A model training and business risk control method and device

Abstract

The specification discloses a model training and business risk control method and device, which can obtain complaint text information and business data of a user in history, and a risk control rule, input the complaint text information into a first encoding layer in a business model, input the business data into a second encoding layer in the business model, perform first training on the business model based on results output by the first encoding layer and the second encoding layer, input the complaint text information into the first encoding layer after training, input the risk control rule into a third encoding layer, perform second training on the business model based on results output by the first encoding layer after training and the third encoding layer, and finally input the complaint text information, the business data and the risk control rule into the business model to perform overall training on the business model, thereby improving the efficiency and accuracy of business risk control for the user and protecting the privacy data of the user.

Description

TECHNICAL FIELD

[0001] The present specification relates to the technical field of computer technology, and particularly relates to a model training and business risk control method and device. BACKGROUND

[0002] Currently, in order to protect the property safety, privacy data safety and the like of users, a business platform can receive complaint information of a user, and based on the complaint information of the user, solve actual business problems for the user, thereby providing better services for the user.

[0003] In actual application, after receiving the complaint information of the user, the business platform can determine whether the complaint information of the user is true through manual auditing, and in the case of being true, perform business processing for the user according to the complaint information of the user, but this way needs to consume a large amount of manual cost and has low efficiency.

[0004] Therefore, how to improve the efficiency of complaint auditing is a problem to be solved urgently. SUMMARY

[0005] The present specification provides a model training and business risk control method and device to improve the efficiency of helping users to perform business risk control.

[0006] The present specification adopts the following technical solution:

[0007] The present specification provides a model training method, comprising:

[0008] Obtaining complaint text information and business data of a user in history when performing a business, and a risk control rule corresponding to the business;

[0009] Inputting the complaint text information into a first encoding layer in a business model, and inputting the business data into a second encoding layer in the business model, and training the first encoding layer and the second encoding layer based on results output by the first encoding layer and the second encoding layer;

[0010] Inputting the complaint text information into the trained first encoding layer in the business model, and inputting the risk control rule into a third encoding layer in the business model, and training the trained first encoding layer and the third encoding layer based on results output by the trained first encoding layer and the third encoding layer;

[0011] inputting the complaint text information into a first encoding layer of the trained business model, inputting the business data into a second encoding layer of the trained business model, and inputting the risk control rule into a third encoding layer of the trained business model, and training the business model based on results output by the first encoding layer, the second encoding layer, and the third encoding layer.

[0012] Optionally, inputting the complaint text information into a first encoding layer of the business model and inputting the business data into a second encoding layer of the business model, and performing first training on the business model based on results output by the first encoding layer and the second encoding layer, includes:

[0013] inputting the complaint text information and the business data into the business model to output feature encoding corresponding to the complaint text information through a first encoding layer of the business model and to determine feature encoding corresponding to the business data through a second encoding layer of the business model;

[0014] determining a prediction result based on the complaint text information and the business data according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the business data, and performing first training on the business model with an optimization objective of minimizing the prediction result and a business label corresponding to the user.

[0015] Optionally, the risk control rule includes a first rule, a second rule, and a third rule, the first rule is used to indicate that the user has consumed business resources for the business corresponding to the complaint text information, the second rule is used to indicate that other users involved in the complaint text information have not returned the business resources consumed by the user, and the third rule is used to indicate that the user and the other users involved in the complaint text information are in a preset relationship.

[0016] inputting the complaint text information into a first encoding layer of the trained business model and inputting the risk control rule into a third encoding layer of the business model, and performing second training on the business model based on results output by the first encoding layer and the third encoding layer, includes:

[0017] inputting the complaint text information into a first encoding layer of the trained business model and inputting the risk control rule into a third encoding layer of the business model, and obtaining feature encoding corresponding to the complaint text information and feature encoding corresponding to the risk control rule;

[0018] According to the feature code corresponding to the complaint text information and the feature code corresponding to the risk control rule, determine the judgment result corresponding to the first rule, the judgment result corresponding to the second rule and the judgment result corresponding to the third rule;

[0019] With minimizing the difference between the judgment result corresponding to the first rule and the first label corresponding to the user, the difference between the judgment result corresponding to the second rule and the second label corresponding to the user and the difference between the judgment result corresponding to the third rule and the third label corresponding to the user as the optimization target, the business model is second trained.

[0020] Optionally, based on the results output by the trained first encoding layer, the trained second encoding layer and the trained third encoding layer, the business model is trained, including:

[0021] According to the feature code output by the trained first encoding layer, the trained second encoding layer and the trained third encoding layer respectively, the feature code corresponding to the complaint text information, the feature code corresponding to the business data and the feature code corresponding to the risk control rule, the prediction result of the business executed by the user in the past is obtained;

[0022] With minimizing the difference between the prediction result and the business label corresponding to the user as the optimization target, the business model is trained.

