CN118261420A

CN118261420A - Data processing method, device and equipment

Info

Publication number: CN118261420A
Application number: CN202410369935.0A
Authority: CN
Inventors: 韩少帅
Original assignee: Alipay Hangzhou Information Technology Co Ltd
Current assignee: Alipay Hangzhou Information Technology Co Ltd
Priority date: 2024-03-28
Filing date: 2024-03-28
Publication date: 2024-06-28

Abstract

The embodiment of the specification provides a data processing method, a device and equipment, wherein the method comprises the following steps: receiving target input data input by a target user aiming at a target service, and triggering and executing target multi-mode service data corresponding to the target service by the target user; determining target feedback data through a pre-trained risk interpretation model based on the target multi-mode service data and the target input data, wherein the risk interpretation model comprises a first module and a pre-trained second module, the pre-trained second module is obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, and parameters of the second module are in a frozen state in the training process of the risk interpretation model; and judging whether the target service is executed with risk or not based on the target feedback data, and outputting the target feedback data and the risk judgment result to the target user.

Description

Data processing method, device and equipment

Technical Field

The present disclosure relates to the field of data processing technologies, and in particular, to a data processing method, apparatus, and device.

Background

Along with the rapid development of the internet industry, the application of the artificial intelligent model is also becoming more and more popular, for example, in a wind control scene, the risk type of executing the service can be determined by performing risk analysis on the service to be executed through the artificial intelligent model, so that a risk management and control personnel can determine a corresponding risk control strategy to perform risk control according to the risk type of the service, and privacy data of a user is protected from being revealed.

However, since the risk management personnel determines the corresponding risk control policy only by executing the risk type of a certain service, which results in low accuracy and efficiency of risk detection, a solution capable of improving the detection efficiency and detection accuracy of the service risk detection to accurately perform risk control is needed.

Disclosure of Invention

The embodiment of the specification aims to provide a solution capable of improving the detection efficiency and the detection accuracy of business risk detection so as to accurately control risks.

In order to achieve the above technical solution, the embodiments of the present specification are implemented as follows:

in a first aspect, an embodiment of the present disclosure provides a data processing method, including: receiving target input data input by a target user aiming at a target service, and triggering the target user to execute target multi-mode service data corresponding to the target service, wherein the target input data is used for acquiring risk information for executing the target service, and the target multi-mode service data comprises service data of at least two modes; determining target feedback data based on the target multi-mode service data and the target input data through a pre-trained risk interpretation model, wherein the target feedback data comprises answer data for the target input data and interpretation data of the answer data, the risk interpretation model comprises a first module and a pre-trained second module, the pre-trained second module is obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data is larger than that of training sample data for training the risk interpretation model, and parameters of the second module are in a frozen state in the training process of the risk interpretation model; and judging whether the target service is executed with risk or not based on the target feedback data, and outputting the target feedback data and the risk judgment result to the target user.

In a second aspect, embodiments of the present disclosure provide a data processing apparatus, the apparatus comprising: the system comprises a data receiving module, a target user input module and a target service executing module, wherein the data receiving module is used for receiving target input data input by the target user aiming at target service and triggering and executing target multi-mode service data corresponding to the target service by the target user, the target input data is used for acquiring risk information for executing the target service, and the target multi-mode service data comprises service data of at least two modes; the first processing module is used for determining target feedback data through a pre-trained risk interpretation model based on the target multi-mode service data and the target input data, the target feedback data comprises answer data for the target input data and interpretation data of the answer data, the risk interpretation model comprises a first module and a pre-trained second module, the pre-trained second module is obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data is larger than that of training sample data for training the risk interpretation model, and parameters of the second module are in a frozen state in the training process of the risk interpretation model; and the risk judging module is used for judging whether the target service is executed with risk or not based on the target feedback data, and outputting the target feedback data and the risk judging result to the target user.

In a third aspect, embodiments of the present specification provide a data processing apparatus, the data processing apparatus comprising: a processor; and a memory arranged to store computer executable instructions that, when executed, cause the processor to: receiving target input data input by a target user aiming at a target service, and triggering the target user to execute target multi-mode service data corresponding to the target service, wherein the target input data is used for acquiring risk information for executing the target service, and the target multi-mode service data comprises service data of at least two modes; determining target feedback data based on the target multi-mode service data and the target input data through a pre-trained risk interpretation model, wherein the target feedback data comprises answer data for the target input data and interpretation data of the answer data, the risk interpretation model comprises a first module and a pre-trained second module, the pre-trained second module is obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data is larger than that of training sample data for training the risk interpretation model, and parameters of the second module are in a frozen state in the training process of the risk interpretation model; and judging whether the target service is executed with risk or not based on the target feedback data, and outputting the target feedback data and the risk judgment result to the target user.

In a fourth aspect, embodiments of the present description provide a storage medium for storing computer-executable instructions that, when executed, implement the following: receiving target input data input by a target user aiming at a target service, and triggering the target user to execute target multi-mode service data corresponding to the target service, wherein the target input data is used for acquiring risk information for executing the target service, and the target multi-mode service data comprises service data of at least two modes; determining target feedback data based on the target multi-mode service data and the target input data through a pre-trained risk interpretation model, wherein the target feedback data comprises answer data for the target input data and interpretation data of the answer data, the risk interpretation model comprises a first module and a pre-trained second module, the pre-trained second module is obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data is larger than that of training sample data for training the risk interpretation model, and parameters of the second module are in a frozen state in the training process of the risk interpretation model; and judging whether the target service is executed with risk or not based on the target feedback data, and outputting the target feedback data and the risk judgment result to the target user.

