Abnormal business identification model training method and device, computer equipment, readable storage medium and program product

Abstract

The invention relates to a training method and device of an abnormal service identification model, computer equipment, a readable storage medium and a program product. And performing first training on a to-be-trained abnormal service identification model through the initial service information negative sample and the initial service information positive sample, and identifying a label-free service information sample based on a first abnormal service identification model obtained by the first training to obtain a corresponding service type identification result, and based on the service type identification result and the data set, performing second training on the first abnormal service identification model until a preset training ending condition is satisfied, and obtaining a trained abnormal service identification model. Compared with a traditional method for training through a large number of clearly-labeled positive and negative samples, the method has the advantages that the initial samples with the labels and the samples without the labels are combined, the abnormal service recognition model is subjected to multiple iterative training, the requirement for the label samples is reduced, and the training efficiency of the abnormal service recognition model is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a training method, apparatus, computer equipment, computer-readable storage medium, and computer program product for an abnormal business identification model. Background Art

[0002] With the development of business scale, financial services have been widely integrated into people's lives. In order to meet the financial business needs of users, financial services need to be managed. When users handle financial services, there are cases where abnormal business handling is carried out for false reasons. In order to ensure the business processing security of enterprises and users, it is necessary to accurately identify abnormal businesses. At present, the way to identify abnormal businesses is usually through artificial intelligence models. However, the current artificial intelligence models need to use a large number of clearly labeled positive and negative samples for training, which reduces the training efficiency of the model.

[0003] Therefore, the current training method for the model for identifying abnormal business has the defect of low training efficiency. Summary of the invention

[0004] Based on this, it is necessary to provide a training method, device, computer equipment, computer-readable storage medium and computer program product for an abnormal business identification model that can improve the efficiency of model training in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for training an abnormal business identification model, the method comprising:

[0006] Acquire an abnormal business identification model to be trained and a corresponding data set; the data set includes initial business information negative samples carrying negative sample labels and initial business information positive samples carrying positive sample labels;

[0007] Based on the initial business information negative sample and the initial business information positive sample in the data set, performing a first training on the abnormal business identification model to obtain a first abnormal business identification model;

[0008] Acquire an unlabeled service information sample, input the service information sample into the first abnormal service identification model, and have the first abnormal service identification model output a corresponding service type identification result based on the service information sample;

[0009] According to the service type identification result and the data set, the first abnormal service identification model is subjected to second training until a preset training end condition is met, thereby obtaining a trained abnormal service identification model.

[0010] In one embodiment, the step of obtaining a data set corresponding to abnormal business identification to be trained includes:

[0011] Obtain each unlabeled business information sample and each initial business information positive sample carrying a positive sample label;

[0012] According to the abnormal information of each of the business information samples, each initial business information negative sample in each of the business information samples is determined and a negative sample label is added to obtain each initial business information negative sample carrying a negative sample label;

[0013] The data set is generated according to each of the initial business information negative samples and each of the initial business information positive samples.

[0014] In one embodiment, determining each initial business information negative sample in each business information sample according to the abnormal information of each business information sample includes:

[0015] Marking each of the business information samples as a candidate business information negative sample, and constructing a classifier according to each of the candidate business information negative samples;

[0016] The classifier classifies each of the business information samples according to the abnormal information, determines each of the initial business information negative samples according to the classification result and adds a negative sample label to obtain each of the initial business information negative samples carrying the negative sample label.

[0017] In one embodiment, determining each initial business information negative sample in each business information sample according to the abnormal information of each business information sample includes:

[0018] Scoring each of the business information samples according to a preset experience database and each of the abnormal information of each of the business information samples to obtain each sample score corresponding to each of the business information samples;

[0019] Each of the initial business information negative samples is determined according to the sample score.

[0020] In one embodiment, the inputting the service information sample into the first abnormal service identification model, and the first abnormal service identification model outputting a corresponding service type identification result based on the service information sample, includes:

[0021] Inputting the service information sample into the first abnormal service identification model, and allowing the first abnormal service identification model to identify the service type corresponding to the service information sample and the confidence level corresponding to the service type;

[0022] Determine a corresponding prediction label according to the business type;

[0023] According to the prediction label and the confidence level, a business type recognition result corresponding to the business information sample is output.

[0024] In one embodiment, the second training of the first abnormal business identification model is performed according to the business type identification result and the data set until a preset training end condition is met to obtain a trained abnormal business identification model, including:

[0025] Determining, according to the business type identification result, a target business information sample whose confidence level is greater than a preset threshold;

[0026] Expanding the data set according to the target business information sample carrying the prediction label to obtain a new data set;

[0027] Inputting the new data set into the first abnormal business identification model, and having the first abnormal business identification model output a new business type identification result based on the new data set;

[0028] Adjust the model parameters of the first abnormal business identification model according to the new business type identification result, and return to the step of determining the target business information sample whose confidence is greater than a preset threshold according to the business type identification result, until the preset training end condition is met, to obtain a trained abnormal business identification model.

