Risk identification method, device and equipment

Abstract

The embodiment of the invention discloses a risk identification method, device and equipment, and relates to the technical field of computers, and the method comprises the steps: obtaining business operation data related to the business operation behavior of a target merchant in a first business operation scene; encoding the business operation data to obtain data representation corresponding to the business operation data; inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks in a plurality of tasks for the target merchant, the multi-task learning network including a plurality of expert models and gating networks corresponding to the tasks, the plurality of tasks at least comprise a risk identification service of a target merchant in a first service operation scene and a risk identification service of the target merchant which is converted from the first service operation scene to a second service operation scene; and based on the risk representation of each task, determining a risk identification result for the target merchant through a pre-trained risk identification network.

Description

Technical Field

[0001] This document relates to the field of computer technology, and in particular to a risk identification method, device and equipment. Background Art

[0002] With the development of informatization and digitalization, and people's increasing attention to their privacy data, the scale and complexity of merchant risks are constantly increasing. Conventional risk decisions are usually very dependent on historical cases and expert opinions. For this reason, an automated risk decision-making system can be established. The premise of establishing an automated risk decision-making system is to be able to better predict the probability of merchant risk. Since merchants in different business operation scenarios (such as merchants in offline business operation scenarios (or offline merchants) and merchants in online business operation scenarios (or online merchants)) often have different behaviors before risks occur, how to predict the risk probability of merchants is an important topic that needs to be paid attention to at present.

[0003] Usually, the entire life cycle of a merchant can be divided into different period intervals, and then a corresponding model can be built for each period interval. However, the above method will consume a lot of resources for model construction and model maintenance. Therefore, it is necessary to provide a multi-stage merchant risk probability prediction mechanism based on the life cycle to improve the overall accuracy of risk probability prediction and enhance the decision-making ability of the automated decision-making system. Summary of the invention

[0004] The purpose of the embodiments of this specification is to provide a multi-stage merchant risk probability prediction mechanism based on the life cycle to improve the overall accuracy of risk probability prediction and enhance the decision-making ability of the automated decision-making system.

[0005] In order to implement the above technical solution, the embodiments of this specification are implemented as follows: A risk identification method provided in an embodiment of the present specification comprises: obtaining business operation data related to the business operation behavior of a target merchant in a first business operation scenario. Encoding the business operation data to obtain a data representation corresponding to the business operation data. Inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including the relationship between different tasks among multiple tasks for the target merchant, wherein the multi-task learning network comprises multiple expert models and a gating network corresponding to the tasks, wherein the multiple tasks at least include the risk identification business of the target merchant in the first business operation scenario and the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario. Based on the risk representation of each task, the risk identification results for the target merchant are determined respectively through the pre-trained risk identification network.

[0006] A risk identification device provided in an embodiment of the present specification includes: a data acquisition module, which acquires business operation data related to the business operation behavior of a target merchant in a first business operation scenario. An encoding module, which encodes the business operation data to obtain a data representation corresponding to the business operation data. A multi-task processing module, which inputs the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including the relationship between different tasks in a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario. A risk identification module, based on the risk representation of each task, determines the risk identification result for the target merchant through a pre-trained risk identification network.

[0007] A risk identification device is provided in an embodiment of the present specification, and the risk identification device includes: a processor; and a memory arranged to store computer executable instructions, wherein the executable instructions, when executed, cause the processor to: obtain business operation data related to the business operation behavior of a target merchant in a first business operation scenario. The business operation data is encoded to obtain a data representation corresponding to the business operation data. The data representation corresponding to the business operation data is input into a pre-trained multi-task learning network to obtain a risk representation of each task including the relationship between different tasks in a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario. Based on the risk representation of each task, the risk identification result for the target merchant is determined respectively through the pre-trained risk identification network.

[0008] The embodiments of this specification also provide a storage medium, which is used to store computer executable instructions, and the executable instructions implement the following process when executed by the processor: obtain business operation data related to the business operation behavior of the target merchant in the first business operation scenario. Encode the business operation data to obtain the data representation corresponding to the business operation data. Input the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including the relationship between different tasks in multiple tasks for the target merchant, and the multi-task learning network includes multiple expert models and a gating network corresponding to the task, and the multiple tasks include at least the risk identification business of the target merchant in the first business operation scenario and the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario. Based on the risk representation of each task, the risk identification results for the target merchant are determined respectively through the pre-trained risk identification network.

[0009] The embodiments of this specification also provide a computer program product, including a computer program, which implements the following process when executed by a processor: obtaining business operation data related to the business operation behavior of a target merchant in a first business operation scenario. Encoding the business operation data to obtain a data representation corresponding to the business operation data. Inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including the relationship between different tasks in a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario. Based on the risk representation of each task, the risk identification results for the target merchant are determined respectively through the pre-trained risk identification network. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings required for use in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative labor. Figure 1 This is an embodiment of a risk identification method of this specification; Figure 2 This is another risk identification method embodiment of this specification; Figure 3This is a conversion diagram of a business operation scenario in this manual; Figure 4 This is another embodiment of a risk identification method of this specification; Figure 5 This is another embodiment of a risk identification method of this specification; Figure 6 This is a schematic diagram of the risk identification process for this manual; Figure 7 This is a schematic diagram of another risk identification process in this manual; Figure 8 This is another embodiment of a risk identification method of this specification; Fig. 9 This is an embodiment of a risk identification device in this specification; Fig.10 This is an embodiment of a risk identification device in this specification. DETAILED DESCRIPTION

[0011] The embodiments of this specification provide a risk identification method, device and equipment.

[0012] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0013] The embodiments of this specification provide a multi-stage risk identification mechanism in the life cycle of a merchant. With the development of informatization and digitalization, the scale and complexity of merchant risks continue to rise. Conventional risk decisions are usually very dependent on historical cases and expert opinions, and after the punishment of merchants is issued, a large subsequent manual operation and maintenance cost is required, resulting in the current situation where the decision-making system is huge, the operation and maintenance flexibility is low, and the manpower and time consumption are large. To this end, an automated risk decision-making system can be established, and the premise of establishing an automated risk decision-making system is to be able to better predict the merchant's risk probability PRO (Probability of Risk Occurrence). Since the behavior and probability of merchants before the occurrence of risks in different business operation scenarios are often quite different, how to predict the risk probability of merchants is an important topic that needs attention at present.

[0014] Usually, the entire life cycle of a merchant can be divided into different cycle intervals, and then, a corresponding model can be constructed for each cycle interval. However, the above method will consume a lot of resources for the construction and maintenance of the model. In addition, the business operation data samples related to the business operation behavior of the merchant can be unified and merged, and a machine learning model or a deep learning model can be trained directly for a certain risk label. This learning method is simple and effective, but it ignores the merchant's life cycle information. The risk basis of merchants in different cycle intervals is quite different, which leads to poor model learning effect and poor risk learning ability, resulting in poor model decision-making efficiency. To this end, the embodiment of this specification proposes a multi-stage merchant risk probability prediction mechanism based on the life cycle to improve the overall accuracy of risk probability prediction and improve the decision-making ability of the automated decision-making system. For specific processing, please refer to the specific content in the following embodiments.

[0015] like Figure 1 As shown, the embodiment of this specification provides a risk identification method, and the execution subject of the method can be a terminal device or a server, etc., wherein the terminal device can be a mobile terminal device such as a mobile phone, a tablet computer, or a computer device such as a laptop or a desktop computer, or an IoT device (specifically such as a smart watch, a car device, etc.), etc., wherein the server can be an independent server, or a server cluster composed of multiple servers, etc., and the server can be a background server for financial services or online shopping services, or a background server for an application, etc. In this embodiment, the execution subject is taken as an example for detailed description. For the case where the execution subject is a terminal device, please refer to the following server situation processing, which will not be repeated here. The method can specifically include the following steps: In step S102, business operation data related to the business operation behavior of the target merchant in the first business operation scenario is obtained.