[0023] The present specification provides a business risk control method, including:

[0024] Obtain the complaint text information input by the user when executing the risk control business, and the business data corresponding to the complaint text information;

[0025] Input the complaint text information, the business data and the risk control rule into the pre-trained business model, to determine the judgment result of whether the user executing the risk control business conforms to the risk control rule through the results output by the first encoding layer, the second encoding layer and the third encoding layer in the business model, and the business model is trained by the above model training method;

[0026] According to the judgment result, the user executes the business risk control.

[0027] The present specification provides a model training device, including:

[0028] The acquisition module is used for acquiring the complaint text information and the business data corresponding to the business executed by the user in the past, and the risk control rule corresponding to the business;

[0029] The first pre-training module is configured to input the complaint text information into a first encoding layer in the business model, input the business data into a second encoding layer in the business model, and perform first training on the business model based on results output by the first encoding layer and the second encoding layer, where the first training process at least includes adjusting model parameters of the first encoding layer and the second encoding layer.

[0030] The second pre-training module is configured to input the complaint text information into the trained first encoding layer in the business model, input the risk control rule into a third encoding layer in the business model, and perform second training on the business model based on results output by the trained first encoding layer and the third encoding layer, where the second training process at least includes adjusting model parameters of the trained first encoding layer and the third encoding layer.

[0031] The training module is configured to input the complaint text information into the trained first encoding layer in the business model, input the business data into the trained second encoding layer in the business model, and input the risk control rule into the trained third encoding layer in the business model, and perform training on the business model based on results output by the trained first encoding layer, the trained second encoding layer, and the trained third encoding layer.

[0032] The present specification provides a device for business risk control, comprising:

[0033] The acquisition module is configured to acquire complaint text information input by a user when performing a risk control business, and business data corresponding to the complaint text information.

[0034] The input module is configured to input the complaint text information, the business data, and a risk control rule into a pre-trained business model, to determine, by the business model, a determination result of whether the user performing the risk control business meets the risk control rule based on results output by a first encoding layer, a second encoding layer, and a third encoding layer in the business model, and the business model is trained by the above-mentioned model training method.

[0035] The risk control module is configured to perform business risk control on the user according to the determination result.

[0036] The present specification provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned model training or business risk control method.

[0037] The specification provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the above-mentioned model training or business risk control method when executing the program.

[0038] The above-mentioned at least one technical solution adopted by the specification can achieve the following beneficial effects:

[0039] In the model training and business risk control method provided in the specification, complaint text information and business data corresponding to a business performed by a user in the past can be obtained, as well as a risk control rule corresponding to the business, the complaint text information is input into a first encoding layer in a business model, and the business data is input into a second encoding layer in the business model, and based on the results output by the first encoding layer and the second encoding layer, the business model is first trained, and the first training process at least includes adjusting the model parameters of the first encoding layer and the second encoding layer. Then, the complaint text information can be input into the first encoding layer of the business model after training, and the risk control rule can be input into the third encoding layer of the business model, and based on the results output by the first encoding layer after training and the third encoding layer, the business model is second trained, and the second training process at least includes adjusting the model parameters of the first encoding layer after training and the third encoding layer. Finally, the complaint text information can be input into the first encoding layer of the business model after training, the business data can be input into the second encoding layer of the business model after training, and the risk control rule can be input into the third encoding layer of the business model after training, and based on the results output by the first encoding layer after training, the second encoding layer after training and the third encoding layer after training, the business model is trained.

[0040] As can be seen from the above method, the complaint text information is used as a bridge between the business data and the risk control rule, that is, after the first encoding layer and the second encoding layer are trained together, the first encoding layer and the third encoding layer are trained together, and finally, the first encoding layer, the second encoding layer and the third encoding layer are trained together, so as to obtain the business model after training. The business model can be used to audit the business described by the complaint text information of the user, thereby improving the efficiency and accuracy of the business risk control for the user. BRIEF DESCRIPTION OF DRAWINGS

[0041] The drawings described herein are used to provide further understanding of the specification, and form a part of the specification. The illustrative embodiments of the specification and their descriptions are used to explain the specification, and do not constitute an improper limitation on the specification. In the drawings:

[0042] Figure 1 A flowchart of a model training method in the specification;

[0043] Figure 2 A flowchart of a process for training the business model provided in the present specification is provided.

[0044] Figure 3 A flowchart of a process for the business risk control method provided in the present specification is provided.

[0045] Figure 4 A device diagram of a model training provided in the present specification is provided.