Drawings

In order to more clearly illustrate the embodiments of the present description or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described below, it being obvious that the drawings in the following description are only some of the embodiments described in the present description, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

FIG. 1 is a schematic diagram of a data processing system of the present specification;

FIG. 2A is a flow chart of an embodiment of a data processing method of the present disclosure;

FIG. 2B is a schematic diagram illustrating a data processing method according to the present disclosure;

FIG. 3 is a schematic diagram of a data acquisition according to the present disclosure;

FIG. 4 is a schematic diagram illustrating a processing procedure of another data processing method according to the present disclosure;

FIG. 5 is a schematic diagram of a training process of an image-text conversion model according to the present specification;

FIG. 6 is a schematic diagram of an application process of an image text conversion model according to the present specification;

FIG. 7 is a schematic diagram of a risk interpretation model according to the present disclosure;

FIG. 8 is a schematic diagram of a risk interpretation model according to another embodiment of the present disclosure;

FIG. 9 is a schematic diagram of a training process of a risk interpretation model of the present specification;

FIG. 10 is a schematic diagram of an embodiment of a data processing apparatus according to the present disclosure;

Fig. 11 is a schematic diagram of a structure of a data processing apparatus of the present specification.

Detailed Description

The embodiment of the specification provides a data processing method, a device and equipment.

In order to make the technical solutions in the present specification better understood by those skilled in the art, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only some embodiments of the present specification, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.

The technical scheme of the specification can be applied to a data processing system, as shown in fig. 1, the data processing system can be provided with terminal equipment and a server, wherein the server can be an independent server or a server cluster formed by a plurality of servers, and the terminal equipment can be equipment such as a personal computer and the like or mobile terminal equipment such as a mobile phone, a tablet personal computer and the like.

The data processing system may include n terminal devices and m servers, where n and m are positive integers greater than or equal to 1, the servers may be background servers of a certain application program, and the terminal devices may be client devices of the application program, where the application program may be an application program capable of providing services such as resource transfer service, video viewing service, and instant messaging service for a user.

The terminal equipment can collect input data input by a user aiming at a certain service, trigger and execute multi-mode service data corresponding to the service, and send the collected data to the server. The server may determine feedback data, i.e. determine answer data to the user input data, and interpretation data of the answer data, based on the user input data and the multimodal service data by means of a pre-trained risk interpretation model. In this way, the server can determine whether the service is at risk according to the feedback data, and send the feedback data and the risk judgment result to the terminal device, and the terminal device can output the feedback data and the risk judgment result to the user.

In addition, the server can also determine the data acquired by the terminal equipment as training sample data, so that the risk interpretation model is trained through the training sample data under the condition that a model training period is reached, and the trained risk interpretation model is obtained.

In addition, a central server (such as the server 1) may be further disposed in the data processing system, and the central server may receive the historical data (i.e., the historical multi-mode service data, the user input data, the corresponding historical feedback data, etc.) stored in the terminal device and/or the server, and determine the historical data as training sample data, so as to train the risk interpretation model based on the training sample data, and obtain a trained risk interpretation model. In this way, the central server can send the model parameters of the trained risk interpretation model to other servers in the data processing system, and the other servers can update the local risk interpretation model according to the received model parameters to obtain the trained risk interpretation model, and further perform risk detection processing and the like through the trained risk interpretation model. The occurrence of conditions such as service interruption and the like caused by the need of training a risk interpretation model is avoided, and the service use requirement of a user is met.

The data processing method in the following embodiments can be implemented based on the above-described data processing system configuration.

Example 1

As shown in fig. 2A and fig. 2B, the embodiment of the present disclosure provides a data processing method, where an execution body of the method may be a server, where the server may be an independent server or may be a server cluster formed by a plurality of servers. The method specifically comprises the following steps:

in S202, target input data input by a target user for a target service is received, and the target user triggers execution of target multi-mode service data corresponding to the target service.

The target input data may be used to obtain risk information of executing the target service, the target multi-mode service data may include service data of at least two modes, the target service may be any service that may involve data risk problems such as user privacy data disclosure, for example, the target service may be a resource transfer service, an identity verification service, etc., the target input data may be risk query information of the target user for executing the target service, the target input data may be one or more of text data, audio data and video data input by the target user, for example, the target input data may be input by the target user: "is there a risk in executing the target service? "," what specific part is there a violation? "," is there a tampered image? The target multi-mode service data may be service data of at least two modes corresponding to the trigger execution target service, specifically, for example, the target multi-mode service data may include text data (such as resource transfer number, resource transfer time, etc.), service data of at least two modes in service data including image data, audio data, video data, etc. included in a resource transfer page, or for example, the target service may include text data (such as identity information input by a target user), collected service data of at least two modes in service data including image data, video data, audio data, etc. of the target user.