[0016] Among them, the first business operation scenario can be a scenario that is converted to another or more business operation scenarios, for example, a live business operation scenario, an offline business operation scenario, an online business operation scenario, etc., which can be set according to actual conditions. The target merchant can be any merchant, and the target merchant can be a merchant that sells designated goods, or a merchant that provides designated services (such as storage services, recharge services, etc. of cloud servers), etc., which can be set according to actual conditions, and the embodiments of this specification do not limit this. Business operation behaviors can include multiple, for example, business operation behaviors can include the transaction behavior of the target merchant, the association behavior between different merchants, the association behavior between the target merchant and the user, the related behavior of the target merchant, etc., which can be set according to actual conditions. Business operation data can include multiple, for example, business operation data can include the historical transaction data of the target merchant, the current transaction data of the target merchant, the association relationship between the target merchant and different merchants, the association relationship between the target merchant and the user, the account information of the target merchant, the transaction risk information of the target merchant, the situation of the target merchant being complained about, etc. One or more. Business operation data may also be composed of multiple different types of data. For example, business operation data may include one or more of text data, image data, video data, tabular data (i.e., Tabular Data), etc. The specific data may be set according to actual conditions, and the embodiments of this specification do not limit this.

[0017] During implementation, when it is necessary to detect or identify whether a merchant (i.e., the target merchant) in the first business operation scenario has specified risks (such as fraud risk, illegal financial transaction risk, privacy leakage risk, etc.), the business operation data related to the business operation behavior of the target merchant recorded in the specified time period or the time period before the current moment can be obtained. Specifically, the target merchant's historical transaction data in the past month or year (which may include transaction identification, user account information, transaction amount and transaction time, etc.), the relationship between the target merchant and different merchants, the relationship between the target merchant and different users, the target merchant's account information, the target merchant's transaction risk information, the target merchant's complaints, and other business operation data can be obtained.

[0018] Alternatively, when a user needs to execute a service (specifically, a resource transaction service (such as a payment service, a transfer service, etc.) provided by a merchant (i.e., a target merchant) in a first business operation scenario, the terminal device may record relevant data involved in the user's execution of the service, and the recorded data may include business operation data related to the business operation behavior of the target merchant in the first business operation scenario. In order to identify whether there is a preset risk in the above business processing, the server may obtain business operation data related to the business operation behavior of the target merchant from the data recorded in the terminal device, or the terminal device may send a risk identification request for the above business to the server, and the server may respond to the risk identification request and obtain business operation data related to the business operation behavior of the target merchant, etc.

[0019] In step S104, the business operation data is coded to obtain data representation corresponding to the business operation data.

[0020] The data representation may be presented in the form of a matrix, a vector, or other forms, which may be set according to actual conditions.

[0021] In implementation, the business operation data may contain a variety of different types of data. For different types of data, corresponding encoders or deep learning network models can be set, and the data of the specified type in the business operation data can be encoded and processed by the corresponding encoders or deep learning network models to obtain the data representation corresponding to each type of data. For example, for text data, the encoder or deep learning network model can be pre-trained, or the trained encoder or deep learning network model can be directly obtained from the specified database, and the above text data can be encoded and processed by the trained encoder or deep learning network model to obtain the data representation corresponding to the text data; for image data, the image encoder or image coding model can be pre-trained, or the trained image encoder or image coding model can be directly obtained from the specified database, and the above image data can be encoded and processed by the trained image encoder or image coding model to obtain the data representation corresponding to the image data, etc.

[0022] In addition, for a certain type of data, which may include multiple different categories of data, a corresponding encoder or deep learning network model can be set for each category of data. For example, for data of the "transaction identification" category, data of the "target merchant's account information" category, and data of the "transaction amount" category, corresponding encoders or deep learning network models can be set respectively. The specific settings can be based on actual conditions.

[0023] The data representations corresponding to different types of data may be fused and processed (for example, the data representations corresponding to different types of data may be concatenated or the data representations corresponding to different types of data may be fused and calculated through a specified algorithm, etc.) to obtain the data representation corresponding to the business operation data. Alternatively, the data representation corresponding to the business operation data may be determined by weighted summation. Specifically, corresponding weights may be set in advance for the data representations corresponding to different types of data, respectively, and then the data representations corresponding to different types of data may be multiplied by the corresponding weights. The data representations corresponding to different types of data may be multiplied by the corresponding weights and added together, and the result obtained may be used as the data representation corresponding to the business operation data. Alternatively, other algorithms may be set in advance, and the data representations corresponding to different types of data may be processed by the algorithm. The result obtained may be used as the data representation corresponding to the business operation data, etc. The specific setting may be based on actual conditions, and the embodiments of this specification are not limited to this.

[0024] In step S106, the data representation corresponding to the business operation data is input into a pre-trained multi-task learning network to obtain a risk representation of each task including the relationship between different tasks in multiple tasks for the target merchant. The multi-task learning network includes multiple expert models and gated networks corresponding to the tasks. The multiple tasks include at least the risk identification business of the target merchant in the first business operation scenario and the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario.

[0025] Among them, the multi-task learning network is a machine learning method that adopts the inductive transfer mechanism. The main goal is to use the specific domain information implicit in the training signals of multiple related tasks to improve the generalization ability of the network. The multi-task learning network achieves this goal by training multiple tasks in parallel using shared representations, that is, the multi-task learning network can be a network that learns multiple tasks in the same system. These tasks may have similar input spaces, output spaces or structures. The goal of the multi-task learning network is to improve the generalization ability and performance of the network model by learning multiple tasks. In order to achieve the above processing, multiple expert models and gated networks corresponding to the task can be set in the multi-task learning network. Specifically, four expert models and three gated networks can be set, or three expert models and three gated networks can be set, or three expert models and two gated networks can be set, etc., which can be set according to actual conditions. The internal structure of the expert model is no longer simply based on the automatic generation of algorithm-fitted data, but is built on rules with business meanings. The decision path of each judgment result can be intuitively displayed, and the triggered rules are clear at a glance. After the expert model is developed, it is equivalent to providing a standard model template for such scenarios, agreeing on the fields, features, and model files required by the model. When it is subsequently applied to the same scenario, it only needs to perform data ETL (i.e., extraction, conversion, loading) and feature calculation according to the constraints of the model, and cross-platform model files can also be quickly deployed and applied. The gated network can assign corresponding weights to the output data of each expert model. Risk characterization can be a characterization that can be input into the risk identification network to obtain the corresponding risk identification results and is targeted at the target merchant. Risk characterization can include multiple forms. Risk characterization can be presented in the form of vectors or matrices (in this case, the vectors or matrices corresponding to each risk can be pre-set), or, it can also be presented in the form of numerical values, etc., which can be set according to actual conditions, and the embodiments of this specification do not limit this. The second business operation scenario may be a business operation scenario different from the first business operation scenario. For example, the first business operation scenario is an offline business operation scenario, and the second business operation scenario may be a live broadcast type of business operation scenario; or, the first business operation scenario is a live broadcast type of business operation scenario, and the second business operation scenario may be an offline business operation scenario, etc. The specific settings may be based on actual conditions.