[0046] Figure 5 A device diagram of a business risk control provided in the present specification is provided.

[0047] Figure 6 A device diagram of an electronic device corresponding to Figure 1 or Figure 3 provided in the present specification is provided. DETAILED DESCRIPTION

[0048] In order to make the purposes, technical solutions and advantages of the present specification clearer, the technical solutions of the present specification will be described in detail below with reference to the embodiments of the present specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present specification, not all the embodiments. Based on the embodiments in the present specification, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of protection of the present specification.

[0049] The technical solutions provided by the embodiments of the present specification will be described in detail below with reference to the drawings.

[0050] Figure 1 A flowchart of a process for the model training method provided in the present specification, specifically comprising the following steps:

[0051] S100: Obtain complaint text information and business data when a user performs a business in history, and a risk control rule corresponding to the business.

[0052] In actual application, in the business risk control, if there is a problem in the process of performing a business by a user, the user can submit a text describing the problem to a business platform as complaint text information, and the business platform can perform auditing based on the complaint text information, and if the auditing is passed, the business risk control can be performed for the user. The present specification mainly proposes a training method for a business model for auditing such complaint text information, and a business risk control method of applying the business model to audit the complaint text information.

[0053] Specifically, in the process of training the business model, complaint text information and business data corresponding to a business executed by a user in the past, and a risk control rule corresponding to the business can be obtained. The complaint text information mentioned here can refer to text describing a problem of the business executed by the user in the past, and the user can apply for business risk control to the business platform through the complaint text information. The business data can refer to data related to the business executed by the user in the past, and the risk control rule can refer to a rule for auditing business risk control applied by the user through the complaint text information in the risk control business.

[0054] The above three concepts will be described below taking the scenario that a user applies for business risk control to a business platform in a transaction scenario as an example. It is assumed that there are some problems when the user transacts with others (such as being cheated by others to transfer money), and then the user can submit complaint text information to the business platform. The complaint text information can describe the problems involved in the transaction of the user, and the business platform can audit the complaint text information submitted by the user based on the complaint text information, obtain business data related to the user (such as the amount of money transferred by the user to others, some information of the user himself, some information of the other party himself, and historical transaction information of the user and the other party, etc.), and a risk control rule.

[0055] S102: input the complaint text information into a first encoding layer in the business model, and input the business data into a second encoding layer in the business model, and perform first training on the business model based on results output by the first encoding layer and the second encoding layer, wherein the first training process at least includes adjusting model parameters of the first encoding layer and the second encoding layer.

[0056] S104: input the complaint text information into the first encoding layer of the business model after training, and input the risk control rule into a third encoding layer in the business model, and perform second training on the business model based on results output by the first encoding layer after training and the third encoding layer, wherein the second training process at least includes adjusting model parameters of the first encoding layer after training and the third encoding layer.

[0057] When training the business model, the first encoding layer for determining feature encoding of the complaint text information, the second encoding layer for determining feature encoding of the business data, and the third encoding layer for determining feature encoding of the risk control rule can be pre-trained respectively.

[0058] Specifically, the complaint text information can be input into a first encoding layer in the business model, and the business data can be input into a second encoding layer in the business model, and based on the results output by the first encoding layer and the second encoding layer, the business model is first trained, and the first training process at least includes adjusting the model parameters of the first encoding layer and the second encoding layer. That is, in the first training process, the first encoding layer and the second encoding layer are mainly trained.

[0059] And the complaint text information is input into the first encoding layer of the trained business model, the risk control rules are input into the third encoding layer in the business model, and based on the results output by the trained first encoding layer and the third encoding layer, the business model is second trained, and the second training process at least includes adjusting the model parameters of the trained first encoding layer and the third encoding layer.

[0060] That is, in the pre-training stage, the first encoding layer and the second encoding layer can be trained through the complaint text information and the business data, so that the first encoding layer and the second encoding layer can learn the relationship between the complaint text information and the business data, and the first encoding layer and the third encoding layer can be trained through the complaint text information and the risk control rules, so that the first encoding layer and the third encoding layer can learn the relationship between the complaint text information and the risk control rules.

[0061] That is, the pre-training process is to try to establish the relationship between the risk control rules and the business data through the complaint text information as much as possible, so the first encoding layer corresponding to the complaint text information is shared in the first training and the second training. It can be understood that the first encoding layer is an indirect medium for establishing the relationship between the risk control rules and the business data, and the three encoding layers will be trained together in the subsequent process.