In implementation, with rapid development of the internet industry, application of an artificial intelligent model is also becoming more and more popular, for example, in a wind control scene, risk analysis can be performed on a service to be executed through the artificial intelligent model to determine a risk type of executing the service, so that a risk management and control personnel can determine a corresponding risk control strategy to perform risk control according to the risk type of the service to protect private data of a user from being revealed and the like. However, since the risk management personnel determines the corresponding risk control policy only by executing the risk type of a certain service, which results in low accuracy and efficiency of risk detection, a solution capable of improving the detection efficiency and detection accuracy of the service risk detection to accurately perform risk control is needed. For this reason, the embodiments of the present specification provide a technical solution that can solve the above-mentioned problems, and specifically, reference may be made to the following.

Taking the target service as a resource transfer service as an example, the target user can trigger and start the resource transfer service through a resource transfer application program installed in the terminal device, that is, the terminal device can acquire target input data input by the target user for the resource transfer service under the condition that the terminal device detects that the target user triggers and starts the resource transfer service through a certain resource transfer application program, and the target user triggers and executes the target multi-mode service number corresponding to the resource transfer service.

For example, as shown in fig. 3, the target user may input information such as the number of resource transfers and the resource transfer object in the resource transfer page, the terminal device may collect text data such as the number of resource transfers and the resource transfer object, and the resource transfer time, and the picture data, the video data, and the audio data included in the resource transfer page, and the terminal device may send the collected service data as target multi-mode service data to the server. In addition, the target user may also input risk inquiry information for triggering execution of the resource transfer service on the resource transfer page, such as "is there a risk in executing the target service? Whether the picture in the page is a tampered picture or not, the terminal device may send the received risk inquiry information to the server as target input data.

Thus, the server can receive the target input data and the target multi-mode service data sent by the terminal equipment.

In addition, the method for determining the target input data and the target multi-mode service data is an optional and implementable determination method, and in an actual application scenario, there may be a plurality of different determination methods, and may be different according to the actual application scenario, which is not specifically limited in the embodiment of the present disclosure.

In S204, based on the target multimodal service data and the target input data, target feedback data is determined by a pre-trained risk interpretation model.

The target feedback data may include answer data to the target input data, and interpretation data of the answer data, for example, "is there a risk of executing the target service? For example, if the picture in the page is a tampered picture, the target feedback data may include reply data to the target input data, such as "there is a risk of executing the target service," the picture is a tampered picture, "and interpretation data to the reply data, such as" there is a risk of privacy disclosure due to a certain area in the picture being a tampered condition, and the area being an important detection area, so that executing the target service may be performed. The risk interpretation model may include a first module and a pre-trained second module, the pre-trained second module may be obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data amount of the historical sample data may be larger than the data amount of the training sample data for training the risk interpretation model, parameters of the second module are in a frozen state in the training process of the risk interpretation model, the first module may be constructed by the deep learning algorithm, that is, the first module and the second module may be constructed based on the same deep learning algorithm or may be constructed based on different deep learning algorithms, and in addition, a model structure of the first module may be smaller than a model structure of the second module, for example, constructed by the first module and the second module based on a plurality of convolution layers, the number of convolution layers contained in the first module may be smaller than the number of convolution layers contained in the second module, and the second module may learn a probability distribution of the text data through the deep learning algorithm, so as to predict a next word, phrase or sentence.

In an implementation, the server may construct training sample data based on the historical data, so as to train the risk interpretation model through the training sample data, so as to obtain a trained risk interpretation model, where, because the second module is obtained through pre-training, and the data size of the historical sample data used for training the second module is larger than the data size of the training sample data used for training the risk interpretation model, in the process of training the risk interpretation model, the parameters of the second module may be frozen, and only the parameters of the first module may be updated, that is, the parameters of the second module may be in a frozen state in the process of training the risk interpretation model. After obtaining the trained risk interpretation model, the server can determine target feedback data through the pre-trained risk interpretation model based on target multi-mode service data and target input data.

In S206, based on the target feedback data, it is determined whether there is a risk in executing the target service, and the target feedback data and the risk determination result are output to the target user.

In implementation, since the target input data is data for acquiring risk information for executing the target service, and the target feedback data includes answer data to the target input data and interpretation data of the answer data, whether there is a risk for executing the target service can be determined through the target input data and the target feedback data, so as to obtain a risk determination result.

For example, "is there a risk in executing the target service? Whether the picture in the page is a tampered picture, the target feedback data may include answer data to the target input data, such as "there is a risk of executing the target service, the picture is a tampered picture", and interpretation data to the answer data, such as "there is a risk of privacy disclosure due to a certain area in the picture being a tamper condition and the area being an important detection area", for example. Based on the target input data and target feedback data, it may be determined that there is a risk of executing the target service.

Or the server can also perform risk detection processing on the target feedback data, the target input data and the target multi-mode service data based on a pre-trained risk type determining model to obtain a corresponding risk type, and determine whether the target service is executed with risk based on the risk type. The risk type determining model may be a model which is constructed based on a preset machine learning algorithm and is used for determining a risk type, the risk type may include a high risk type, a medium risk type, a low risk type and the like, and the risk type may be determined based on a service type corresponding to the target service.

In addition, the method for determining whether the risk exists in the execution target service is an optional and realizable determination method, and in the actual application scenario, there may be a plurality of different determination methods, and may be different according to the actual application scenario, which is not specifically limited in the embodiment of the present disclosure.

The server may send the target feedback data and the risk determination result to the terminal device, and the terminal device may output the target feedback data and the risk determination result to the target user.