[0026] In implementation, the multi-task learning network can be processed in the following ways: construct a shared feature representation: first, it is necessary to construct a shared feature representation, which can be a low-dimensional representation of the input data or a high-dimensional representation obtained by a certain feature extraction algorithm; learn task-specific output layers: in a neural network, it is necessary to learn the output layer of each task, which can be different tasks such as classification, regression, and semantic labeling; train the multi-task learning network. When training the multi-task learning network, it is necessary to consider the loss function of each task, and the above loss functions can be added to obtain the total loss function. By optimizing the total loss function, multiple tasks can be learned simultaneously; evaluate model performance: when evaluating model performance, the performance of each task can be evaluated through the test set, and the average performance of the multi-task learning network can be calculated.

[0027] In practical applications, the multi-task learning network may include multiple expert models and a gating network. The data representations corresponding to the business operation data may be input into the multiple expert models in the multi-task learning network respectively. After each expert model processes the data representation, the corresponding output data may be obtained. At the same time, the data representation corresponding to the business operation data may be input into the gating network. The gating network may assign corresponding weights to each expert model. Then, the product between the output data of the expert model and the corresponding weight may be calculated by weighted summation. The results of the multiplication may then be added. The above method may be used to determine the relationship between different tasks in the multiple tasks for the target merchant, thereby obtaining the result of the weighted summation. The risk representation of each task may be determined based on the obtained weighted summation result.

[0028] In step S108, based on the risk characterization of each task, the risk identification results for the target merchants are determined by using the pre-trained risk identification network.

[0029] Among them, the risk identification network can be constructed in a variety of ways. For example, the risk identification network can be constructed through specified network layers (such as fully connected layers, convolutional layers, activation function layers, etc.), or through multi-layer perceptrons, etc. The specific settings can be based on actual conditions.

[0030] In implementation, the risk characterization of each task can be input into the subnetwork corresponding to each task in the risk identification network, and the risk identification results output by the subnetwork corresponding to each task can be obtained. The risk identification results for the target merchant can be determined based on the risk identification results output by the subnetwork corresponding to each task. For example, the first business operation scenario is an offline business operation scenario, and the second business operation scenario can be a live broadcast type business operation scenario. Based on this, two tasks can be set, namely, the risk identification task under the offline business operation scenario and the risk identification task under the business operation scenario converted from the offline business operation scenario to the live broadcast type. Through the above processing, the risk identification results of the target merchant in the offline business operation scenario and the risk identification results of the target merchant in the business operation scenario converted to the live broadcast type can be determined. After obtaining the risk identification results for the target merchant in the above manner, the risk identification results for the target merchant can be provided to the subsequent system or business for processing. For example, the risk identification results for the target merchant can be provided to the automated decision-making system to execute the corresponding risk strategy for the target merchant, such as limiting the target merchant's collection amount or freezing the target merchant's account, etc., which can be set according to actual conditions.

[0031] It should be noted that the conversion of the target merchant between the first business operation scenario and the second business operation scenario can be a complete conversion, for example, the target merchant is a merchant in the first business operation scenario, and the target merchant is only converted to the second business operation scenario; or, it can also be a part of a conversion process, such as the target merchant is a merchant in the first business operation scenario, and the target merchant is converted to the second business operation scenario, and then the target merchant is converted to the first business operation scenario (or the third business operation scenario, etc.), etc.

[0032] The embodiment of the present specification provides a risk identification method, which obtains business operation data related to the business operation behavior of a target merchant in a first business operation scenario, and then encodes and processes the business operation data to obtain data representations corresponding to the business operation data. After that, the data representations corresponding to the business operation data can be input into a pre-trained multi-task learning network to obtain risk representations of each task including the relationship between different tasks in multiple tasks for the target merchant. The multi-task learning network includes multiple expert models and gated networks corresponding to the tasks. The multiple tasks at least include risk identification services for target merchants in the first business operation scenario and risk identification services for target merchants converted from the first business operation scenario to the second business operation scenario. Finally, based on the risk representation of each task, the risk identification results for the target merchant can be determined by the pre-trained risk identification network. In this way, a multi-task learning network is used to learn for different life cycles of the target merchant, thereby greatly improving the risk identification efficiency of multi-stage merchants in the life cycle, and at the same time greatly reducing the maintenance difficulty of the risk identification system. In addition, the subsequent merchant risk decision system relies on the accuracy of merchant risk prediction or risk identification. The improvement of merchant risk prediction accuracy can improve the merchant decision-making efficiency. Moreover, through multi-task learning, we can simultaneously learn: the prediction results of the preset risks that occur directly without conversion (that is, in the first business operation scenario); the prediction results of the preset risks that occur to merchants after conversion to the second business operation scenario, etc. Since the behavioral performance of merchants before the risk occurs in the first business operation scenario is often quite different from that in the second business operation scenario, the above method can greatly improve the model effect.

[0033] In actual applications, in the process of risk identification for target merchants, since the behavior performance and probability of target merchants before the preset risk occurs in offline business operation scenarios and online business operation scenarios are often quite different, if the entire life cycle of the target merchant is directly combined and unified to directly predict the probability of subsequent risk occurrence, the learning difficulty is relatively large and the learning effect is often poor. Based on this, the first business operation scenario is an offline business operation scenario, and the second business operation scenario is an online business operation scenario, or the first business operation scenario is an online business operation scenario, and the second business operation scenario is an offline business operation scenario. Among them, the offline business operation scenario can be an operation scenario through an offline physical store, and the online business operation scenario can be an operation scenario through an online store in an online trading platform. In the above life cycle of the target merchant, the merchant in the online business operation scenario is often registered through the applet of the offline business operation scenario to the online business operation scenario, and the behavior performance and probability of the target merchant before the preset risk occurs in the offline business operation scenario and the online business operation scenario are often quite different.

[0034] In practical applications, the specific processing methods of the above step S108 can be varied. An optional processing method is provided below, such as Figure 2 As shown, the processing may specifically include the following steps S1082 to S1086.

[0035] In step S1082, the risk representation corresponding to the risk identification business of the target merchant in the first business operation scenario is input into the corresponding risk identification network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario.

[0036] In step S1084, based on the risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario is determined through the risk identification network corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario.

[0037] In step S1086, the risk identification result for the target merchant is determined based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario.

[0038] During implementation, the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario can be used as the risk identification result for the target merchant, or the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario can be fused and calculated, and the calculated result can be used as the risk identification result for the target merchant, etc. The specific setting can be based on actual conditions, and the embodiments of this specification do not limit this.

[0039] In actual applications, the life cycle of target merchants in offline business operation scenarios can be described as follows: Figure 3As shown, the target merchant may generate preset risks in the identity of an offline business operation scenario. If the target merchant is converted from an offline business operation scenario to an online business operation scenario, the target merchant may also generate preset risks after converting to an online business operation scenario. For the life cycle of the target merchant in the offline business operation scenario, the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes the first sub-business of the business operation scenario conversion (that is, the first sub-business can be a sub-business converted from the first business operation scenario to the second business operation scenario (such as the first sub-business can be a sub-business converted from the offline business operation scenario to the online business operation scenario)) and the second sub-business of the target merchant that converts the business operation scenario (that is, the second sub-business can be a sub-business converted from the second business operation scenario (such as the second sub-business can be a sub-business converted from the offline business operation scenario to the online business operation scenario)). The risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes the first sub-business The corresponding first sub-representation (which may include one or more of the target merchant's transaction behavior, the association behavior between the target merchant and different merchants, the association behavior between the target merchant and the user, the target merchant's related behavior, etc.) and the second sub-representation corresponding to the second sub-business (which may include one or more of the target merchant's transaction behavior, the association behavior between the target merchant and different merchants, the association behavior between the target merchant and the user, the target merchant's related behavior, etc.), the risk identification network corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes the first sub-network corresponding to the first sub-business and the second sub-network corresponding to the second sub-business. Based on this, the specific processing methods of the above step S1084 can be various. The following provides an optional processing method, such as Figure 4 As shown, the processing may specifically include the following steps S10842 to S10846.