[0062] When pre-training the first encoding layer and the second encoding layer, the complaint text information and the business data can be input into the business model, so as to output the feature encoding corresponding to the complaint text information through the first encoding layer in the business model, and determine the feature encoding corresponding to the business data through the second encoding layer in the business model, and according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the business data, determine the prediction result based on the complaint text information and the business data, and minimize the prediction result and the business label corresponding to the user as the optimization target. The business model is first trained, and it can also be understood that this process is to pre-train the first encoding layer and the second encoding layer in the business model.

[0063] The prediction result can be used to represent whether the predicted risk control business (or complaint event) corresponding to the complaint text information passes the audit, or whether the historical business performed by the user involved in the complaint text information complies with the risk control rule. The labeling result can represent whether the risk control business (or complaint event) corresponding to the complaint text information actually passes the audit, or whether the historical business performed by the user involved in the complaint text information actually complies with the risk control rule. In the risk control scenario in which the user complains about his own transaction, the prediction result can be used to represent whether the complaint of the user predicted by the business model passes the audit, and the labeling information can be used to represent whether the complaint of the user actually passes the audit.

[0064] Therefore, it can be easily thought that in the first training process, the positive sample can be obtained from the information related to the historical risk control business that passes the audit, and the negative sample can be obtained from the information related to the historical risk control business that does not pass the audit. Of course, the negative sample can also be determined in other ways, that is, it can be determined from the complaint text information of the user and the business data of other users unrelated to the user.

[0065] Of course, the training samples of the subsequent training processes can be generated in the above manner, and the three training processes can share the training samples. That is, the labeling information of the training samples is needed in the three training processes, and the first training process can use the complaint text information and the business data in the training samples, the second training process can use the complaint text information and the risk control rule in the training samples, and the last training process can use the complaint text information, the business data and the risk control rule in the training samples.

[0066] When the first encoding layer and the third encoding layer are continuously trained, the training can be continuously performed in a manner similar to the above, that is, the complaint text information and the risk control rule can be input into the business model to output the feature encoding corresponding to the complaint text information through the first encoding layer in the business model, and to determine the feature encoding corresponding to the risk control rule through the third encoding layer in the business model, and to determine the prediction result based on the complaint text information and the risk control rule according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the risk control rule, and to perform the second training on the business model with the optimization target of minimizing the prediction result and the business label corresponding to the user. This training process is equivalent to training the first encoding layer and the third encoding layer together.

[0067] Wherein, since the risk control rules can contain multiple rules, the labeling information can not only be a business label indicating whether the review is passed, but also can contain labels corresponding to each risk control rule (i.e., whether the user's historical business execution conforms to the corresponding risk control rule).

[0068] Specifically, the risk control rules can include a first rule, a second rule, and a third rule. The first rule can be used to indicate that the user has consumed business resources for the business corresponding to the complaint text information. The second rule can be used to indicate that the other user involved in the complaint text information has not returned the business resources consumed by the user. The third rule can be used to indicate that the user and the other user involved in the complaint text information are in a preset relationship. Then, the label corresponding to one risk control rule indicates whether the user's historical business execution conforms to the risk control rule.

[0069] Here, the first rule, the second rule, and the third rule correspond to actual business scenarios, which can mean that the first rule is that the user has spent money on the business involved in the complaint text information, the second rule is that the user has not been repaid for the money spent on the other user in the business involved in the complaint text information, and the third rule is that the user and the other user are strangers.

[0070] Specifically, the training process can be: inputting the complaint text information into the first encoding layer of the business model after training, and inputting the risk control rules into the third encoding layer of the business model, to obtain the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the risk control rules. Then, according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the risk control rules, determine the determination result corresponding to the first rule (i.e., whether the complaint event corresponding to the complaint text information conforms to the first rule), the determination result corresponding to the second rule (i.e., whether the complaint event corresponding to the complaint text information conforms to the second rule), and the determination result corresponding to the third rule (i.e., whether the complaint event corresponding to the complaint text information conforms to the third rule). Finally, taking the minimization of the difference between the determination result corresponding to the first rule and the first label corresponding to the user, the difference between the determination result corresponding to the second rule and the second label corresponding to the user, and the difference between the determination result corresponding to the third rule and the third label corresponding to the user as the optimization goal, training the first encoding layer and the third encoding layer after training.

[0071] That is, when training the first encoding layer and the third encoding layer again, the labeling information can contain three labels corresponding to the first rule, the second rule, and the third rule, respectively. Therefore, when training the first encoding layer and the third encoding layer, there is a loss function corresponding to each label, and the first encoding layer and the third encoding layer are trained by minimizing the difference between each label and its corresponding prediction result.

[0072] The label corresponding to the first rule can indicate whether the complaint event corresponding to the complaint text information actually complies with the first rule. The label corresponding to the second rule can indicate whether the complaint event corresponding to the complaint text information actually complies with the second rule. The label corresponding to the third rule can indicate whether the complaint event corresponding to the complaint text information actually complies with the third rule.