The embodiment of the specification provides a data processing method, by receiving target input data input by a target user for target service and target multi-mode service data corresponding to the target service, wherein the target input data is used for acquiring risk information for executing the target service, the target multi-mode service data comprises at least two modes of service data, based on the target multi-mode service data and the target input data, target feedback data is determined through a pre-trained risk interpretation model, the target feedback data can comprise answer data for the target input data and interpretation data of the answer data, the risk interpretation model can comprise a first module and a pre-trained second module, the pre-trained second module can be obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data can be larger than the data volume of training sample data for training the risk interpretation model, parameters of the second module are in a frozen state in the training process of the risk interpretation model, whether the target service is executed or not is judged based on the target feedback data, and a risk judgment result is output to the target user. In this way, in the training process of the risk interpretation model, the parameters of the second module are in a frozen state, so that the server only needs to train the parameters of the first module, and the second module is obtained by training historical sample data with larger data volume, so that the detection accuracy of the risk interpretation model obtained by training can be ensured while the training efficiency of the risk interpretation model is improved. In addition, the risk interpretation model can generate corresponding answer data and interpretation data (namely feedback data) of the answer data based on multi-mode service data of at least two modes and input data of a user, so that the problem of poor risk detection accuracy caused by risk detection based on service data of a single mode can be avoided, whether the risk exists in an execution target service can be accurately judged through the interpretation data of the answer data, namely, the detection efficiency and the detection accuracy of the service risk detection can be improved, and the risk control can be accurately performed.

Example two

As shown in fig. 4, the embodiment of the present disclosure provides a data processing method, where an execution body of the method may be a server, where the server may be an independent server or may be a server cluster formed by a plurality of servers. The method specifically comprises the following steps:

In S402, training sample data for a target service is acquired.

The training sample data may include historical multi-mode service data corresponding to the target service, user input data for acquiring risk information of the target service, and historical feedback data of the input data.

In implementation, the server may obtain historical data corresponding to the target service (such as historical multi-modal service data corresponding to the target service of three months, user input data for obtaining risk information of the target service, and historical feedback data of the input data) from a local database or a database of other servers in the data processing system, and construct training sample data according to the historical data.

In S404, predictive feedback data for the input data is determined by a risk interpretation model based on the historical multimodal service data and the user input data.

In practice, the above-mentioned processing manner of S402 may be varied, and the following provides an alternative implementation manner, which can be seen from the following steps one to three:

and step one, performing text conversion processing on the historical multi-mode service data to obtain historical text service data corresponding to the historical multi-mode service data.

The multi-modal service data may include image data, video data, and the like.

In an implementation, since the second module may not be capable of directly processing non-text data, text conversion processing may be performed on multi-mode service data, for example, using image data (or video data) as an example, and the server may perform text conversion processing on the image data to obtain corresponding text service data. For example, the server may perform text conversion processing on the image data in the historical multi-mode service data through the following steps A1 to A8 to obtain corresponding historical text service data.

And step A1, acquiring historical image data and historical text data.

And step A2, carrying out feature extraction processing on the historical image data based on an image feature extraction module in the image text conversion model to obtain a third feature vector corresponding to the historical image data.

And step A3, based on a text feature extraction module in the image text conversion model, performing text feature extraction processing on the historical text data to obtain a fourth feature vector corresponding to the historical text data.

And A4, acquiring a distance value between the third feature vector and the fourth feature vector.

And step A5, determining a loss value based on a preset loss function and a distance value, determining whether the image text conversion model is converged based on the loss value, and if the image text conversion model is determined not to be converged based on the loss value, continuing training the image text conversion model based on the historical image data and the historical text data until the image text conversion model is converged, so as to obtain the trained image text conversion model.

IN implementation, as shown IN fig. 5, assuming that there are N pieces of history image data and N pieces of history text data IN total, feature extraction processing may be performed on the history image data by the image feature extraction module IN the image text conversion model, so as to obtain third feature vectors (i.e., I1, I2, I3,..and IN) corresponding to the history image data. The feature extraction processing may be performed on the historical text data by using a text feature extraction module in the image text conversion model, so as to obtain fourth feature vectors (i.e., T1, T2, T3,..and TN) corresponding to the historical text data.

The server may then obtain the distance values (i.e., I1 x T1, I1 x T2,..once., IN x TN) between the third feature vector and the fourth feature vector, further, based on the preset loss function and the distance value, determining a loss value, and based on the loss value, determining whether the image text conversion model converges, if the image text conversion model is determined not to be converged based on the loss value, training the image text conversion model based on the historical image data and the historical text data until the image text conversion model is converged, and obtaining a trained image text conversion model.

In this way, the server can align the image data and the text data through the trained image text conversion model, that is, the output of the image feature extraction module and the output of the text feature extraction module in the image text conversion model have fixed mapping.

And step A6, carrying out feature extraction processing on the image data in the historical multi-mode service data based on an image feature extraction module in a pre-trained image text conversion model to obtain a first feature vector corresponding to the image data.

And A7, acquiring candidate text data corresponding to the target service, and carrying out text feature extraction processing on the candidate text data based on a text feature extraction module in a pre-trained image text conversion model to obtain a second feature vector corresponding to the candidate text data.

And step A8, selecting target text data corresponding to the image data from the candidate text data based on the first feature vector and the second feature vector, and determining the target text data as text service data corresponding to the image data in the historical multi-mode service data.