[0040] In step S10842, the first sub-representation corresponding to the first sub-business is input into the first network to obtain a conversion probability of the target merchant converting from the first business operation scenario to the second business operation scenario.

[0041] In step S10844, the second sub-representation corresponding to the second sub-business is input into the second sub-network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant converted into the second business operation scenario.

[0042] In step S10846, based on the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario, the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario is determined.

[0043] During implementation, the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario can be multiplied by the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario to obtain a corresponding result, and the corresponding result obtained can be determined as the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario. Alternatively, a corresponding algorithm can be pre-set according to actual conditions, and the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario are calculated by the algorithm. The calculated result is used as the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, etc. It can be specifically set according to actual conditions, and the embodiments of this specification do not limit this.

[0044] In practical applications, business operation data includes tabular data, and the data representation corresponding to the business operation data includes data representation corresponding to various different tabular features contained in the tabular data, wherein the tabular data (TabularData) can be data in a tabular style, and the tabular data can include categorical tabular features and numerical tabular features, as shown in Table 1.

[0045] Table 1

[0046] In the above Table 1, "account information of target merchant", "account information of user", "transaction bill identification", "transaction amount" and "transaction time" can be categorical table features, and the remaining data in Table 1 such as "10000000", "0010005", "0058296", "32000003", "42000204", "58701258", "500", "3000", "1000", "10:50:42" can be numerical table features. In resource trading business, information recommendation business, data mining business and other businesses, the most commonly used data mode is tabular data. In addition, in risk identification and risk prevention and control, tabular data can also be the most commonly used data form. Based on this, the tabular data of users in the process of executing the target business can be obtained.

[0047] Based on the above content, the specific processing methods of the above step S106 can be various. The following is an optional processing method, such as Figure 5 As shown, the processing may specifically include the following steps S1062 to S1066.

[0048] In step S1062, data representations corresponding to various table features contained in the business operation data are spliced ​​to obtain a spliced ​​data representation.

[0049] In implementation, the data representations corresponding to various different table features may include data representations corresponding to categorical table features and data representations corresponding to numerical table features. The data representations corresponding to categorical table features and data representations corresponding to numerical table features may be presented in the form of a matrix, a vector, or other forms, etc., which may be set specifically according to actual conditions. For categorical table features, since categorical table features are usually text or corresponding codes, which are generally relatively fixed, the categorical table features may be converted into corresponding Embedding values ​​through the EmbeddingLook-Up operation to project them into a high-dimensional feature space to enhance their expression capabilities. Specifically, the mapping relationship between categorical table features and embedding vectors may be pre-set, and the mapping relationship may be stored in a specified database or storage device. When the table data is obtained, the categorical table features may be extracted from it, and the embedding vectors corresponding to the categorical table features in the table data may be searched from the above mapping relationship, so that the categorical table features in the table data may be converted into corresponding embedding vectors, and the embedding vectors obtained after the conversion may be used as the data representations corresponding to the categorical table features. For numerical table features, the data representation corresponding to the numerical table features can usually be obtained by inputting the original value or performing bucket or bin processing (which can include multiple types, such as equal frequency bin processing, equal width bin processing, etc., which can be set according to actual conditions) and then performing embedding processing. After that, the data representations corresponding to the various different table features contained in the business operation data can be spliced ​​to obtain the spliced ​​data representation.

[0050] In step S1064, the concatenated data representation is input into each expert model in the multi-task learning network respectively to obtain output data of each expert model.

[0051] Among them, multiple expert models (i.e., Experts) can be parallel network models, each of which learns for the entire sample set, but as the model training proceeds, each expert model will usually specialize and become an "expert" in processing a specific type of data or a specific task.

[0052] In step S1066, the concatenated data representation and the output data of each expert model are input into the gating network corresponding to the task to obtain the weight assigned to each expert model by the gating network, and based on the output data of each expert model and the weight corresponding to each expert model, the risk representation of each task including the relationship between different tasks in multiple tasks for the target merchant is determined.

[0053] Among them, the task of the gating network is to dynamically assign the weight of the expert model to each data point. The gating network decides which expert model's output data should dominate the final prediction based on the input data.

[0054] In implementation, the concatenated data representation and the output data of each expert model can be input into the gating network corresponding to the task, and the gating network corresponding to the task can be used to determine which expert model output data should dominate the final prediction, thereby obtaining the weight assigned to each expert model by the gating network. The output data of a certain expert model can be multiplied by the weight corresponding to the expert model to obtain the corresponding result. The above method can be used to obtain the result of multiplying the output data of each expert model by its corresponding weight, and the calculated results can be added together to obtain the result as the risk representation of each task.

[0055] In practical applications, the number of gated networks in the above step S1066 is determined by the number of multiple tasks, and each task corresponds to a gated network. The corresponding structure can be a MMOE network structure, such as Figure 6 As shown, in the above step S1066, the spliced ​​data representation and the output data of each expert model are input into the gating network corresponding to the task, and the specific processing method for obtaining the weight assigned by the gating network to each expert model can be various. An optional processing method is provided below, which may specifically include the following contents: the spliced ​​data representation and the output data of each expert model are respectively input into the gating network corresponding to each task, and the weight assigned by each gating network to each expert model is obtained.

[0056] In addition, the multiple expert models in the above step S106 include an expert model corresponding to each task and a shared expert model shared by multiple tasks. The corresponding structure can be a PLE network structure, such as Figure 7 As shown, based on this, the specific processing methods of the above step S106 can be various, and an optional processing method is provided below, which can specifically include the processing of the following steps A2 and A4.

[0057] In step A2, the data representation corresponding to the business operation data is input into the expert model corresponding to each task and the shared expert model shared by multiple tasks, respectively, to obtain the output data of the expert model corresponding to each task and the output data of the shared expert model.

[0058] In implementation, the multiple expert models may include two types, one is an expert model corresponding to each task, and the other is a shared expert model shared by multiple tasks, wherein the shared expert model may include one or more, and may be selected specifically according to actual conditions. In order to simplify the processing process, the shared expert model may include one, and in this case, all tasks may share one shared expert model. The shared expert model and the expert model corresponding to each task are basically the same as the expert model mentioned above in processing data, and there is no obvious difference. Therefore, the specific processing process of the above step A2 can refer to the above-mentioned related content, and will not be repeated here.

[0059] In step A4, for the first task among multiple tasks, the data representation corresponding to the business operation data is input into the gating network corresponding to the first task, and the weights assigned by the first gating network to the output data of the expert model corresponding to the first task and the output data of the shared expert model are obtained; based on the output data and the corresponding weights of the expert model corresponding to the first task, and the output data and the corresponding weights of the shared expert model, the risk representation of the first task is determined to determine the risk representation of each task.

[0060] The first task may be any one of the multiple tasks.

[0061] In implementation, for the first task, the data representation corresponding to the business operation data can be input into the gating network corresponding to the first task, and the weights assigned by the first gating network to the output data of the expert model corresponding to the first task and the output data of the shared expert model can be obtained. Then, the output data of the expert model corresponding to the first task can be multiplied by the corresponding weights to obtain the corresponding results. The output data of the shared expert model can be multiplied by the corresponding weights to obtain the corresponding results. The above two results can be added together, and the obtained result is used as the risk representation of the first task. In the above manner, the risk representations of other tasks in multiple tasks can be obtained, and then the risk representation of each task can be obtained.