[0073] S106: Input the complaint text information into the trained first coding layer in the business model, input the business data into the trained second coding layer, and input the risk control rules into the trained third coding layer in the business model, and train the business model based on the output results of the trained first coding layer, the trained second coding layer, and the trained third coding layer.

[0074] After pre-training the first coding layer, the second coding layer and the third coding layer in the above manner, the complaint text information can be input into the trained first coding layer in the business model, the business data can be input into the trained second coding layer, and the risk control rules can be input into the trained third coding layer in the business model, and the business model can be trained based on the output results of the trained first coding layer, the trained second coding layer and the trained third coding layer.

[0075] Similarly, there are many ways to train the business model as a whole. For example, based on the feature codes corresponding to the complaint text information, the feature codes corresponding to the business data, and the feature codes corresponding to the risk control rules output by the trained first coding layer, the trained second coding layer, and the trained third coding layer respectively, the prediction results for the business performed historically by the user are obtained; and the business model is trained with the optimization goal of minimizing the difference between the prediction result and the business label corresponding to the user.

[0076] For another example, similar to the second pre-training process mentioned above, in the annotation information, each risk control rule corresponds to the label of the risk control rule. Then, when training the business model, it is necessary to make the business model output the prediction results corresponding to each risk control rule, and minimize the difference between the prediction results corresponding to each risk control rule and the label corresponding to the corresponding risk control rule as the optimization goal to train the business model as a whole.

[0077] From the above content, it can be seen that the process of training the business model in this specification is to pre-train the first coding layer, the second coding layer and the third coding layer, and then uniformly train the business model as a whole, such as Figure 2 shown.

[0078] Figure 2 This is a flow chart of a training process for the above business model provided in this specification.

[0079] As can be seen from Figure 2 In the pre-training process, the complaint text information serves to build a connection between the risk control rules and the business data. Therefore, in the two pre-training processes, the first encoding layer corresponding to the complaint text information is involved in the pre-training. First, the first encoding layer and the second encoding layer are trained together. Then, the first encoding layer and the third encoding layer are trained together. In this way, after the first encoding layer, the second encoding layer, and the third encoding layer are pre-trained, the business model can be trained as a whole. That is, the first encoding layer, the second encoding layer, and the third encoding layer are trained together. After this training, the business model can be applied to actual risk control business.

[0080] The method for business risk control provided in the specification will be described below from the perspective of a risk control scenario in which a business model is used to audit the business involved in the complaint text information of a user.

[0081] Figure 3 The flowchart of the method for business risk control in the specification specifically includes the following steps:

[0082] S300: Obtain complaint text information input by a user when performing a risk control business and business data corresponding to the complaint text information.

[0083] S302: Input the complaint text information, the business data, and a risk control rule into a pre-trained business model to determine, through a result output by a first encoding layer, a second encoding layer, and a third encoding layer in the business model, whether the user performing the risk control business complies with a judgment result of the risk control rule. The business model is trained by a model training method.

[0084] S304: Perform business risk control on the user according to the judgment result.

[0085] In actual application scenarios, after the business model is trained in the above manner, the complaint text information submitted by a user in a risk control business can be audited by the business model. That is, whether the risk control business involved in the complaint text information submitted by the user needs to be subjected to business risk control is determined.

[0086] Specifically, the complaint text information input by the user when performing the risk control business and the business data corresponding to the complaint text information can be obtained, and the complaint text information, the business data, and the risk control rule are input into the pre-trained business model, so as to determine the judgment result of whether the user performs the risk control business in accordance with the risk control rule through the result output by the first encoding layer, the second encoding layer, and the third encoding layer in the business model. In this way, the business platform can determine whether to perform business risk control for the user through the judgment result. The business model is obtained through the above model training method.

[0087] The above process can be understood as that the prediction result output by the business model can represent whether the business involved in the complaint text information of the user needs to be subjected to business risk control, and the risk control rule mentioned here can be consistent with the risk control rule in the above model training process, that is, if the business model needs to be applied in the audit scene of business risk control, the risk control rule in the model training process is the risk control rule.

[0088] It should be noted that in the prediction process, the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the business data output by the first encoding layer and the third encoding layer can be fused first to obtain the fused business features corresponding to the user, and then the judgment result of whether the user performs the risk control business in accordance with the risk control rule is determined through the fused business features and the feature encoding corresponding to the risk control rule.

[0089] It should be further noted that if the business model outputs the judgment results corresponding to the first rule, the second rule, and the third rule respectively in the training process, when determining whether to perform business risk control for the user, it can be determined that business risk control needs to be performed for the user in the case that the complaint text information is determined to comply with the first rule, the second rule, and the third rule according to the judgment result.

[0090] As can be seen from the above method, the complaint text information is used as a bridge to build a connection between the business data and the risk control rule, that is, the first encoding layer and the second encoding layer are trained together, then the first encoding layer and the third encoding layer are trained together, and finally the first encoding layer, the second encoding layer, and the third encoding layer are trained together, so as to obtain the trained business model. The business model can be used to audit the business described by the complaint text information of the user, thereby improving the efficiency and accuracy of performing business risk control for the user.

[0091] The above is the model training and business risk control method provided by one or more embodiments of the present specification. Based on the same idea, the present specification also provides a model training and business risk control device, as shown in Figure 4 、 Figure 5 .

[0092] Figure 4 A device schematic diagram of model training is provided for the present specification, and specifically comprises:

[0093] The acquisition module 401 is configured to acquire complaint text information and service data corresponding to a service performed by a user in the past, and a risk control rule corresponding to the service;

[0094] The first pre-training module 402 is configured to input the complaint text information into a first encoding layer in a service model, and input the service data into a second encoding layer in the service model, and perform first training on the service model based on results output by the first encoding layer and the second encoding layer, wherein the first training process at least includes adjusting model parameters of the first encoding layer and the second encoding layer;

[0095] The second pre-training module 403 is configured to input the complaint text information into the trained first encoding layer in the service model, and input the risk control rule into a third encoding layer in the service model, and perform second training on the service model based on results output by the trained first encoding layer and the third encoding layer, wherein the second training process at least includes adjusting model parameters of the trained first encoding layer and the third encoding layer;

[0096] The training module 404 is configured to input the complaint text information into the trained first encoding layer in the service model, input the service data into the trained second encoding layer in the service model, and input the risk control rule into the trained third encoding layer in the service model, and perform training on the service model based on results output by the trained first encoding layer, the trained second encoding layer, and the trained third encoding layer.

[0097] Optionally, the first pre-training module 402 is specifically configured to input the complaint text information and the service data into a service model, to output feature encoding corresponding to the complaint text information through a first encoding layer in the service model, and to determine feature encoding corresponding to the service data through a second encoding layer in the service model; determine a prediction result based on the complaint text information and the service data according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the service data, and perform first training on the service model with the optimization target of minimizing the prediction result and a service label corresponding to the user.

[0098] Optionally, the risk control rule includes a first rule, a second rule, and a third rule, the first rule is used to indicate that the user has consumed business resources for the business corresponding to the complaint text information, the second rule is used to indicate that other users involved in the complaint text information have not returned the business resources consumed by the user, and the third rule is used to indicate that the user and other users involved in the complaint text information are in a preset relationship.

[0099] The second pre-training module 403 is specifically configured to input the complaint text information into the first encoding layer of the trained business model, and input the risk control rule into the third encoding layer of the business model, to obtain feature encoding corresponding to the complaint text information and feature encoding corresponding to the risk control rule; determine the judgment result corresponding to the first rule, the judgment result corresponding to the second rule, and the judgment result corresponding to the third rule according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the risk control rule; and perform second training on the business model, with the optimization target of minimizing the difference between the judgment result corresponding to the first rule and the first label corresponding to the user, the difference between the judgment result corresponding to the second rule and the second label corresponding to the user, and the difference between the judgment result corresponding to the third rule and the third label corresponding to the user.

[0100] Optionally, the training module 404 is specifically configured to obtain a prediction result of the business executed by the user according to the feature encoding corresponding to the complaint text information, the feature encoding corresponding to the business data, and the feature encoding corresponding to the risk control rule output by the first encoding layer, the second encoding layer, and the third encoding layer after training, and perform training on the business model with the optimization target of minimizing the difference between the prediction result and the business label corresponding to the user.

[0101] Figure 5 A device for business risk control is provided, and specifically includes:

[0102] The acquisition module 501 is configured to acquire complaint text information input by a user when performing a risk control business, and business data corresponding to the complaint text information.

[0103] The input module 502 is configured to input the complaint text information, the business data, and a risk control rule into a pre-trained business model, to determine a judgment result of whether the user performing the risk control business meets the risk control rule through a result output by a first encoding layer, a second encoding layer, and a third encoding layer in the business model, and the business model is trained by a model training method.

[0104] The risk control module 503 is configured to perform business risk control on the user according to the judgment result.

[0105] Optionally, the risk control rules include a first rule, a second rule, and a third rule. The first rule is used to indicate that the user has consumed business resources for the business corresponding to the complaint text information. The second rule is used to indicate that other users involved in the complaint text information have not returned the business resources consumed by the user. The third rule is used to indicate that the user and the other users involved in the complaint text information have a preset relationship.