[0029] In a second aspect, the present application provides a training device for an abnormal business identification model, the device comprising:

[0030] An acquisition module is used to acquire an abnormal business identification model to be trained and a corresponding data set; the data set includes initial business information negative samples carrying negative sample labels and initial business information positive samples carrying positive sample labels;

[0031] A first training module, configured to perform a first training on the abnormal business identification model based on the initial business information negative sample and the initial business information positive sample in the data set to obtain a first abnormal business identification model;

[0032] An input module, used to obtain an unlabeled business information sample, input the business information sample into the first abnormal business identification model, and the first abnormal business identification model outputs a corresponding business type identification result based on the business information sample;

[0033] The second training module is used to perform a second training on the first abnormal business identification model according to the business type identification result and the data set until a preset training end condition is met, thereby obtaining a trained abnormal business identification model.

[0034] In a third aspect, the present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0035] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0036] In a fifth aspect, the present application provides a computer program product, including a computer program, which implements the steps of the above method when executed by a processor.

[0037] The training method, apparatus, computer equipment, computer-readable storage medium and computer program product of the abnormal business identification model mentioned above perform a first training on the abnormal business identification model to be trained through initial business information negative samples and initial business information positive samples, identify the unlabeled business information samples based on the first abnormal business identification model obtained by the first training, and obtain the corresponding business type identification results, and perform a second training on the first abnormal business identification model based on the business type identification results and the data set, until the preset training end conditions are met, and a trained abnormal business identification model is obtained. Compared with the traditional training through a large number of clearly labeled positive and negative samples, this solution combines labeled initial samples and unlabeled samples to perform multiple iterative training on the abnormal business identification model, which reduces the demand for labeled samples and improves the training efficiency of the abnormal business identification model. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0039] FIG1 is a schematic diagram of a flow chart of a method for training an abnormal business identification model in one embodiment;

[0040] FIG2 is a schematic diagram of a training method for an abnormal business identification model in one embodiment;

[0041] FIG3 is a structural block diagram of a training device for an abnormal business identification model in one embodiment;

[0042] FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0043] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0044] In one embodiment, as shown in FIG1 , a training method for an abnormal business identification model is provided. This embodiment uses the method applied to a server as an example for illustration. It is understandable that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server, including the following steps S202 to S208. Among them:

[0045] Step S202, obtaining an abnormal business identification model to be trained and a corresponding data set; the data set includes initial business information negative samples carrying negative sample labels and initial business information positive samples carrying positive sample labels.

[0046] The abnormal business identification model may be a model for identifying abnormal business in the financial field. To ensure the normal processing of financial business, it is necessary to train the abnormal business identification model to identify and process abnormal business in financial business. When training the abnormal business identification model, the server may obtain the abnormal business identification model to be trained and obtain the data set used to train the abnormal business identification model.

[0047] Among them, the server can obtain the initial business information negative sample and the initial business information positive sample, and construct the above data set according to the initial business information negative sample and the initial business information positive sample. The above initial business negative sample carries a negative sample label, and the above initial business information positive sample carries a positive sample label. The initial business information negative sample carrying the negative sample label represents a business information sample that is not an abnormal business, and the initial business information positive sample carrying the positive sample label represents a business information sample that is an abnormal business.

[0048] Step S204: Based on the initial business information negative sample and the initial business information positive sample in the data set, the abnormal business identification model is first trained to obtain a first abnormal business identification model.

[0049] The data set may include initial business information negative samples carrying negative sample labels and initial business information positive samples carrying positive sample labels. The server may use the initial business information negative samples and the initial business information positive samples in the data set to perform a first training on the abnormal business identification model to be trained, and obtain a first abnormal business identification model.

[0050] Among them, the above-mentioned first training can be a supervised training, and the server can use each business information sample with a sample label, such as the above-mentioned initial business information negative sample with a negative sample label and the initial business information positive sample with a positive sample label, to perform supervised iterative training on the abnormal business identification model to be trained, and then obtain a trained first abnormal business identification model. For example, the server inputs the above-mentioned initial business information negative sample with a negative sample label and the initial business information positive sample with a positive sample label into the abnormal business identification model to be trained, and the abnormal business identification model outputs the corresponding abnormal business identification result based on the initial business information negative sample with a negative sample label and the initial business information positive sample with a positive sample label, so that the server adjusts the model parameters of the abnormal business identification model according to the comparison between the abnormal business identification result and the corresponding business information negative sample or business information positive sample, until within the preset number of training times, the similarity between the above-mentioned abnormal business identification result and the corresponding business information negative sample or business information positive sample is greater than or equal to the preset similarity threshold, or the number of training times reaches the preset number of training times, and the first abnormal business identification model is obtained.