In implementation, as shown in fig. 6, assuming that there are M candidate text data in total, the server may perform text feature extraction processing on the M candidate text data by using a text feature extraction module in the trained image text conversion model, and may obtain M second feature vectors (i.e., A1, A2, A3, AN). And carrying out image feature extraction processing on the image data in the historical multi-mode service data by an image feature extraction module in the image text conversion model obtained through training to obtain a first feature vector (namely B1).

The server may acquire a distance value between the first feature vector and each of the second feature vectors, select target text data (e.g., candidate text data corresponding to A3) corresponding to the image data from the candidate text data based on the distance value, and determine the target text data as text service data corresponding to the image data in the historical multimodal service data.

In addition, audio data can be also included in the historical multi-mode service data, the original data of the audio data is a waveform signal, and as the voice and the text have a one-to-one correspondence, the server can directly perform voice recognition processing on the audio data to obtain the corresponding text service data.

The method for voice recognition processing may be multiple, for example, the server may perform voice recognition processing on the streaming audio data through a voice recognition framework of a transducer-transducer to obtain corresponding text service data, and in addition, there may be multiple different voice recognition processing methods, and different voice recognition methods may be selected according to different practical application scenarios, which is not limited in this embodiment of the present disclosure.

And step two, inputting the historical text service data into a linear projection layer of the risk interpretation model to obtain a target feature vector of the historical text service data.

And thirdly, inputting the target feature vector and user input data into a risk interpretation model to obtain predictive feedback data aiming at the input data.

In implementation, the first module and the second module in the risk interpretation model may be configured in various manners, and the following two alternative implementations are provided, and specifically, the following steps B1 to B3, or the steps C1 to C2 may be referred to as processing:

And B1, inputting the target feature vector and user input data into a first module of the risk interpretation model to obtain a first sub-result.

And step B2, inputting the target feature vector and the user input data into a second module trained in advance in the risk interpretation model to obtain a second sub-result.

And step B3, determining prediction feedback data for the input data based on the first sub-result and the second sub-result.

The first module may include a first sub-module and a second sub-module, the first sub-module may be configured to process the target feature vector and the user input data, the second sub-module may be configured to process an output result of the first sub-module to obtain the first sub-result, an output dimension of the first sub-module is the same as an input dimension of the second sub-module, an output dimension of the second sub-module is the same as an output dimension of the pre-trained second module, and an input dimension of the second sub-module may be smaller than the output dimension of the second sub-module.

In implementation, as shown in fig. 7, the first module includes a first sub-module and a second sub-module that have different input dimensions and different output dimensions, and the output dimension of the first sub-module is smaller than the input dimension, so that parameters required to be trained in the risk interpretation model can be reduced by a low-rank adaptation (low-rank adaptation) manner, and the training efficiency of the risk interpretation model is improved.

And step C1, inputting the target feature vector and the user input data into a first module of the risk interpretation model to obtain an intermediate result.

And step C2, inputting the intermediate result into a second module trained in advance in the risk interpretation model to obtain predictive feedback data aiming at the input data.

The number of layers of the neural network contained in the first module may be smaller than the number of layers of the neural network contained in the second module.

In an implementation, as shown in fig. 8, the first module may include a neural network structure of 1024×8, and the second module may include a neural network structure of 1024×1024, so that the server may process the target feature vector and the user input data through the first module to obtain an intermediate result, and then input the intermediate result to the second module to obtain the prediction feedback data.

In addition, the above model structure is an optional and realizable structure, and in addition, there may be a plurality of different model structures, for example, the first module may also be inside the second module, specifically, for example, the 1024 x 8 neural network structure included in the first module may be a 1024 x 1024 neural network structure included in the second module, and the first module may be included in any position of the second module, or the first module may also be a next data processing module of the second module, that is, the server may input the target feature vector and the user input data into the pre-trained second module of the risk interpretation model, to obtain an intermediate result, input the intermediate result into the first module in the risk interpretation model, to obtain the predicted feedback data for the input data, and so on, and different model structures may be selected for the risk interpretation model according to the actual application scenario, which is not limited in this embodiment of the present disclosure.

In S406, based on the historical feedback data and the predictive feedback data, it is determined whether the risk interpretation model converges, and if the risk interpretation model does not converge, the parameters of the first module are updated based on the historical feedback data and the predictive feedback data, so as to obtain an updated risk interpretation model.

In implementation, as shown in fig. 9, when the server determines that the risk interpretation model does not converge based on the historical feedback data and the predictive feedback data, the server may update the parameters of the first module based on the historical feedback data and the predictive feedback data to obtain an updated risk interpretation model.

In S408, training the updated risk interpretation model is continued based on the historical multimodal service data, the user input data, and the historical feedback data until the risk interpretation model converges, to obtain a trained risk interpretation model.

In practice, on the basis of an already trained model (i.e. a second module which is pre-trained), a new data set (i.e. a training sample data set) is used for retraining, i.e. when the risk interpretation model is trained, only the parameters of the first module can be updated in a supervised fine tuning manner, so that the obtained trained risk interpretation model can adapt to specific requirements of a new task or field.

The target input data may be used to obtain risk information for executing the target service, and the target multi-mode service data may include service data of at least two modes.

Example III

The data processing method provided in the embodiment of the present disclosure is based on the same concept, and the embodiment of the present disclosure further provides a data processing device, as shown in fig. 10.