[0106] The specification also provides a computer-readable storage medium storing a computer program, which can be used to execute the above-mentioned model training and business risk control method.

[0107] The specification also provides a computer program product Figure 6 The schematic structural diagram of the electronic device is shown. As shown in Figure 6 According to the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and of course, other hardware required by the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs to implement the above-mentioned model training and business risk control method. Of course, in addition to the software implementation, the specification does not exclude other implementation manners, such as a logic device or a combination of software and hardware, and so on, that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or a logic device.

[0108] In the 1990s, it was quite obvious to distinguish whether an improvement in a technology was in hardware (e.g., improvement in circuit structures of diodes, transistors, switches, etc.) or in software (improvement in method flow). However, as technology has evolved, many improvements in method flow today can be considered as direct improvements in hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structures by programming the improved method flow into hardware circuits. Therefore, it cannot be said that an improvement in a method flow cannot be implemented by hardware entity modules. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is an integrated circuit whose logic function is determined by user programming of the device. A digital system is "integrated" on a PLD by the designer programming it, rather than by asking a chip manufacturer to design and fabricate a custom integrated circuit chip. Moreover, instead of manually fabricating integrated circuit chips, this programming is now mostly implemented by "logic compiler" software, which is similar to software compilers used in program development, and the original code to be compiled is written in a specific programming language, which is called a hardware description language (HDL), and there are many such languages, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., and the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should be aware that, as long as the method flow is logically programmed in the above-mentioned hardware description languages and programmed into an integrated circuit, a hardware circuit implementing the logical method flow can be easily obtained.

[0109] The controller can be implemented in any suitable way, e.g. the controller can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code, e.g. software or firmware, executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, examples of controllers include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91 SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to being implemented in pure computer readable program code form, the controller can perfectly well be implemented by means of logic programmed into logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded microcontrollers, etc. to perform the same functions. The controller can thus be considered as a hardware component, and the means comprised therein for performing various functions can be considered as structures within the hardware component. Alternatively, or even, the means for performing various functions can be considered as both a software module implementing a method and a structure within a hardware component.

[0110] The systems, apparatuses, modules or units illustrated by the above embodiments can be implemented by computer chips or entities, or products with certain functions. A typical implementation device is a computer. Specifically, the computer can be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0111] For the sake of description, the above apparatuses are described in functional division and are described respectively. Of course, the functions of the units can be implemented in the same or multiple software and / or hardware when implementing the present specification.

[0112] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0113] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0114] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0115] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0116] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0117] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory, etc. The memory is an example of computer-readable media.

[0118] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0119] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0120] Those skilled in the art will appreciate that embodiments of the present specification can be provided as methods, systems or computer program products. Therefore, the present specification can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0121] The present specification can be described in the general context of computer-executable instructions, such as program modules, executed by computers. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present specification can also be practiced in distributed computing environments where tasks are performed by remote processing nodes connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage nodes.

[0122] The various embodiments described in this specification are described using a numbering of embodiments approach: these are each individually integrated contributions pertaining to different but related aspects of the description. Each of the various embodiments can stand on its own, and each can be combined with the subject matter of other embodiments to produce further embodiments. Where the same numbers appear in different embodiments, such numbers are used for the sake of ease of understanding only and do not imply that the embodiments in which such numbers appear are the same or similar.

[0123] The above description is embodied in the form of only a description of embodiments of the present specification, and is not intended to limit the present specification. Various changes and modifications can be made by those skilled in the art based on the present specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.