[0051] Specifically, the first abnormal business identification model can be an initial abnormal business identification model, and the abnormal business identification model can be a machine learning model. The data set can be a labeled data set. The server can perform initial training on the abnormal business identification model to be trained based on the labeled data set to obtain an initial first abnormal business identification model. For example, the server uses a supervised learning algorithm, including but not limited to a classification algorithm and a regression algorithm, to train the abnormal business identification model to be trained using a labeled data set to obtain a first abnormal business identification model. Among them, the abnormal business identification model can be trained based on any machine learning algorithm, such as a logistic regression algorithm, a random forest algorithm, and a neural network algorithm.

[0052] Step S206, obtaining a label-free business information sample, inputting the business information sample into the first abnormal business identification model, and having the first abnormal business identification model output a corresponding business type identification result based on the business information sample.

[0053] Among them, the unlabeled business information samples can be business information samples that do not carry negative sample labels or positive sample labels, that is, there can be negative samples or positive samples in these business information samples. The server can use the above-mentioned unlabeled business information samples to train the first abnormal business identification model. For example, the server inputs the business information sample into the above-mentioned first abnormal business identification model, and the above-mentioned first abnormal business identification model outputs the corresponding business type identification result based on the above-mentioned business information sample. Among them, the business type identification result includes the predicted label of the business type identified by the first abnormal business identification model for the business information sample and the confidence corresponding to the business type.

[0054] Step S208, performing a second training on the first abnormal business identification model according to the business type identification result and the data set, until a preset training end condition is met, thereby obtaining a trained abnormal business identification model.

[0055] After obtaining the service type recognition result based on the first abnormal service recognition model and the above-mentioned unlabeled service information sample, the server can perform a second training on the first abnormal service recognition model in combination with the service type recognition result and the above-mentioned data set until the preset training end condition is met, thereby obtaining a trained abnormal service recognition model. The preset training end condition includes but is not limited to the output result of the above-mentioned first abnormal service recognition model converging within the preset training times, or the training times reaching the preset training times.

[0056] Among them, the above-mentioned second training can be an unsupervised training. The server can determine the target business information samples with high confidence and carrying prediction labels based on the business type identification results, and then use the target business information samples carrying prediction labels, combined with the above-mentioned data set, to further train the first abnormal business identification model, so as to obtain a trained abnormal business identification model.

[0057] Among them, the server can use the trained abnormal business identification model to identify abnormalities in the business information of the financial business. For example, the server can input the business information to be identified into the trained abnormal business identification model, and the abnormal business identification model outputs the corresponding business type identification result based on the business information, including the business type corresponding to the business information, the prediction label corresponding to the business type, etc. Among them, the business type includes normal business and abnormal business. If the server determines that the business information is abnormal business based on the above abnormal business identification model, it can output the alarm information corresponding to the above abnormal business.

[0058] In the training method of the above abnormal business identification model, the abnormal business identification model to be trained is first trained by using the initial business information negative samples and the initial business information positive samples, and the unlabeled business information samples are identified based on the first abnormal business identification model obtained by the first training to obtain the corresponding business type identification results, and the first abnormal business identification model is secondly trained based on the business type identification results and the data set until the preset training end conditions are met to obtain the trained abnormal business identification model. Compared with the traditional training with a large number of clearly labeled positive and negative samples, this solution combines labeled initial samples and unlabeled samples to perform multiple iterative training on the abnormal business identification model, which reduces the demand for labeled samples and improves the training efficiency of the abnormal business identification model.

[0059] In one embodiment, the step of obtaining a data set corresponding to abnormal business identification to be trained includes: obtaining each unlabeled business information sample and each initial business information positive sample carrying a positive sample label; according to the abnormal information of each of the above business information samples, determining each initial business information negative sample in each of the above business information samples and adding a negative sample label to obtain each initial business information negative sample carrying a negative sample label; generating the above data set according to each of the above initial business information negative samples and each of the above initial business information positive samples.

[0060] In this embodiment, the server can construct a data set of business information samples with pre-labeled labels. The data set contains initial business information negative samples with negative sample labels and initial business information positive samples with positive sample labels. When constructing the data set, the server can obtain each unlabeled business information sample and each initial business information positive sample with positive sample labels. The initial business information positive sample can be obtained from a historical sample database. The historical sample database stores multiple business information samples with pre-labeled labels.

[0061] The server can determine each initial business information negative sample in each of the above business information samples based on the abnormal information of each of the above business information samples, and add a negative sample label to each of the initial business information negative samples to obtain each initial business information negative sample carrying a negative sample label. The above abnormal information may include information related to abnormal business in the business information sample. The more abnormal information there is, the greater the probability that the business sample information belongs to the initial business information negative sample. Thus, the server can generate the above data set based on each of the above initial business information negative samples and each of the above initial business information positive samples.

[0062] Through this embodiment, the server can use the initial business information negative samples and initial business information positive samples with pre-labeled labels to construct a corresponding data set, and use the data set to perform a first training on the abnormal business identification model, and then perform a second training on the first abnormal business identification model after the first training based on the unlabeled business information samples, thereby improving the training efficiency of the abnormal business identification model.

[0063] In one embodiment, based on the abnormal information of each of the above-mentioned business information samples, each initial business information negative sample in each of the above-mentioned business information samples is determined, including: marking each of the above-mentioned business information samples as a candidate business information negative sample, and constructing a classifier based on each of the above-mentioned candidate business information negative samples; the above-mentioned classifier classifies each of the above-mentioned business information samples according to the abnormal information, and determines each of the above-mentioned initial business information negative samples according to the results of the above-mentioned classification.

[0064] In this embodiment, when the server is labeling the initial business information negative samples, it can use a two-step iterative method to determine. For example, the server can mark each of the above business information samples as each candidate business information negative sample, and construct a classifier based on each of the above candidate business information negative samples. Among them, each candidate business information negative sample can represent a business information sample that is regarded as a negative sample, so that the server can analyze the content of abnormal information in each business information sample from the perspective of a negative sample through the above-constructed classifier. For example, the server uses the above-mentioned classifier to classify the abnormal information of each of the above business information samples to obtain the corresponding classification results. Among them, the classification results include each sample type corresponding to each business information sample, and the sample score of each sample type of each business information sample. The sample score represents the score of the above business information sample given by the classifier from the perspective of a negative sample. Thus, the server can determine each of the above initial business information negative samples based on the results of the above classification.

[0065] Specifically, the above-mentioned step of determining the data set can be regarded as a two-step iterative method. The server can regard all unlabeled business information samples as negative samples, that is, each candidate business information negative sample, so that the server builds a preliminary classifier based on each candidate business information negative sample. The server uses this classifier to score and classify the unlabeled business information samples to obtain the corresponding classification results, including the business type corresponding to the business information sample, and the score of the business information sample. The server can select a business information sample whose score is less than a first preset score threshold as an initial business information negative sample. Among them, the first preset score threshold can be set according to actual conditions. Thereby, the server can complete the selection of the initial business information negative sample. In addition, the above-mentioned initial business information negative sample can also be determined based on the input score.

[0066] In one embodiment, each initial business information negative sample in each of the above business information samples is determined based on the abnormal information of each of the above business information samples, including: scoring each of the above business information samples based on a preset experience library and each of the abnormal information of each of the above business information samples to obtain each sample score corresponding to each of the above business information samples; determining each of the above initial business information negative samples based on the above sample scores and adding negative sample labels to obtain each initial business information negative sample carrying a negative sample label.

[0067] In this embodiment, the server can also determine the score of the sample based on experience, and then determine the initial business information negative sample. For example, the server can score each of the above business information samples according to the preset experience library and each abnormal information of each of the above business information samples, and obtain each sample score corresponding to each of the above business information samples. Among them, the sample score represents the score determined according to the preset experience library combined with the abnormality degree of the abnormal information of the business information sample. The higher the score, the greater the probability of belonging to the abnormal business information sample. Then the server can determine each of the above initial business information negative samples according to the above sample score, and add a negative sample label to obtain each initial business information negative sample carrying a negative sample label.

[0068] Specifically, the server can build expert experience based on the user's basic account information and transaction information, also known as a preset experience library, so that the server can stratify and refine user risks and transaction risks, and quantitatively calculate risk scores. The higher the score, the higher the corresponding risk, and obtain the scores of each sample corresponding to each business information sample. The server can select each business information sample with a score less than the second preset score threshold as the initial business information negative sample, and add the corresponding negative sample label to obtain each initial business information negative sample. Among them, the second preset score threshold can be set according to actual conditions. The server can combine each initial business information negative sample with each initial business information positive sample to form a labeled data set.

[0069] Through the above embodiments, the server uses multiple methods to determine the initial labeled samples, and uses the initial samples to perform the first labeled training on the abnormal business identification model, and then performs the second training on the first abnormal business identification model after the first training based on the unlabeled business information samples, thereby improving the training efficiency of the abnormal business identification model. In addition, the server uses the experience of business domain experts to select the initial negative samples, which also reduces the complexity of the model.

[0070] In one embodiment, the business information sample is input into the first abnormal business identification model, and the first abnormal business identification model outputs a corresponding business type identification result based on the business information sample, including: inputting the business information sample into the first abnormal business identification model, and the first abnormal business identification model identifies the business type corresponding to the business information sample and the confidence level corresponding to the business type; determining a corresponding prediction label according to the business type; and outputting a business type identification result corresponding to the business information sample according to the prediction label and the confidence level.

[0071] In this embodiment, the above-mentioned business type identification result includes the business type of the business information sample identified by the first abnormal business identification model, the corresponding label and the confidence of the business type. That is, the server uses the first abnormal business identification model to identify the business type of the unlabeled business information sample. For example, the server inputs the business information sample into the above-mentioned first abnormal business identification model, and the above-mentioned first abnormal business identification model identifies the business type corresponding to the above-mentioned business information sample, and identifies the confidence corresponding to the above-mentioned business type. Among them, the confidence indicates the degree of determination of the first abnormal business identification model on the above-mentioned identified business type. The higher the confidence, the higher the degree of determination of the first abnormal business identification model that the business information sample belongs to this business type.

[0072] The first abnormal business identification model can determine the corresponding prediction label according to the above business type, for example, it can include negative sample labels and positive sample labels. Therefore, the first abnormal business identification model outputs the business type identification result corresponding to the above business information sample according to the above prediction label and the above confidence level. In this way, the first abnormal business identification model is used to predict the unlabeled business information sample.

[0073] Specifically, the server can use the first abnormal business identification model trained with the initial data set to predict the unlabeled business information samples. For example, the server can use the first abnormal business identification model trained with the initial data set to predict the unlabeled business information samples and obtain the business type identification result corresponding to each unlabeled business information sample. Among them, the above-mentioned business type identification result can be a classification result, and the business type identification result can obtain the predicted business type, the confidence corresponding to the business type, and the label corresponding to the business type. Among them, the confidence can be a probability value based on the output of softmax (normalized exponential function).

[0074] Through this embodiment, the server can use unlabeled business information samples to predict the first abnormal business identification model obtained through supervised training of the initial labeled data set, so that the server can use the predicted business type identification results to perform semi-supervised training on the first abnormal business identification model, thereby improving the efficiency of training the abnormal business identification model.

[0075] In one embodiment, the first abnormal business identification model is subjected to a second training based on the business type identification result and the data set until a preset training end condition is met, thereby obtaining a trained abnormal business identification model, including: determining the target business information sample whose confidence is greater than a preset threshold value based on the business type identification result; expanding the data set based on the target business information sample carrying the prediction label to obtain a new data set; inputting the new data set into the first abnormal business identification model, and the first abnormal business identification model outputs a new business type identification result based on the new data set; adjusting the model parameters of the first abnormal business identification model based on the new business type identification result, and returning to the step of determining the target business information sample whose confidence is greater than the preset threshold value based on the business type identification result until the preset training end condition is met, thereby obtaining a trained abnormal business identification model.

[0076] In this embodiment, the server can use the business type recognition result output by the first abnormal business recognition model to expand the data set, and use the expanded data set to implement semi-supervised training of the first abnormal business recognition model. Among them, the business type recognition result output by the first abnormal business recognition model based on the recognition of unlabeled business information samples may include the business type of the business information sample, the corresponding prediction label and the confidence of the business type. The server can then determine the target business information sample whose confidence is greater than the preset threshold based on the above business type recognition result, that is, compare the confidence in each business type recognition result with the preset threshold. If it is greater than the preset threshold, the business information sample corresponding to the confidence is determined to be the target business information sample.

[0077] The server can add the target business information samples carrying the above-mentioned prediction labels to the above-mentioned data set, thereby obtaining a new expanded data set. The server can input the new data set into the above-mentioned first abnormal business identification model, and the above-mentioned first abnormal business identification model outputs a new business type identification result based on the new data set. The server adjusts the model parameters of the above-mentioned first abnormal business identification model according to the new business type identification result, and returns to the step of determining the target business information samples with confidence greater than the preset threshold according to the above-mentioned business type identification result, that is, determining the target business information samples with higher confidence from the new business type identification result, and expanding the above-mentioned data set together with the prediction labels carried by them. The server further trains the first abnormal business identification model with the expanded data set until the above-mentioned preset training end condition is met, and the server obtains the trained abnormal business identification model. Among them, the preset training end condition includes but is not limited to the convergence of the business type identification result output by the above-mentioned first abnormal business identification model within the preset training times, or the training times reach the preset training times.

[0078] Specifically, the above-mentioned prediction label is also called a pseudo-label. For the business type identification results output by the first abnormal business identification model, the server can select the identification results with higher confidence as pseudo-labels. For example, the server sets a preset threshold of confidence and only selects data of target business information samples whose identification probability of the first abnormal business identification model is higher than the preset threshold of confidence, such as target business information samples with confidence higher than 0.9. In this way, the server assigns corresponding pseudo-labels to the target business information samples that the above-mentioned first abnormal business identification model is more confident in, thereby avoiding noise caused by low-confidence predictions.

[0079] Thus, the server can merge the selected unlabeled target business information samples and their corresponding pseudo-labels with the original labeled data set to form an expanded new data set, thereby increasing the size of the training data set. The server uses the expanded data set including labeled business information samples and target business information samples with pseudo-labels to retrain the first abnormal business identification model, so that the first abnormal business identification model will consider more data in the new training and improve the generalization ability of unlabeled business information samples. The server can iterate and repeat the above process multiple times to continuously update the model and generate new pseudo-labels. After each round of iteration outputs the business type identification result, the first abnormal business identification model can obtain more target business information samples with pseudo-labels, so that the first abnormal business identification model can continuously improve its accuracy during the training process.

[0080] The server can also set a termination condition. For example, if the model performance of the first abnormal business identification model is no longer significantly improved on the validation set, or reaches a preset maximum number of iterations, the server can terminate the training and output the final trained abnormal business identification model. Thus, the server can use the trained abnormal business identification model to identify abnormal business in financial business.

[0081] Through this embodiment, the server can use the business type identification results output by the first abnormal business identification model to select target business information samples with higher confidence to expand the data set, and use the expanded data set to iteratively train the first abnormal business identification model, thereby achieving the effect of improving the efficiency of training the abnormal business identification model.

[0082] In an exemplary embodiment, as shown in FIG2 , FIG2 is a schematic diagram of a training method for an abnormal business identification model in an embodiment. In this embodiment, the server can pre-train to obtain an initial model, that is, the first abnormal business identification model mentioned above. The server can first obtain reliable negative samples, for example, by a two-step iterative method or an empirical method.

[0083] Among them, for the two-step iterative method, the server can regard all unlabeled business information samples as negative samples, that is, each candidate business information negative sample, so that the server builds a preliminary classifier based on each candidate business information negative sample. The server uses this classifier to score and classify the unlabeled business information samples, thereby obtaining the corresponding classification results, including the business type corresponding to the business information sample, and the score of the business information sample. The server can select the business information sample with a score less than the first preset score threshold as the initial business information negative sample.

[0084] For the empirical method, the server can build expert experience based on the user's basic account information and transaction information, also known as a preset experience library, so that the server can stratify and refine user risks and transaction risks, and quantitatively calculate risk scores. The higher the score, the higher the corresponding risk, and obtain the sample scores corresponding to each business information sample. The server can select each business information sample with a score less than the second preset score threshold as the initial business information negative sample, and add the corresponding negative sample label to obtain each initial business information negative sample.

[0085] The server can use the data set containing the initial business information negative sample and the initial business information positive sample to perform labeled supervised training on the abnormal business identification model to be trained to obtain an initial first abnormal business identification model. The model can be any learnable algorithm, such as a logistic regression algorithm, a random forest algorithm, and a neural network.

[0086] The server can use the first abnormal business identification model trained with the initial data set to predict the unlabeled business information samples. For example, the server can use the first abnormal business identification model trained with the initial data set to predict the unlabeled business information samples and obtain the business type identification result corresponding to each unlabeled business information sample. The above-mentioned business type identification result can be a classification result, and the business type identification result can obtain the predicted business type, the confidence corresponding to the business type, and the label corresponding to the business type. The confidence can be a probability value based on the output of softmax (normalized exponential function).

[0087] The server can select the recognition results with higher confidence as pseudo labels. For example, the server sets a preset threshold of confidence and only selects the data of the target business information samples whose recognition probability of the first abnormal business recognition model is higher than the preset threshold of confidence, such as the target business information samples with confidence higher than 0.9, so that the server assigns corresponding pseudo labels to the target business information samples that the first abnormal business recognition model is more confident of, avoiding the noise caused by low confidence prediction. In addition, by determining pseudo labels, a large amount of unlabeled data can be utilized, while maintaining the accuracy of the model and reducing the need for a large amount of labeled data.