The data processing apparatus includes: a data receiving module 1001, a first processing module 1002 and a wind judging module 1003, wherein:

The data receiving module 1001 is configured to receive target input data input by a target user for a target service, and target multi-mode service data corresponding to the target service triggered by the target user, where the target input data is used to obtain risk information for executing the target service, and the target multi-mode service data includes service data of at least two modes;

a first processing module 1002, configured to determine target feedback data based on the target multimodal service data and the target input data through a pre-trained risk interpretation model, where the target feedback data includes answer data for the target input data, and interpretation data of the answer data, and the risk interpretation model includes a first module and a pre-trained second module, where the pre-trained second module is obtained by training, through historical sample data, a module for generating text data constructed by a deep learning algorithm, where a data amount of the historical sample data is greater than a data amount of training sample data for training the risk interpretation model, and parameters of the second module are in a frozen state during a training process of the risk interpretation model;

And a risk judging module 1003, configured to judge whether the target service is executed with risk based on the target feedback data, and output the target feedback data and a risk judging result to the target user.

In an embodiment of the present disclosure, the apparatus further includes:

the data acquisition module is used for acquiring training sample data aiming at the target service, wherein the training sample data comprises historical multi-mode service data corresponding to the target service, user input data used for acquiring risk information of the target service and historical feedback data of the input data;

The second processing module is used for determining prediction feedback data aiming at the input data through the risk interpretation model based on the historical multi-mode service data and the user input data;

The parameter updating module is used for determining whether the risk interpretation model is converged based on the historical feedback data and the prediction feedback data, and updating the parameters of the first module based on the historical feedback data and the prediction feedback data if the risk interpretation model is not converged to obtain an updated risk interpretation model;

And the first training module is used for continuing to train the updated risk interpretation model based on the historical multi-mode service data, the user input data and the historical feedback data until the risk interpretation model converges to obtain a trained risk interpretation model.

In an embodiment of the present disclosure, the second processing module is configured to:

Performing text conversion processing on the historical multi-mode service data to obtain historical text service data corresponding to the historical multi-mode service data;

Inputting the historical text service data into a linear projection layer of the risk interpretation model to obtain a target feature vector of the historical text service data;

And inputting the target feature vector and the user input data into the risk interpretation model to obtain predictive feedback data aiming at the input data.

In this embodiment of the present disclosure, the multi-modal service data includes image data, and the second processing module is configured to:

Based on an image feature extraction module in a pre-trained image text conversion model, carrying out feature extraction processing on image data in the historical multi-mode service data to obtain a first feature vector corresponding to the image data;

acquiring candidate text data corresponding to the target service, and carrying out text feature extraction processing on the candidate text data based on a text feature extraction module in the pre-trained image text conversion model to obtain a second feature vector corresponding to the candidate text data;

And selecting target text data corresponding to the image data from the candidate text data based on the first feature vector and the second feature vector, and determining the target text data as the text service data corresponding to the image data in the historical multi-mode service data.

In an embodiment of the present disclosure, the apparatus further includes:

the first acquisition module is used for acquiring historical image data and historical text data;

The first extraction module is used for carrying out feature extraction processing on the historical image data based on the image feature extraction module in the image text conversion model to obtain a third feature vector corresponding to the historical image data;

the second extraction module is used for carrying out text feature extraction processing on the historical text data based on the text feature extraction module in the image text conversion model to obtain a fourth feature vector corresponding to the historical text data;

the distance determining module is used for obtaining a distance value between the third characteristic vector and the fourth characteristic vector;

And the second training module is used for determining a loss value based on a preset loss function and the distance value, determining whether the image text conversion model is converged based on the loss value, and if the image text conversion model is determined not to be converged based on the loss value, continuing training the image text conversion model based on the historical image data and the historical text data until the image text conversion model is converged, so as to obtain the trained image text conversion model.

Inputting the target feature vector and the user input data into a first module of the risk interpretation model to obtain a first sub-result;

inputting the target feature vector and the user input data to a second module trained in advance in the risk interpretation model to obtain a second sub-result;

based on the first sub-result and the second sub-result, the predictive feedback data for the input data is determined.

In this embodiment of the present disclosure, the first module includes a first sub-module and a second sub-module, where the first sub-module is configured to process the target feature vector and the user input data, and the second sub-module is configured to process an output result of the first sub-module to obtain the first sub-result, an output dimension of the first sub-module is the same as an input dimension of the second sub-module, an output dimension of the second sub-module is the same as an output dimension of the pre-trained second module, and an input dimension of the second sub-module is smaller than the output dimension of the second sub-module.

inputting the target feature vector and the user input data into a first module of the risk interpretation model to obtain an intermediate result;

and inputting the intermediate result to a second module which is trained in advance in the risk interpretation model to obtain the predictive feedback data aiming at the input data, wherein the number of layers of the neural network contained in the first module is smaller than that of the neural network contained in the second module.