Claims

1. A method for model training, comprising: obtaining complaint text information and business data corresponding to a business performed by a user in the past, and a risk control rule corresponding to the business; inputting the complaint text information into a first encoding layer in a business model, and inputting the business data into a second encoding layer in the business model, and performing first training on the business model based on results output by the first encoding layer and the second encoding layer, wherein the first training process at least comprises adjusting model parameters of the first encoding layer and the second encoding layer; inputting the complaint text information into the first encoding layer after training in the business model, and inputting the risk control rule into a third encoding layer in the business model, and performing second training on the business model based on results output by the first encoding layer after training and the third encoding layer, wherein the second training process at least comprises adjusting model parameters of the first encoding layer after training and the third encoding layer; inputting the complaint text information into the first encoding layer after training in the business model, inputting the business data into the second encoding layer after training in the business model, and inputting the risk control rule into the third encoding layer after training in the business model, and performing training on the business model based on results output by the first encoding layer after training, the second encoding layer after training, and the third encoding layer after training. 2.The method of claim 1, wherein inputting the complaint text information into a first encoding layer in a business model, and inputting the business data into a second encoding layer in the business model, and performing first training on the business model based on results output by the first encoding layer and the second encoding layer, comprises: inputting the complaint text information and the business data into the business model, to output feature encoding corresponding to the complaint text information by the first encoding layer in the business model, and to determine feature encoding corresponding to the business data by the second encoding layer in the business model; determining a prediction result based on the complaint text information and the business data according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the business data, and performing first training on the business model with an optimization objective of minimizing the prediction result and a business label corresponding to the user. 3.The method of claim 1, wherein the risk control rule comprises a first rule, a second rule, and a third rule, the first rule is used to represent that the user has consumed business resources for the business corresponding to the complaint text information, the second rule is used to represent that other users involved in the complaint text information have not returned the business resources consumed by the user, and the third rule is used to represent that the user and the other users involved in the complaint text information are in a preset relationship. inputting the complaint text information into the first encoding layer trained in the business model, and inputting the risk control rule into the third encoding layer in the business model, and performing second training on the business model based on the results output by the first encoding layer and the third encoding layer, comprising: inputting the complaint text information into the first encoding layer trained in the business model, and inputting the risk control rule into the third encoding layer in the business model, and obtaining the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the risk control rule; determining the judgment result corresponding to the first rule, the judgment result corresponding to the second rule, and the judgment result corresponding to the third rule according to the feature encoding corresponding to the complaint text information and the feature encoding corresponding to the risk control rule; performing second training on the business model with the optimization target of minimizing the difference between the judgment result corresponding to the first rule and the first label corresponding to the user, the difference between the judgment result corresponding to the second rule and the second label corresponding to the user, and the difference between the judgment result corresponding to the third rule and the third label corresponding to the user.

4. The method of claim 1, wherein the business model is trained based on the results output by the first encoding layer, the second encoding layer, and the third encoding layer after training, comprising: obtaining the prediction result of the business performed by the user according to the feature encoding corresponding to the complaint text information, the feature encoding corresponding to the business data, and the feature encoding corresponding to the risk control rule output by the first encoding layer, the second encoding layer, and the third encoding layer after training, respectively; training the business model with the optimization target of minimizing the difference between the prediction result and the business label corresponding to the user.

5. A method for business risk control, comprising: obtaining complaint text information input by a user when performing a risk control business, and business data corresponding to the complaint text information; inputting the complaint text information, the business data, and a risk control rule into a pre-trained business model to determine a judgment result of whether the user performing the risk control business complies with the risk control rule based on the results output by a first encoding layer, a second encoding layer, and a third encoding layer in the business model, the business model being trained by the method of any one of claims 1-4; performing business risk control on the user according to the judgment result.

6. The method of claim 5, wherein the risk control rule comprises a first rule, a second rule, and a third rule, the first rule being used to indicate that the user has consumed business resources for the business corresponding to the complaint text information, the second rule being used to indicate that other users involved in the complaint text information have not returned the business resources consumed by the user, and the third rule being used to indicate that the user and the other users involved in the complaint text information are in a preset relationship.

7. An apparatus for model training, comprising: The acquisition module is used to obtain complaint text information and business data corresponding to the business performed by the user in the past, as well as the risk control rules corresponding to the business; a first pre-training module, configured to input the complaint text information into a first coding layer of a business model, and input the business data into a second coding layer of the business model, and perform a first training on the business model based on outputs of the first coding layer and the second coding layer, wherein the first training process at least includes adjusting model parameters of the first coding layer and the second coding layer; A second pre-training module is configured to input the complaint text information into the trained first coding layer of the business model, and input the risk control rules into the third coding layer of the business model, and perform a second training on the business model based on the output results of the trained first coding layer and the third coding layer, wherein the second training process at least includes adjusting the model parameters of the trained first coding layer and the third coding layer; A training module is used to input the complaint text information into the trained first coding layer in the business model, input the business data into the trained second coding layer in the business model, and input the risk control rules into the trained third coding layer in the business model, and train the business model based on the output results of the trained first coding layer, the trained second coding layer, and the trained third coding layer.

8. A business risk control device, comprising: An acquisition module is used to obtain the complaint text information entered by the user when performing risk control business, as well as the business data corresponding to the complaint text information; An input module, configured to input the complaint text information, the business data, and the risk control rules into a pre-trained business model, so as to determine, based on the output of the business model through the first coding layer, the second coding layer, and the third coding layer in the business model, whether the user complies with the risk control rules when performing the risk control business, wherein the business model is trained by the method according to any one of claims 1 to 4; The risk control module is used to perform business risk control on the user based on the judgment result.

9. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1 to 6.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 6 when executing the program.

Images