[0088] Thus, the server can merge the selected unlabeled target business information samples and their corresponding pseudo-labels with the original labeled data set to form an expanded new data set, thereby increasing the size of the training data set. The server uses the expanded data set including labeled business information samples and target business information samples with pseudo-labels to retrain the first abnormal business identification model, so that the first abnormal business identification model will consider more data in the new training and improve the generalization ability of unlabeled business information samples. The server can iterate and repeat the above process multiple times, so as to continuously update the model and generate new pseudo-labels. After each round of iteration outputs the business type identification result, the first abnormal business identification model can obtain more target business information samples with pseudo-labels, so that the first abnormal business identification model can continuously improve its accuracy during the training process. By iteratively training the first abnormal business identification model, the server can gradually enhance the capabilities of the model and make full use of pseudo-labels to improve performance.

[0089] The server can also set a termination condition. For example, if the model performance of the first abnormal business identification model is no longer significantly improved on the validation set, or reaches a preset maximum number of iterations, the server can terminate the training and output the final trained abnormal business identification model. Thus, the server can use the trained abnormal business identification model to identify abnormal business in financial business.

[0090] Through the above embodiments, the server performs multiple iterative training on the abnormal business identification model by combining labeled initial samples and unlabeled samples, thereby reducing the demand for labeled samples and improving the training efficiency of the abnormal business identification model. In addition, the server selects initial negative samples using the experience of domain experts, effectively utilizes a large amount of unlabeled data through iteration, reduces the demand for a large amount of labeled data while maintaining the accuracy of the model, gradually enhances the model's capabilities, and makes full use of pseudo labels to improve performance.

[0091] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0092] Based on the same inventive concept, the embodiment of the present application also provides a training device for an abnormal business identification model for implementing the training method of the abnormal business identification model involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of the training device for one or more abnormal business identification models provided below can refer to the limitations of the training method for the abnormal business identification model above, and will not be repeated here.

[0093] In an exemplary embodiment, as shown in FIG3 , a training device for an abnormal business identification model is provided, comprising: an acquisition module 500, a first training module 502, an input module 504, and a second training module 506, wherein:

[0094] The acquisition module 500 is used to acquire the abnormal business identification model to be trained and the corresponding data set; the above data set includes initial business information negative samples carrying negative sample labels and initial business information positive samples carrying positive sample labels.

[0095] The first training module 502 is used to perform a first training on the abnormal business identification model based on the initial business information negative sample and the initial business information positive sample in the data set to obtain a first abnormal business identification model.

[0096] The input module 504 is used to obtain unlabeled business information samples, input the business information samples into the first abnormal business identification model, and the first abnormal business identification model outputs corresponding business type identification results based on the business information samples.

[0097] The second training module 506 is used to perform a second training on the first abnormal business identification model according to the business type identification result and the data set until a preset training end condition is met, thereby obtaining a trained abnormal business identification model.

[0098] In one embodiment, the acquisition module 500 is used to acquire each unlabeled business information sample and each initial business information positive sample carrying a positive sample label; based on the abnormal information of each of the above business information samples, each initial business information negative sample in the above business information samples is determined and a negative sample label is added to obtain each initial business information negative sample carrying a negative sample label; based on each of the above initial business information negative samples and each of the above initial business information positive samples, the data set is generated.

[0099] In one embodiment, the acquisition module 500 is used to mark each of the business information samples as a candidate business information negative sample, and to construct a classifier based on each of the candidate business information negative samples; the classifier is used to classify each of the business information samples based on abnormal information, and to determine each of the initial business information negative samples based on the classification results and add negative sample labels to obtain each initial business information negative sample carrying a negative sample label.

[0100] In one embodiment, the acquisition module 500 is used to score each of the business information samples according to a preset experience database and each abnormal information of each of the business information samples to obtain each sample score corresponding to each of the business information samples; and determine each of the initial business information negative samples according to the sample score.

[0101] In one embodiment, the input module 504 is used to input the business information sample into the first abnormal business identification model, and the first abnormal business identification model identifies the business type corresponding to the business information sample and the confidence level corresponding to the business type; determines the corresponding prediction label according to the business type; and outputs the business type identification result corresponding to the business information sample according to the prediction label and the confidence level.

[0102] In one embodiment, the second training module 506 is used to determine the target business information samples whose confidence level is greater than a preset threshold value based on the business type identification results; expand the data set based on the target business information samples carrying the prediction labels to obtain a new data set; input the new data set into the first abnormal business identification model, and the first abnormal business identification model outputs a new business type identification result based on the new data set; adjust the model parameters of the first abnormal business identification model based on the new business type identification result, and return to the step of determining the target business information samples whose confidence level is greater than the preset threshold value based on the business type identification results, until the preset training end condition is met to obtain a trained abnormal business identification model.

