of the following components connected to the input / output (I / O) interface 605: an input portion 606 including a keyboard, a mouse, etc.; an output portion 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 608 including a hard disk, etc.; and a communication portion 609 including a network interface card such as a LAN card, a modem, etc. The communication portion 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage portion 608 as needed.
[0125] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.
[0126] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 602 and / or RAM 603 described above and / or one or more memories other than ROM 602 and RAM 603.
[0127] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiment of the present disclosure.
[0128] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 601. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.
[0129] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 609, and / or installed from a removable medium 611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.
[0130] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the processor 601, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, apparatus, module, unit, etc. described above can be implemented by a computer program module.
[0131] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).
[0132] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0133] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.
[0134] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.
Claims
1. A data processing method, characterized in that: The method comprises: Acquiring data to be identified, wherein the data to be identified includes data without a label and / or data with a noise label; and Based on a preset prediction model, the data to be identified is predicted, and a prediction result of the data to be identified is obtained as a label of the data to be identified, wherein the preset prediction model integrates N classifiers, and the classifiers are selected by a greedy algorithm in a semi-supervised learning process, wherein N is a positive integer greater than 2.
2. The method according to claim 1, characterized in that The method for establishing the preset prediction model includes: Obtain a first training data set carrying real labels and a second training data set carrying noise labels; Based on the first training data set and the second training data set, performing two or more training rounds of training using semi-supervised learning until a preset cutoff condition is met, to obtain L classifiers, where L is a positive integer greater than or equal to N; Selecting the N classifiers from the L classifiers using a greedy algorithm; and A meta-classifier is established based on the N classifiers to form the prediction model.
3. The method according to claim 2, characterized in that The step of performing two or more training rounds of training based on the first training data set and the second training data set using semi-supervised learning until a preset cutoff condition occurs includes: For the first round of training, a first round classifier and a model value corresponding to the first round classifier are obtained based on the first training data set.
4. The method according to claim 3, characterized in that The adopting a greedy algorithm to select the N classifiers from the L classifiers includes: The first round classifier is selected as one of the N classifiers.
5. The method according to claim 3 or 4, characterized in that: After obtaining a first-round classifier and a model value corresponding to the first-round classifier through training based on the first training data set, The method of performing two or more training rounds of training based on the first training data set and the second training data set using semi-supervised learning until a preset cutoff condition is met further includes: For training other than the first round, the following confidence prediction operation, training data screening operation, and semi-supervised training operation are repeatedly performed until the preset cutoff condition is reached: The confidence prediction operation: predicting the second training data set based on the classifier generated in the previous round to obtain a confidence set of labels in the second training data set, wherein the classifier generated in the previous round includes the classifier of the first round and the classifier not in the first round; The training data screening operation: selecting the confidences in the confidence set that meet the preset confidence screening rules and the corresponding training data as the third training data set; The semi-supervised training operation: training based on the third training data set to obtain the classifier generated in the current round and the model value corresponding to the classifier generated in the current round.
6. The method according to claim 5, characterized in that The adopting a greedy algorithm to select the N classifiers from the L classifiers further includes: Compare the model value corresponding to the classifier generated in the current round with the highest model value corresponding to the classifier generated in the previous round; and When the model value corresponding to the classifier generated in the current round is greater than the highest model value corresponding to the classifier generated in the previous round, the classifier generated in the current round is retained as one of the N classifiers.
7. The method according to any one of claims 3, 4 and 6, characterized in that The step of establishing a meta-classifier based on the N classifiers and forming the prediction model comprises: Assigning weights to corresponding classifiers based on the model value; and The meta-classifier established based on the weights and the N classifiers completes weight adjustment through a preset validation set to obtain the prediction model.
8. The method according to claim 5, characterized in that The preset cutoff condition includes: the number of training iterations reaches a preset number of rounds, or the confidence that meets the preset confidence screening rule is less than a preset number; The preset confidence screening rule includes: if the confidence is greater than a preset confidence threshold, the screening is passed.
9. A data processing device, characterized in that: The device comprises: A data acquisition module, used to acquire data to be identified, wherein the data to be identified includes data without labels and / or data with noise labels; and A label prediction module is used to predict the data to be identified based on a preset prediction model, and obtain the prediction result of the data to be identified as the label of the data to be identified, wherein the preset prediction model integrates N classifiers, and the classifiers are selected based on a greedy algorithm in a semi-supervised learning process, wherein N is a positive integer greater than 2.
10. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.
11. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.
12. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.