The embodiment of the specification provides a data processing device, by receiving target input data input by a target user for target service and target multi-mode service data corresponding to the target service, wherein the target input data is used for acquiring risk information for executing the target service, the target multi-mode service data comprises at least two modes of service data, based on the target multi-mode service data and the target input data, target feedback data is determined through a pre-trained risk interpretation model, the target feedback data can comprise answer data for the target input data and interpretation data of the answer data, the risk interpretation model can comprise a first module and a pre-trained second module, the pre-trained second module can be obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data can be larger than the data volume of training sample data for training the risk interpretation model, parameters of the second module are in a frozen state in the training process of the risk interpretation model, whether the target service is executed or not is judged based on the target feedback data, and a risk judgment result is output to the target user. In this way, in the training process of the risk interpretation model, the parameters of the second module are in a frozen state, so that the server only needs to train the parameters of the first module, and the second module is obtained by training historical sample data with larger data volume, so that the detection accuracy of the risk interpretation model obtained by training can be ensured while the training efficiency of the risk interpretation model is improved. In addition, the risk interpretation model can generate corresponding answer data and interpretation data (namely feedback data) of the answer data based on multi-mode service data of at least two modes and input data of a user, so that the problem of poor risk detection accuracy caused by risk detection based on service data of a single mode can be avoided, whether the risk exists in an execution target service can be accurately judged through the interpretation data of the answer data, namely, the detection efficiency and the detection accuracy of the service risk detection can be improved, and the risk control can be accurately performed.

Example IV

Based on the same idea, the embodiment of the present disclosure further provides a data processing apparatus, as shown in fig. 11.

The data processing apparatus may vary considerably in configuration or performance and may include one or more processors 1101 and memory 1102, with one or more stored applications or data stored in memory 1102. Wherein the memory 1102 may be transient storage or persistent storage. The application programs stored in memory 1102 may include one or more modules (not shown) each of which may include a series of computer executable instructions for use in a data processing apparatus. Still further, the processor 1101 may be arranged to communicate with a memory 1102, a series of computer executable instructions in the memory 1102 being executed on a data processing device. The data processing device can also include one or more power supplies 1103, one or more wired or wireless network interfaces 1104, one or more input output interfaces 1105, one or more keyboards 1106.

In particular, in this embodiment, the data processing apparatus 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 for the data processing apparatus, and the one or more programs configured to be executed by the one or more processors comprise instructions for:

Receiving target input data input by a target user aiming at a target service, and triggering the target user to execute target multi-mode service data corresponding to the target service, wherein the target input data is used for acquiring risk information for executing the target service, and the target multi-mode service data comprises service data of at least two modes;

Determining target feedback data based on the target multi-mode service data and the target input data through a pre-trained risk interpretation model, wherein the target feedback data comprises answer data for the target input data and interpretation data of the answer data, the risk interpretation model comprises a first module and a pre-trained second module, the pre-trained second module is obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data is larger than that of training sample data for training the risk interpretation model, and parameters of the second module are in a frozen state in the training process of the risk interpretation model;

And judging whether the target service is executed with risk or not based on the target feedback data, and outputting the target feedback data and the risk judgment result to the target user.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for data processing apparatus embodiments, the description is relatively simple, as it is substantially similar to method embodiments, with reference to the description of method embodiments in part.

Example five

The embodiments of the present disclosure further provide a computer readable storage medium, where a computer program is stored, where the computer program when executed by a processor implements each process of the embodiments of the data processing method, and the same technical effects can be achieved, and for avoiding repetition, a detailed description is omitted herein. The computer readable storage medium is, for example, a Read-Only Memory (ROM), a random access Memory (Random Access Memory RAM), a magnetic disk or an optical disk.

The embodiment of the specification provides a computer readable storage medium, by receiving target input data input by a target user for target service and triggering target multi-mode service data corresponding to execution target service by the target user, wherein the target input data is used for acquiring risk information of execution target service, the target multi-mode service data comprises at least two modes of service data, based on the target multi-mode service data and the target input data, target feedback data is determined through a pre-trained risk interpretation model, the target feedback data can comprise answer data to the target input data and interpretation data of the answer data, the risk interpretation model can comprise a first module and a pre-trained second module, the pre-trained second module can be obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data can be larger than the data volume of training sample data for training the risk interpretation model, parameters of the second module are in a frozen state in the training process of the risk interpretation model, whether the execution target service is at risk exists or not is judged based on the target feedback data, and a judgment result is output to the target user. In this way, in the training process of the risk interpretation model, the parameters of the second module are in a frozen state, so that the server only needs to train the parameters of the first module, and the second module is obtained by training historical sample data with larger data volume, so that the detection accuracy of the risk interpretation model obtained by training can be ensured while the training efficiency of the risk interpretation model is improved. In addition, the risk interpretation model can generate corresponding answer data and interpretation data (namely feedback data) of the answer data based on multi-mode service data of at least two modes and input data of a user, so that the problem of poor risk detection accuracy caused by risk detection based on service data of a single mode can be avoided, whether the risk exists in an execution target service can be accurately judged through the interpretation data of the answer data, namely, the detection efficiency and the detection accuracy of the service risk detection can be improved, and the risk control can be accurately performed.