[0103] Each module in the training device of the abnormal business identification model can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0104] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG4. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store business information sample data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a training method for an abnormal business identification model is implemented.

[0105] Those skilled in the art will understand that the structure shown in FIG. 4 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0106] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned training method for the abnormal business identification model when executing the computer program.

[0107] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned training method of the abnormal business identification model is implemented.

[0108] In one embodiment, a computer program product is provided, including a computer program, which implements the above-mentioned training method for abnormal business identification model when executed by a processor.

[0109] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0110] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0111] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0112] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A training method for an abnormal business identification model, characterized in that: The method comprises: Acquire an abnormal business identification model to be trained and a corresponding data set; the data set includes initial business information negative samples carrying negative sample labels and initial business information positive samples carrying positive sample labels; Based on the initial business information negative sample and the initial business information positive sample in the data set, performing a first training on the abnormal business identification model to obtain a first abnormal business identification model; Acquire an unlabeled service information sample, input the service information sample into the first abnormal service identification model, and have the first abnormal service identification model output a corresponding service type identification result based on the service information sample; According to the service type identification result and the data set, the first abnormal service identification model is subjected to second training until a preset training end condition is met, thereby obtaining a trained abnormal service identification model.

2. The method according to claim 1, characterized in that The steps of obtaining a data set corresponding to abnormal business identification to be trained include: Obtain each unlabeled business information sample and each initial business information positive sample carrying a positive sample label; According to the abnormal information of each of the business information samples, each initial business information negative sample in each of the business information samples is determined and a negative sample label is added to obtain each initial business information negative sample carrying a negative sample label; The data set is generated according to each of the initial business information negative samples and each of the initial business information positive samples.

3. The method according to claim 2, characterized in that The determining, according to the abnormal information of each of the business information samples, each of the initial business information negative samples in each of the business information samples comprises: Marking each of the business information samples as a candidate business information negative sample, and constructing a classifier according to each of the candidate business information negative samples; The classifier classifies each of the business information samples according to the abnormal information, determines each of the initial business information negative samples according to the classification result and adds a negative sample label to obtain each of the initial business information negative samples carrying the negative sample label.

4. The method according to claim 2, characterized in that: The determining, according to the abnormal information of each of the business information samples, each of the initial business information negative samples in each of the business information samples comprises: Scoring each of the business information samples according to a preset experience database and each of the abnormal information of each of the business information samples to obtain each sample score corresponding to each of the business information samples; Each of the initial business information negative samples is determined according to the sample score.

5. The method according to any one of claims 1 to 4, characterized in that: The step of inputting the service information sample into the first abnormal service identification model, and having the first abnormal service identification model output a corresponding service type identification result based on the service information sample, includes: Inputting the service information sample into the first abnormal service identification model, and allowing the first abnormal service identification model to identify the service type corresponding to the service information sample and the confidence level corresponding to the service type; Determine a corresponding prediction label according to the business type; According to the prediction label and the confidence level, a business type recognition result corresponding to the business information sample is output.

6. The method according to claim 5, characterized in that The second training of the first abnormal business identification model is performed according to the business type identification result and the data set until a preset training end condition is met to obtain a trained abnormal business identification model, including: Determining, according to the business type identification result, a target business information sample whose confidence level is greater than a preset threshold; Expanding the data set according to the target business information sample carrying the prediction label to obtain a new data set; Inputting the new data set into the first abnormal business identification model, and having the first abnormal business identification model output a new business type identification result based on the new data set; Adjust the model parameters of the first abnormal business identification model according to the new business type identification result, and return to the step of determining the target business information sample whose confidence is greater than a preset threshold according to the business type identification result, until the preset training end condition is met, to obtain a trained abnormal business identification model.

7. A training device for an abnormal business identification model, characterized in that: The device comprises: An acquisition module is used to acquire an abnormal business identification model to be trained and a corresponding data set; the data set includes initial business information negative samples carrying negative sample labels and initial business information positive samples carrying positive sample labels; A first training module, configured to perform a first training on the abnormal business identification model based on the initial business information negative sample and the initial business information positive sample in the data set to obtain a first abnormal business identification model; An input module, used to obtain an unlabeled business information sample, input the business information sample into the first abnormal business identification model, and the first abnormal business identification model outputs a corresponding business type identification result based on the business information sample; The second training module is used to perform a second training on the first abnormal business identification model according to the business type identification result and the data set until a preset training end condition is met, thereby obtaining a trained abnormal business identification model.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images