[0062] The multi-task learning network and the risk identification network may be trained in various ways. An optional processing method is provided below. For details, please refer to the processing of the following steps B2 to B10.

[0063] In step B2, a business operation data sample related to the business operation behavior of the first merchant in the first business operation scenario is obtained for model training.

[0064] The first merchant may be any merchant, and further, the first merchant may include a target merchant.

[0065] In step B4, the business operation data sample is encoded to obtain a sample data representation corresponding to the business operation data sample.

[0066] The specific processing procedures of the above steps B2 and B4 can be found in the above related contents and will not be repeated here.

[0067] In step B6, the sample data representation corresponding to the business operation data sample is input into the multi-task learning network to obtain a risk representation sample for each task including the relationship between different tasks in multiple tasks for the first merchant, and the multiple tasks include at least the risk identification business of the target merchant in the first business operation scenario and the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario.

[0068] The specific processing process of the above step B6 can refer to the relevant content of the above step S106, which will not be repeated here.

[0069] In practical applications, the processing of step B6 may also include: multiple expert models are expert models shared by multiple tasks, and the sample data representations corresponding to the business operation data samples are respectively input into each expert model in the multi-task learning network to obtain the output data samples of each expert model; the sample data representations corresponding to the business operation data samples and the output data samples of each expert model are input into the gating network corresponding to the task to obtain the weight assigned by the gating network to each expert model; based on the output data samples of each expert model and the sample weights corresponding to each expert model, the risk representation samples of each task are determined.

[0070] In addition, the number of gated networks is determined by the number of multiple tasks, and each task corresponds to a gated network. The sample data representation corresponding to the business operation data sample and the output data sample of each expert model are respectively input into the gated network corresponding to the task, and the sample weight assigned by the gated network to each expert model is obtained, including: the sample data representation corresponding to the business operation data sample and the output data sample of each expert model are respectively input into the gated network corresponding to each task, and the sample weight assigned by each gated network to each expert model is obtained.

[0071] In addition, the multiple expert models include an expert model corresponding to each task and a shared expert model shared by multiple tasks. The processing of step B6 may also include: inputting the sample data representation corresponding to the business operation data sample into the expert model corresponding to each task and the shared expert model shared by multiple tasks, respectively, to obtain the output data sample of the expert model corresponding to each task and the output data sample of the shared expert model; for the first task among the multiple tasks, inputting the sample data representation corresponding to the business operation data sample into the gating network corresponding to the first task, to obtain the weights assigned by the first gating network to the output data sample of the expert model corresponding to the first task and the output data sample of the shared expert model; based on the output data sample and the corresponding sample weight of the expert model corresponding to the first task, as well as the output data sample and the corresponding sample weight of the shared expert model, determining the risk representation sample of the first task, so as to determine the risk representation sample of each task.

[0072] In step B8, based on the risk characterization of each task, the risk identification result of the first merchant is determined through the risk identification network.

[0073] The specific processing process of the above step B8 can refer to the relevant content of the above step S108, which will not be repeated here.

[0074] In step B10, based on the risk identification result of the first merchant and a preset loss function, the multi-task learning network and the risk identification network are jointly trained to obtain a trained multi-task learning network and a trained risk identification network.

[0075] Among them, the loss function may include multiple types, for example, a mean square error loss function or a cross entropy loss function, etc., which can be set specifically according to actual conditions, and the embodiments of this specification do not limit this.

[0076] In actual applications, the above-mentioned loss function includes a first sub-loss function and a second sub-loss function. The first sub-loss function is constructed through a cross-entropy loss function based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk label corresponding to the target merchant in the first business operation scenario (i.e., whether the target merchant incurs preset risks in the first business operation scenario (such as an offline business operation scenario)). The second sub-loss function is constructed through a cross-entropy loss function based on the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario, as well as the risk label corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario (i.e., whether the target merchant incurs preset risks in the second business operation scenario (such as an online business operation scenario) after the conversion).

[0077] Among them, the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario may include a first sub-business and a second sub-business. Based on this, the label corresponding to the first sub-business can be whether the target merchant has converted from the first business operation scenario to the second business operation scenario, and the label corresponding to the second sub-business can be whether the target merchant has incurred a preset risk in the identity of the second business operation scenario after the conversion, etc. Based on this, the output data of the first sub-business can be multiplied by the output data of the second sub-business and a conditional probability can be calculated, and then the cross entropy loss can be calculated with the label corresponding to the second sub-business, thereby obtaining the above-mentioned loss function.

[0078] The following is a detailed description of a risk identification method provided by an embodiment of this specification in combination with specific application scenarios, wherein the first business operation scenario can be an offline business operation scenario, the second business operation scenario can be an online business operation scenario, and the business operation data can include tabular data, which can include historical transaction data between the target merchant and different users (specifically, it can include user behavior information, account information of both parties to the transaction, transaction amount, transaction time and transaction location, etc.), target merchant account information, target merchant historical risk information, etc. The multi-task learning network includes four expert models and three gated networks.

[0079] like Figure 8 As shown, the embodiment of this specification provides a risk identification method, and the execution subject of the method can be a terminal device or a server, etc., wherein the terminal device can be a mobile terminal device such as a mobile phone, a tablet computer, or a computer device such as a laptop or a desktop computer, or an IoT device (specifically such as a smart watch, a car device, etc.), etc., wherein the server can be an independent server, or a server cluster composed of multiple servers, etc., and the server can be a background server for financial services or online shopping services, or a background server for an application, etc. In this embodiment, the execution subject is taken as an example for detailed description. For the case where the execution subject is a terminal device, please refer to the following server situation processing, which will not be repeated here. The method can specifically include the following steps: In step S802, a tabular data sample related to the business operation behavior of a first merchant in an offline business operation scenario is obtained for model training.

[0080] In step S804, encoding processing is performed on the table data sample to obtain a sample data representation corresponding to the table data sample.

[0081] In step S806, the sample data representation corresponding to the tabular data sample is input into the multi-task learning network to obtain a risk representation sample for each task including the relationship between different tasks in multiple tasks for the first merchant, and the multiple tasks include at least the risk identification business of the target merchant in the offline business operation scenario and the risk identification business of the target merchant converted from the offline business operation scenario to the online business operation scenario.

[0082] In step S808, based on the risk characterization samples of each task, the risk identification results of the first merchant are determined through the risk identification network.

[0083] In step S810, based on the risk identification result of the first merchant and a preset loss function, the multi-task learning network and the risk identification network are jointly trained to obtain a trained multi-task learning network and a trained risk identification network.

[0084] Among them, the loss function includes a first sub-loss function and a second sub-loss function. The first sub-loss function is based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the offline business operation scenario and the risk label corresponding to the target merchant in the offline business operation scenario, and is constructed through the cross entropy loss function. The second sub-loss function is based on the conversion probability of the target merchant from the offline business operation scenario to the online business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the online business operation scenario, as well as the risk label corresponding to the risk identification business of the target merchant converted from the offline business operation scenario to the online business operation scenario, and is constructed through the cross entropy loss function.

[0085] In step S812, tabular data related to the business operation behavior of the target merchant in the offline business operation scenario is obtained.

[0086] In step S814, encoding processing is performed on the table data to obtain data representation corresponding to the table data.

[0087] In step S816, data representations corresponding to a plurality of different table features contained in the table data are spliced ​​to obtain a spliced ​​data representation.

[0088] In step S818, the concatenated data representation is input into each expert model in the multi-task learning network respectively to obtain output data of each expert model.

[0089] In step S820, the concatenated data representation and the output data of each expert model are respectively input into the gating network corresponding to each task to obtain the weight assigned by each gating network to each expert model, and based on the output data of each expert model and the weight corresponding to each expert model, the risk representation of each task including the relationship between different tasks in multiple tasks for the target merchant is determined.

[0090] In step S822, the risk representation corresponding to the risk identification business of the target merchant in the offline business operation scenario is input into the corresponding risk identification network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant in the offline business operation scenario.

[0091] In step S824, the representation corresponding to the business (i.e., the first sub-business mentioned above) of the business operation scenario conversion from the offline business operation scenario to the online business operation scenario is input into the corresponding risk identification network to obtain the conversion probability of the target merchant from the offline business operation scenario to the online business operation scenario.

[0092] In step S826, the representation corresponding to the risk identification business of the target merchant converted into the online business operation scenario (i.e., the second sub-business mentioned above) is input into the corresponding risk identification network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant converted into the online business operation scenario.

[0093] In step S828, based on the conversion probability of the target merchant from the offline business operation scenario to the online business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the online business operation scenario, the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the offline business operation scenario is determined.

[0094] In step S830, the risk identification result for the target merchant is determined based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the offline business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the offline business operation scenario to the online business operation scenario.

[0095] It should be noted that the conversion of the target merchant between the offline business operation scenario and the online business operation scenario can be a complete conversion. For example, if the target merchant is a merchant in an offline business operation scenario, the target merchant is only converted to an online business operation scenario. Alternatively, it can be a part of a conversion process. For example, if the target merchant is a merchant in an offline business operation scenario, the target merchant is converted to an online business operation scenario, and then the target merchant is converted to an offline business operation scenario, etc. Then, the conversion of the target merchant between the offline business operation scenario and the online business operation scenario can be the conversion of the offline business operation scenario in the first half to the online business operation scenario, or the conversion of the online business operation scenario in the second half to the offline business operation scenario.

[0096] The embodiment of the present specification provides a risk identification method, which obtains business operation data related to the business operation behavior of a target merchant in a first business operation scenario, and then encodes and processes the business operation data to obtain data representations corresponding to the business operation data. After that, the data representations corresponding to the business operation data can be input into a pre-trained multi-task learning network to obtain risk representations of each task including the relationship between different tasks in multiple tasks for the target merchant. The multi-task learning network includes multiple expert models and gated networks corresponding to the tasks. The multiple tasks at least include risk identification services for target merchants in the first business operation scenario and risk identification services for target merchants converted from the first business operation scenario to the second business operation scenario. Finally, based on the risk representation of each task, the risk identification results for the target merchant can be determined by the pre-trained risk identification network. In this way, a multi-task learning network is used to learn for different life cycles of the target merchant, thereby greatly improving the risk identification efficiency of multi-stage merchants in the life cycle, and at the same time greatly reducing the maintenance difficulty of the risk identification system. In addition, the subsequent merchant risk decision system relies on the accuracy of merchant risk prediction or risk identification. The improvement of merchant risk prediction accuracy can improve the merchant decision-making efficiency. Moreover, through multi-task learning, we can simultaneously learn: the prediction results of the preset risks that occur directly without conversion (that is, in the first business operation scenario); the prediction results of the preset risks that occur to merchants after conversion to the second business operation scenario, etc. Since the behavioral performance of merchants before the risk occurs in the first business operation scenario is often quite different from that in the second business operation scenario, the above method can greatly improve the model effect.

[0097] The above is a risk identification method provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a risk identification device, such as Fig. 9 shown.

[0098] The risk identification device includes: a data acquisition module 901, a coding module 902, a multi-task processing module 903 and a risk identification module 904, wherein: A data acquisition module 901 acquires business operation data related to business operation behaviors of a target merchant in a first business operation scenario; The encoding module 902 encodes the business operation data to obtain a data representation corresponding to the business operation data; The multi-task processing module 903 inputs the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks among a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of the target merchant in a first business operation scenario and a risk identification business of the target merchant converted from the first business operation scenario to a second business operation scenario; The risk identification module 904 determines the risk identification result for the target merchant through a pre-trained risk identification network based on the risk characterization of each task.

[0099] In the embodiments of this specification, the first business operation scenario is an offline business operation scenario, and the second business operation scenario is an online business operation scenario, or the first business operation scenario is an online business operation scenario, and the second business operation scenario is an offline business operation scenario.

[0100] In the embodiment of this specification, the risk identification module 904 includes: The first prediction unit inputs the risk representation corresponding to the risk identification business of the target merchant in the first business operation scenario into the corresponding risk identification network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario; A second prediction unit determines, based on the risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, a risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario through a risk identification network corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario; The risk identification unit determines a risk identification result for the target merchant based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario.

[0101] In an embodiment of the present specification, the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes a first sub-business of the business operation scenario conversion and a second sub-business of the target merchant of the converted business operation scenario, the risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes a first sub-characterization corresponding to the first sub-business and a second sub-characterization corresponding to the second sub-business, and the risk identification network corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes a first sub-network corresponding to the first sub-business and a second sub-network corresponding to the second sub-business, The second prediction unit inputs the first sub-characterization corresponding to the first sub-business into the first network to obtain the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario; inputs the second sub-characterization corresponding to the second sub-business into the second sub-network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario; and determines the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario based on the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario and the risk occurrence probability corresponding to the second sub-business of the target merchant converted to the second business operation scenario.

[0102] In the embodiment of the present specification, the business operation data includes tabular data, and the data representation corresponding to the business operation data includes data representations corresponding to a plurality of different tabular features contained in the tabular data. The multi-task processing module 903 includes: A splicing unit splices data representations corresponding to a plurality of different table features contained in the business operation data to obtain a spliced ​​data representation; A first processing unit, inputting the concatenated data representation into each expert model in the multi-task learning network respectively, to obtain output data of each expert model; The second processing unit inputs the spliced ​​data representation and the output data of each expert model into the gating network corresponding to the task, obtains the weight assigned by the gating network to each expert model, and determines the risk representation of each task including the relationship between different tasks in the multiple tasks for the target merchant based on the output data of each expert model and the weight corresponding to each expert model.

[0103] In an embodiment of the present specification, the number of the gating networks is determined by the number of the multiple tasks, and each task corresponds to a gating network. The second processing unit inputs the spliced ​​data representation and the output data of each expert model into the gating network corresponding to each task, respectively, to obtain the weight assigned by each gating network to each expert model.

[0104] In the embodiment of this specification, the device further includes: A sample acquisition module, which acquires a business operation data sample related to the business operation behavior of a first merchant in a first business operation scenario for model training; A sample encoding module encodes the business operation data sample to obtain a sample data representation corresponding to the business operation data sample; a sample prediction module, inputting the sample data representation corresponding to the business operation data sample into the multi-task learning network, and obtaining a risk representation sample of each task including a relationship between different tasks in a plurality of tasks for the first merchant, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to the second business operation scenario; A sample risk identification module, based on the risk characterization samples of each task, determines the risk identification result of the first merchant through the risk identification network; The training module jointly trains the multi-task learning network and the risk identification network based on the risk identification result of the first merchant and a preset loss function to obtain a trained multi-task learning network and a trained risk identification network.

[0105] In an embodiment of the present specification, the loss function includes a first sub-loss function and a second sub-loss function, the first sub-loss function being constructed through a cross-entropy loss function based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk label corresponding to the target merchant in the first business operation scenario, and the second sub-loss function being constructed through a cross-entropy loss function based on the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario, and the risk label corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario.

[0106] The embodiment of the present specification provides a risk identification device, which obtains business operation data related to the business operation behavior of a target merchant in a first business operation scenario, and then encodes and processes the business operation data to obtain data representations corresponding to the business operation data. After that, the data representations corresponding to the business operation data can be input into a pre-trained multi-task learning network to obtain risk representations of each task including the relationship between different tasks in multiple tasks for the target merchant. The multi-task learning network includes multiple expert models and gated networks corresponding to the tasks. The multiple tasks at least include risk identification services for target merchants in the first business operation scenario and risk identification services for target merchants converted from the first business operation scenario to the second business operation scenario. Finally, based on the risk representation of each task, the risk identification results for the target merchant can be determined by the pre-trained risk identification network. In this way, a multi-task learning network is used to learn for different life cycles of the target merchant, thereby greatly improving the risk identification efficiency of merchants in multiple stages of the life cycle, and at the same time greatly reducing the maintenance difficulty of the risk identification system. In addition, the subsequent merchant risk decision system relies on the accuracy of merchant risk prediction or risk identification. The improvement of merchant risk prediction accuracy can improve the merchant decision-making efficiency. Moreover, through multi-task learning, we can simultaneously learn: the prediction results of the preset risks that occur directly without conversion (that is, in the first business operation scenario); the prediction results of the preset risks that occur to merchants after conversion to the second business operation scenario, etc. Since the behavioral performance of merchants before the risk occurs in the first business operation scenario is often quite different from that in the second business operation scenario, the above method can greatly improve the model effect.

[0107] The above is a risk identification device provided in the embodiment of this specification. Based on the same idea, the embodiment of this specification also provides a risk identification device, such as Fig.10 shown.

[0108] The risk identification device may provide a terminal device or a server, etc. for the above-mentioned embodiment.

[0109] The risk identification device may have relatively large differences due to different configurations or performances, and may include one or more processors 1001 and memory 1002, and the memory 1002 may store one or more storage applications or data. Among them, the memory 1002 may be a short-term storage or a persistent storage. The application stored in the memory 1002 may include one or more modules (not shown in the figure), and each module may include a series of computer executable instructions in the risk identification device. Furthermore, the processor 1001 may be configured to communicate with the memory 1002 and execute a series of computer executable instructions in the memory 1002 on the risk identification device. The risk identification device may also include one or more power supplies 1003, one or more wired or wireless network interfaces 1004, one or more input and output interfaces 1005, and one or more keyboards 1006.

[0110] Specifically in this embodiment, the risk identification device includes a memory and one or more programs, wherein the one or more programs are stored in the memory, and the one or more programs may include one or more modules, and each module may include a series of computer executable instructions in the risk identification device, and the one or more programs are configured to be executed by one or more processors, including computer executable instructions for performing the following: Acquire business operation data related to business operation behavior of the target merchant in the first business operation scenario; Encoding the business operation data to obtain a data representation corresponding to the business operation data; Inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks among a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario; Based on the risk characterization of each task, a risk identification result for the target merchant is determined respectively through a pre-trained risk identification network.

[0111] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the risk identification device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0112] The embodiment of the present specification provides a risk identification device, which obtains business operation data related to the business operation behavior of a target merchant in a first business operation scenario, and then encodes and processes the business operation data to obtain data representations corresponding to the business operation data. After that, the data representations corresponding to the business operation data can be input into a pre-trained multi-task learning network to obtain risk representations of each task including the relationship between different tasks in multiple tasks for the target merchant. The multi-task learning network includes multiple expert models and gated networks corresponding to the tasks. The multiple tasks at least include risk identification services for target merchants in the first business operation scenario and risk identification services for target merchants converted from the first business operation scenario to the second business operation scenario. Finally, based on the risk representation of each task, the risk identification results for the target merchant can be determined by the pre-trained risk identification network. In this way, a multi-task learning network is used to learn for different life cycles of the target merchant, thereby greatly improving the risk identification efficiency of multi-stage merchants in the life cycle, and at the same time greatly reducing the maintenance difficulty of the risk identification system. In addition, the subsequent merchant risk decision system relies on the accuracy of merchant risk prediction or risk identification. The improvement of merchant risk prediction accuracy can improve the merchant decision-making efficiency. Moreover, through multi-task learning, we can simultaneously learn: the prediction results of the preset risks that occur directly without conversion (that is, in the first business operation scenario); the prediction results of the preset risks that occur to merchants after conversion to the second business operation scenario, etc. Since the behavioral performance of merchants before the risk occurs in the first business operation scenario is often quite different from that in the second business operation scenario, the above method can greatly improve the model effect.

[0113] Furthermore, based on the above Figures 1 to 8 In one embodiment, the present specification further provides a storage medium for storing computer executable instruction information. In a specific embodiment, the storage medium may be a USB flash drive, an optical disk, a hard disk, etc. When the computer executable instruction information stored in the storage medium is executed by the processor, the following process can be implemented: Acquire business operation data related to business operation behavior of the target merchant in the first business operation scenario; Encoding the business operation data to obtain a data representation corresponding to the business operation data; Inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks among a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario; Based on the risk characterization of each task, a risk identification result for the target merchant is determined respectively through a pre-trained risk identification network.

[0114] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above-mentioned storage medium embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0115] The embodiment of the present specification provides a storage medium, which obtains business operation data related to the business operation behavior of a target merchant in a first business operation scenario, and then encodes and processes the business operation data to obtain data representations corresponding to the business operation data. After that, the data representations corresponding to the business operation data can be input into a pre-trained multi-task learning network to obtain risk representations of each task including the relationship between different tasks in multiple tasks for the target merchant. The multi-task learning network includes multiple expert models and gated networks corresponding to the tasks. The multiple tasks at least include risk identification services for target merchants in the first business operation scenario and risk identification services for target merchants converted from the first business operation scenario to the second business operation scenario. Finally, based on the risk representation of each task, the risk identification results for the target merchant can be determined by the pre-trained risk identification network. In this way, a multi-task learning network is used to learn for different life cycles of the target merchant, thereby greatly improving the risk identification efficiency of merchants in multiple stages of the life cycle, and at the same time greatly reducing the maintenance difficulty of the risk identification system. In addition, the subsequent merchant risk decision system relies on the accuracy of merchant risk prediction or risk identification. The improvement of merchant risk prediction accuracy can improve the merchant decision-making efficiency. Moreover, through multi-task learning, we can simultaneously learn: the prediction results of the preset risks that occur directly without conversion (that is, in the first business operation scenario); the prediction results of the preset risks that occur to merchants after conversion to the second business operation scenario, etc. Since the behavioral performance of merchants before the risk occurs in the first business operation scenario is often quite different from that in the second business operation scenario, the above method can greatly improve the model effect.

[0116] Furthermore, based on the above Figures 1 to 8 In one or more embodiments of the present specification, a computer program product is provided, including a computer program. When the computer program in the computer program product is executed by a processor, the following process can be implemented: Acquire business operation data related to business operation behavior of the target merchant in the first business operation scenario; Encoding the business operation data to obtain a data representation corresponding to the business operation data; Inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks among a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario; Based on the risk characterization of each task, a risk identification result for the target merchant is determined respectively through a pre-trained risk identification network.

[0117] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the above-mentioned computer program product embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0118] The embodiment of the present specification provides a computer program product, which obtains business operation data related to the business operation behavior of a target merchant in a first business operation scenario, and then encodes and processes the business operation data to obtain data representations corresponding to the business operation data. After that, the data representations corresponding to the business operation data can be input into a pre-trained multi-task learning network to obtain risk representations of each task including the relationship between different tasks in multiple tasks for the target merchant. The multi-task learning network includes multiple expert models and gated networks corresponding to the tasks. The multiple tasks at least include risk identification services for target merchants in the first business operation scenario and risk identification services for target merchants converted from the first business operation scenario to the second business operation scenario. Finally, based on the risk representation of each task, the risk identification results for the target merchant can be determined by the pre-trained risk identification network. In this way, a multi-task learning network is used to learn for different life cycles of the target merchant, thereby greatly improving the risk identification efficiency of merchants in multiple stages of the life cycle, and at the same time greatly reducing the maintenance difficulty of the risk identification system. In addition, the subsequent merchant risk decision system relies on the accuracy of merchant risk prediction or risk identification. The improvement of merchant risk prediction accuracy can improve the merchant decision-making efficiency. Moreover, through multi-task learning, we can simultaneously learn: the prediction results of the preset risks that occur directly without conversion (that is, in the first business operation scenario); the prediction results of the preset risks that occur to merchants after conversion to the second business operation scenario, etc. Since the behavioral performance of merchants before the risk occurs in the first business operation scenario is often quite different from that in the second business operation scenario, the above method can greatly improve the model effect.

[0119] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0120] In the 1990s, it was very clear whether the improvement of a technology was hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, 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. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0121] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0122] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, 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.

[0123] For the convenience of description, the above devices are described in terms of functions and are divided into various units. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0124] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0125] The embodiments of this specification are described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable fraud case serial and parallel device to produce a machine, so that the instructions executed by the processor of the computer or other programmable fraud case serial and parallel device generate instructions for implementing the processes in the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable fraud case serial and parallel device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0127] These computer program instructions may also be loaded onto a computer or other programmable device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0128] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0129] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0130] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. 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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0131] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0132] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, one or more embodiments of this specification may be in the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Furthermore, one or more embodiments of this specification may be in the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0134] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0135] The above description is only an embodiment of this specification and is not intended to limit this document. For those skilled in the art, this specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification should be included in the scope of the claims of this specification.

Claims

1. A risk identification method, the method comprising: Acquire business operation data related to business operation behavior of the target merchant in the first business operation scenario; Encoding the business operation data to obtain a data representation corresponding to the business operation data; Inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks among a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario; Based on the risk characterization of each task, a risk identification result for the target merchant is determined respectively through a pre-trained risk identification network.

2. According to the method of claim 1, the first business operation scenario is an offline business operation scenario, and the second business operation scenario is an online business operation scenario, or the first business operation scenario is an online business operation scenario, and the second business operation scenario is an offline business operation scenario.

3. The method according to claim 1 or 2, wherein the risk characterization of each task is based on, and the risk identification result for the target merchant is determined by a pre-trained risk identification network, respectively, comprising: Inputting the risk characterization corresponding to the risk identification business of the target merchant in the first business operation scenario into the corresponding risk identification network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario; Based on the risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, determining the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario through the risk identification network corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario; Based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, the risk identification result for the target merchant is determined.

4. According to the method of claim 3, the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes a first sub-business of the business operation scenario conversion and a second sub-business of the target merchant of the converted business operation scenario, the risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes a first sub-characterization corresponding to the first sub-business and a second sub-characterization corresponding to the second sub-business, and the risk identification network corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario includes a first sub-network corresponding to the first sub-business and a second sub-network corresponding to the second sub-business, The risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario is based on the risk characterization corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, and determining the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario through the risk identification network corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, including: Inputting a first sub-representation corresponding to the first sub-business into a first network to obtain a conversion probability of the target merchant from the first business operation scenario to the second business operation scenario; Inputting the second sub-representation corresponding to the second sub-business into the second sub-network to obtain the risk occurrence probability corresponding to the risk identification business of the target merchant converted into the second business operation scenario; Based on the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario and the risk occurrence probability corresponding to the second sub-business of the target merchant converted to the second business operation scenario, the risk occurrence probability corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario is determined.

5. The method according to claim 1, wherein the business operation data comprises tabular data, and the data representation corresponding to the business operation data comprises data representation corresponding to a plurality of different tabular features contained in the tabular data, The step of inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including the relationship between different tasks in the multiple tasks for the target merchant includes: splicing data representations corresponding to a plurality of different table features contained in the business operation data to obtain a spliced ​​data representation; Inputting the concatenated data representations into each expert model in the multi-task learning network respectively to obtain output data of each expert model; The concatenated data representation and the output data of each expert model are input into the gating network corresponding to the task to obtain the weight assigned by the gating network to each expert model, and based on the output data of each expert model and the weight corresponding to each expert model, the risk representation of each task including the relationship between different tasks in the multiple tasks for the target merchant is determined.

6. The method according to claim 5, wherein the number of the gating networks is determined by the number of the plurality of tasks, and each task corresponds to a gating network, and the step of inputting the concatenated data representation and the output data of each expert model into the gating network corresponding to the task to obtain the weight assigned by the gating network to each expert model comprises: The concatenated data representation and the output data of each expert model are respectively input into the gating network corresponding to each task to obtain the weight assigned by each gating network to each expert model.

7. The method according to claim 1, further comprising: Acquire a business operation data sample related to a business operation behavior of a first merchant in a first business operation scenario for model training; Encoding the business operation data sample to obtain a sample data representation corresponding to the business operation data sample; Inputting the sample data representation corresponding to the business operation data sample into the multi-task learning network to obtain a risk representation sample of each task including a relationship between different tasks in a plurality of tasks for the first merchant, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to the second business operation scenario; Based on the risk characterization samples of each task, determining the risk identification result of the first merchant through the risk identification network respectively; Based on the risk identification result of the first merchant and a preset loss function, the multi-task learning network and the risk identification network are jointly trained to obtain a trained multi-task learning network and a trained risk identification network.

8. According to the method described in claim 7, the loss function includes a first sub-loss function and a second sub-loss function, the first sub-loss function is based on the risk occurrence probability corresponding to the risk identification business of the target merchant in the first business operation scenario and the risk label corresponding to the target merchant in the first business operation scenario, and is constructed through a cross-entropy loss function, the second sub-loss function is based on the conversion probability of the target merchant from the first business operation scenario to the second business operation scenario and the risk occurrence probability corresponding to the risk identification business of the target merchant converted to the second business operation scenario, and the risk label corresponding to the risk identification business of the target merchant converted from the first business operation scenario to the second business operation scenario, and is constructed through a cross-entropy loss function.

9. A risk identification device, comprising: A data acquisition module, which acquires business operation data related to the business operation behavior of the target merchant in the first business operation scenario; An encoding module, which performs encoding processing on the business operation data to obtain a data representation corresponding to the business operation data; a multi-task processing module, inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks among a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario; The risk identification module determines the risk identification result for the target merchant through a pre-trained risk identification network based on the risk characterization of each task.

10. A risk identification device, comprising: processor; as well as a memory arranged to store computer executable instructions which, when executed, cause the processor to: Acquire business operation data related to business operation behavior of the target merchant in the first business operation scenario; Encoding the business operation data to obtain a data representation corresponding to the business operation data; Inputting the data representation corresponding to the business operation data into a pre-trained multi-task learning network to obtain a risk representation of each task including a relationship between different tasks among a plurality of tasks for the target merchant, wherein the multi-task learning network includes a plurality of expert models and a gating network corresponding to the tasks, wherein the plurality of tasks at least include a risk identification business of a target merchant in a first business operation scenario and a risk identification business of a target merchant converted from the first business operation scenario to a second business operation scenario; Based on the risk characterization of each task, a risk identification result for the target merchant is determined respectively through a pre-trained risk identification network.

Images