The foregoing describes specific embodiments of the present disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

In the 90 s of the 20 th century, improvements to one technology could clearly be distinguished as improvements in hardware (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software (improvements to the process flow). However, with the development of technology, many improvements of the current method flows can be regarded as direct improvements of hardware circuit structures. Designers almost always obtain corresponding hardware circuit structures by programming improved method flows into hardware circuits. Therefore, an improvement of a method flow cannot be said to be realized by a hardware entity module. For example, a programmable logic device (Programmable Logic Device, PLD) (e.g., field programmable gate array (Field Programmable GATE ARRAY, FPGA)) is an integrated circuit whose logic functions are determined by user programming of the device. A designer programs to "integrate" a digital system onto a PLD without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, such programming is mostly implemented with "logic compiler (logic compiler)" software, which is similar to the software compiler used in program development and writing, and the original code before being compiled is also written in a specific programming language, which is called hardware description language (Hardware Description Language, HDL), but HDL is not just one, but a plurality of 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), and VHDL (Very-High-SPEED INTEGRATED Circuit Hardware Description Language) and Verilog are currently most commonly used. It will also be apparent to those skilled in the art that a hardware circuit implementing the logic method flow can be readily obtained by merely slightly programming the method flow into an integrated circuit using several of the hardware description languages described above.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, application SPECIFIC INTEGRATED Circuits (ASICs), programmable logic controllers, and embedded microcontrollers, examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in a pure computer readable program code, it is well possible to implement the same functionality by logically programming the method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Such a controller may thus be regarded as a kind of hardware component, and means for performing various functions included therein may also be regarded as structures within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.

The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. One typical implementation is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, 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.

For convenience of description, the above devices are described as being functionally divided into various units, respectively. Of course, the functionality of the units may be implemented in one or more software and/or hardware when implementing one or more embodiments of the present description.

It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, one or more embodiments of the present description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Embodiments of the present description are described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the specification. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

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

The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer 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 disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.

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

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, as relevant to see a section of the description of method embodiments.

The foregoing is merely exemplary of the present disclosure and is not intended to limit the disclosure. Various modifications and alterations to this specification will become apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, or the like, which are within the spirit and principles of the present description, are intended to be included within the scope of the claims of the present description.

Claims

1. A data processing method, comprising:

2. The method of claim 1, further comprising, prior to the determining, by a pre-trained risk interpretation model, target feedback data for answering the target input data based on the target multimodal service data and the target input data:

Acquiring training sample data aiming at the target service, wherein the training sample data comprises historical multi-mode service data corresponding to the target service, user input data for acquiring risk information of the target service and historical feedback data of the input data;

determining predictive feedback data for the input data by the risk interpretation model based on the historical multimodal service data and the user input data;

Determining whether the risk interpretation model converges based on the historical feedback data and the predictive feedback data, and if the risk interpretation model does not converge, updating parameters of the first module based on the historical feedback data and the predictive feedback data to obtain an updated risk interpretation model;

And training the updated risk interpretation model continuously based on the historical multi-mode service data, the user input data and the historical feedback data until the risk interpretation model converges to obtain a trained risk interpretation model.

3. The method of claim 2, the determining, based on the historical multimodal traffic data and the user input data, predictive feedback data for the input data by the risk interpretation model, comprising:

4. The method of claim 3, wherein the multi-modal service data includes image data, the text conversion processing is performed on the historical multi-modal service data to obtain text service data corresponding to the historical multi-modal service data, and the method includes:

5. The method according to claim 4, further comprising, before the feature extraction module in the pre-trained image text conversion model performs feature extraction processing on the image data in the historical multimodal service data to obtain a first feature vector corresponding to the image data:

Acquiring historical image data and historical text data;

Based on an image feature extraction module in the image text conversion model, carrying out feature extraction processing on the historical image data to obtain a third feature vector corresponding to the historical image data;

Based on a text feature extraction module in the image text conversion model, performing text feature extraction processing on the historical text data to obtain a fourth feature vector corresponding to the historical text data;

Acquiring a distance value between the third feature vector and the fourth feature vector;

And determining a loss value based on a preset loss function and the distance value, determining whether the image text conversion model is converged based on the loss value, and if the image text conversion model is not converged based on the loss value, continuing training the image text conversion model based on the historical image data and the historical text data until the image text conversion model is converged, so as to obtain the trained image text conversion model.

6. A method according to claim 3, said inputting said target feature vector and said user input data into said risk interpretation model resulting in predictive feedback data for said input data, comprising:

7. The method of claim 6, the first module comprising a first sub-module for processing the target feature vector and the user input data and a second sub-module for processing an output result of the first sub-module to obtain the first sub-result, an output dimension of the first sub-module being the same as an input dimension of the second sub-module, an output dimension of the second sub-module being the same as an output dimension of the pre-trained second module, the input dimension of the second sub-module being less than the output dimension of the second sub-module.

8. A method according to claim 3, said inputting said target feature vector and said user input data into said risk interpretation model resulting in predictive feedback data for said input data, comprising:

9. A data processing apparatus comprising:

The system comprises a data receiving module, a target user input module and a target service executing module, wherein the data receiving module is used for receiving target input data input by the target user aiming at target service and triggering and executing target multi-mode service data corresponding to the target service by the target user, the target input data is used for acquiring risk information for executing the target service, and the target multi-mode service data comprises service data of at least two modes;

The first processing module is used for determining target feedback data through a pre-trained risk interpretation model based on the target multi-mode service data and the target input data, the target feedback data comprises answer data for the target input data and interpretation data of the answer data, the risk interpretation model comprises a first module and a pre-trained second module, the pre-trained second module is obtained by training a module for generating text data constructed by a deep learning algorithm through historical sample data, the data volume of the historical sample data is larger than that of training sample data for training the risk interpretation model, and parameters of the second module are in a frozen state in the training process of the risk interpretation model;

and the risk judging module is used for judging whether the target service is executed with risk or not based on the target feedback data, and outputting the target feedback data and the risk judging result to the target user.

10. A data processing apparatus, the data processing apparatus comprising:

A processor; and

A memory arranged to store computer executable instructions that, when executed, cause the processor to: