WO2021212377A1

WO2021212377A1 - Method and apparatus for determining risky attribute of user data, and electronic device

Info

Publication number: WO2021212377A1
Application number: PCT/CN2020/086194
Authority: WO
Inventors: 李森林
Original assignee: 深圳市欢太数字科技有限公司
Priority date: 2020-04-22
Filing date: 2020-04-22
Publication date: 2021-10-28
Also published as: CN115066699A

Abstract

Disclosed in the present application are a method and apparatus for determining a risky attribute of user data, and an electronic device. The method comprises: obtaining user data, the user data comprising time data and feature data having a corresponding relationship; encoding the user data to obtain encoded data of the user data, the encoded data comprising a time space hidden state and a feature space hidden state; calculating the time space hidden state and the feature space hidden state on the basis of a bidirectional attention mechanism to obtain a bidirectional attention matrix; on the basis of the bidirectional attention matrix and the encoded data, generating data to be decoded; and decoding the data to be decoded to obtain decoded data of the user data, and determining the risky attribute of the user data on the basis of the decoded data. In the present application, the bidirectional attention mechanism is embedded into an encoding-decoding structure for mining the risky attribute of the user data, and the time space hidden state and the feature space hidden state are integrated to characterize the attention, thereby improving the determining accuracy of the risky attribute of the user data.

Description

Method, device and electronic equipment for determining dangerous attributes of user data

Technical field

This application relates to the technical field of electronic equipment, and more specifically, to a method and device for determining dangerous attributes of user data, and electronic equipment.

Background technique

In the past two years, China's vigorous development of inclusive finance has enabled the Internet financial industry to advance by leaps and bounds. While achieving inclusiveness, it has also provided the black industry with a more convenient, fast and low-cost means of committing crimes, which has brought considerable pressure to relevant Internet financial institutions.

Summary of the invention

In view of the above-mentioned problems, this application proposes a method, device and electronic equipment for determining the dangerous attributes of user data to solve the above-mentioned problems.

In the first aspect, an embodiment of the present application provides a method for determining a dangerous attribute of user data. The method includes: acquiring user data, where the user data includes time data and characteristic data that have a corresponding relationship; Encoding to obtain the encoded data of the user data, the encoded data including the temporal and spatial hidden state and the feature space hidden state; based on the two-way attention mechanism, the temporal and spatial hidden state and the feature space hidden state are calculated to obtain Two-way attention matrix; generate data to be decoded based on the two-way attention matrix and the encoded data; decode the data to be decoded to obtain the decoded data of the user data, and determine the decoded data based on the decoded data Dangerous attributes of user data.

In a second aspect, an embodiment of the present application provides a device for determining a dangerous attribute of user data. The device includes: a user data acquisition module for acquiring user data. The user data includes time data and characteristic data that have a corresponding relationship. Coded data obtaining module, used to code said user data to obtain coded data of said user data, said coded data including time-space hidden state and feature space hidden state; two-way attention matrix obtaining module, used for A two-way attention mechanism, which calculates the hidden state of the time space and the hidden state of the feature space to obtain a two-way attention matrix; a data generation module to be decoded is used to generate the two-way attention matrix and the encoded data Data to be decoded; a dangerous attribute determination module, configured to decode the data to be decoded, obtain decoded data of the user data, and determine the dangerous attribute of the user data based on the decoded data.

In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, the memory is coupled to the processor, the memory stores instructions, and the instructions are executed when the instructions are executed by the processor. The processor executes the above method.

In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores program code, and the program code can be invoked by a processor to execute the above method.

The method, device, and electronic device for determining the dangerous attributes of user data provided in the embodiments of the present application obtain user data, and the user data includes time data and characteristic data that have a corresponding relationship, and the user data is encoded to obtain the encoded data of the user data. The coded data includes time and space hidden state and feature space hidden state. Based on the two-way attention mechanism, the time and space hidden state and feature space hidden state are calculated to obtain a two-way attention matrix. Based on the two-way attention matrix and the coded data, generate to be decoded Data, decode the data to be decoded, obtain the decoded data of the user data, and determine the dangerous attributes of the user data based on the decoded data, so as to design the two-way attention mechanism and embed it into the encoding-decoding structure to mine the dangerous attributes of the user data. The hidden state of time and space and the hidden state of feature space are integrated to represent attention, which improves the accuracy of judging the dangerous attributes of user data.

Description of the drawings

In order to more clearly describe the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings that need to be used in the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained from these drawings without creative work.

FIG. 1 shows a schematic flowchart of a method for determining a dangerous attribute of user data provided by an embodiment of the present application;

Figure 2 shows a schematic diagram of encoding-decoding of user data provided by an embodiment of the present application;

FIG. 3 shows a schematic flowchart of a method for determining a dangerous attribute of user data provided by another embodiment of the present application;

FIG. 4 shows a schematic flowchart of step S203 of the method for determining the dangerous attribute of user data shown in FIG. 3 of the present application;

FIG. 5 shows a schematic flowchart of step S204 of the method for determining a dangerous attribute of user data shown in FIG. 3 of the present application;

FIG. 6 shows a schematic flowchart of step S205 of the method for determining the dangerous attributes of user data shown in FIG. 3 of the present application;

FIG. 7 shows a schematic flowchart of a method for determining a dangerous attribute of user data provided by still another embodiment of the present application;

Fig. 8 shows a block diagram of a device for determining a dangerous attribute of user data provided by an embodiment of the present application;

FIG. 9 shows a block diagram of an electronic device used to execute a method for determining a dangerous attribute of user data according to an embodiment of the present application;

Fig. 10 shows a storage unit for storing or carrying program codes for implementing the method for determining the dangerous attributes of user data according to the embodiment of the present application.

Detailed ways

In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application.

In the past two years, China's vigorous development of inclusive finance has enabled the Internet financial industry to advance by leaps and bounds. While achieving inclusiveness, it has also provided the black industry with a more convenient, fast and low-cost means of committing crimes, which has brought considerable pressure to relevant Internet financial institutions. The construction of a risk control system based on the blacklist strategy is the first step, which can effectively prevent non-cold start and black production attacks through association relationships. In the risk identification of cold start users, based on user static information (such as equipment, environment, location-based services (LBS) information, application software (APP) usage, etc.) and dynamic information ( (Such as the user behavior series) of the risk control model played a key role. In recent years, due to the rapid development of research in the field of natural language processing (NLP) and the implementation of successful cases, the application of sequence models has been paid attention to. More and more financial institutions also hope to model user behavior series data , To discover more black users who cannot be captured due to blacklists and static models.

The current behavioral series models are divided into two categories: 1) Behavioral sequence models based on non-linear autoregressive models (NARX), such as recurrent neural networks (RNN) and its variants, long and short-term memory networks (LSTM), gated recurrent units ( GRU), etc.; 2) On the basis of NARX, an attention mechanism is introduced to control the dependence of tags on different inputs. However, the inventor found through research that for category 1), the behavioral sequence model based on the nonlinear autoregressive model is usually modeled based on the encoder-decoder method. The data is encoded into a hidden state by the encoder, and then decoded by the decoder and passed through the corresponding The fully connected layer and softmax are used for label mapping. However, when the behavior series are getting longer and longer, only relying on the hidden state for information transfer cannot explicitly represent the global information. For category 2), because the NARX model is insufficient for processing long sequences, the attention mechanism-based model has received more attention because it can perceive global information. The attention mechanism is a type of neural network that is embedded in the neural network to judge differences. The algorithm for input weights, although the attention mechanism has a global vision, but the current attention mechanism-based models can only be modeled from the time dimension or feature dimension alone, thus ignoring the behavioral series in the time-feature cross dimension. information.

In response to the above problems, the inventor has discovered through long-term research and proposed the method, device, and electronic device for determining the dangerous attributes of user data provided by the embodiments of this application. The two-way attention mechanism is designed to be embedded in the encoding-decoding structure for mining The dangerous attributes of user data integrate the hidden state of time and space and the hidden state of feature space to represent attention, and improve the accuracy of judging the dangerous attributes of user data. Among them, the specific method for determining the dangerous attribute of the user data will be described in detail in the subsequent embodiments.

Please refer to FIG. 1. FIG. 1 shows a schematic flowchart of a method for determining a dangerous attribute of user data provided by an embodiment of the present application. The method for determining the dangerous data of user data is used to mine the dangerous attributes of user data by designing a two-way attention mechanism and embedding it into the encoding-decoding structure, and integrate the hidden state of time and space and the hidden state of feature space to represent attention, Improve the accuracy of determining the dangerous attributes of user data. In a specific embodiment, the method for determining the dangerous attribute of user data is applied to the device 200 for determining the dangerous attribute of user data as shown in FIG. 8 and the electronic equipment 100 (FIG. 9 ). The following will take an electronic device as an example to describe the specific process of this embodiment. Of course, it is understandable that the electronic device applied in this embodiment may be a smart phone, a tablet computer, a wearable electronic device, etc., which is not limited here. The facet will elaborate on the process shown in Figure 1. The method for determining the dangerous attributes of the user data may specifically include the following steps:

Step S101: Obtain user data, where the user data includes time data and characteristic data that have a corresponding relationship.

In this embodiment, user data can be obtained. Among them, user data can be obtained in real time, user data can be obtained at preset time intervals, user data can be obtained at a specified time, or user data can be obtained according to other preset rules. This is not limited. In some embodiments, user data can be obtained locally from the electronic device (the electronic device pre-records and stores user data), user data can be obtained from a server connected to the electronic device (the server pre-records and stores user data), etc. Not limited.

In some embodiments, the user data includes time data and characteristic data that have a corresponding relationship, where the time data can use months as the time period, weeks as the time period, days as the time period, or hours as the time. Period, etc., for example, when the time data takes months as the time period, the user data can include the characteristic data in xx months; when the time data takes the week as the time period, the user data can include the characteristic data in xx weeks ; When the time data takes the day as the time period, the user data can include the characteristic data in xx days; when the user data takes the hour as the time period, the user data can include the characteristic data in the xx time period, etc., here Not limited.

In some embodiments, the characteristic data may include attribute data, behavior data, and the like. Among them, the attribute data may include, but is not limited to: age data, gender data, geographic location data, and hobby data. Among them, the behavior data may include, but is not limited to: login data, browsing data, click data, jump data, payment data, and evaluation data. In this embodiment, the acquired user data may include: the user’s age data, gender data, geographic location data, hobby data, login data, browsing data, click data, jump data, payment data, Evaluation data; the user’s age data, gender data, geographic location data, hobby data, login data, browsing data, click data, jump data, payment data, evaluation data, etc. within xx months are not limited here.

Step S102: Encode the user data to obtain encoded data of the user data, where the encoded data includes a time-space hidden state and a feature-space hidden state.

In this embodiment, after acquiring user data, the user data may be encoded (Encoder) to obtain the encoded data of the user data, where the acquired encoded data may include temporal and spatial hidden states and feature space hidden states. Please refer to FIG. 2. FIG. 2 shows a schematic diagram of encoding-decoding of user data provided by an embodiment of the present application. As shown in FIG. 2, in some embodiments, the user data is Encoder, and the encoded data output during the Encoder process Is the hidden state of RNN

Among them, T represents the length of the sequence, M represents the length of the hidden state,

Identify the hidden state of encoded data with sequence length T and hidden state length M, where,

Indicates the hidden state of all channels at ^{time t th} , that is, the hidden state of the feature space,

Indicates the hidden state of the m ^th channel at all time points, that is, the hidden state of time and space, as shown in A in Figure 2.

Step S103: Based on the two-way attention mechanism, calculate the time-space hidden state and the feature space hidden state to obtain a two-way attention matrix.

In this embodiment, after the time-space hidden state and the feature-space hidden state are obtained, the time-space hidden state and the feature-space hidden state can be calculated based on the two-way attention mechanism to obtain a two-way attention matrix. In some embodiments, after obtaining the hidden state of time and space and the hidden state of feature space, based on the two-way attention mechanism, the first weighted calculation is performed on the hidden state of time and space, and the second weighted calculation is performed on the hidden state of feature space to obtain Two-way attention matrix. In some embodiments, after obtaining the time-space hidden state and the feature-space hidden state, the time-space hidden state and the feature-space hidden state can be weighted based on the two-way attention mechanism, and the corresponding fully connected layer and softmax can be used for labeling. Mapping to obtain a two-way attention matrix.

Step S104: Generate data to be decoded based on the two-way attention matrix and the encoded data.

In this embodiment, after obtaining the bidirectional attention matrix and the coded data, the data to be decoded can be generated based on the bidirectional attention matrix and the coded data. In some embodiments, after obtaining the bidirectional attention matrix and the encoded data, the bidirectional attention matrix and the encoded data can be calculated to obtain the data to be decoded.

Step S105: Decode the data to be decoded, obtain decoded data of the user data, and determine the dangerous attribute of the user data based on the decoded data.

In this embodiment, after the data to be decoded is generated, the data to be decoded can be decoded (Decoder) to obtain the decoded data of the user data, and the dangerous attributes of the user data can be determined based on the decoded data. Among them, because the two-way attention mechanism integrates time-space and feature-space information, it can extend the dimensionality of the attention mechanism's global vision, thereby more accurately mapping label information from different features at the same time and the same feature series at different times. Obtaining the decoded data makes the risk attributes of the user data determined based on the decoded data more accurate.

In some embodiments, determining the risk attribute of the user data based on the decoded data may be determining that the user data is dangerous based on the decoded data, or determining that the user data is not dangerous based on the decoded data. In some embodiments, determining the risk attribute of the user data based on the decoded data may be determining the risk level of the user data based on the decoded data, for example, determining that the risk level of the user data is high based on the decoded data, and determining the risk level of the user data based on the decoded data It is medium to high, the risk level of user data is determined to be medium based on decoded data, the risk level of user data is determined to be medium to low based on decoded data, and the risk level of user data is determined to be low based on decoded data, etc., which are not limited here.

The method for determining the dangerous attributes of user data provided in an embodiment of the present application obtains user data. The user data includes time data and characteristic data that have a corresponding relationship. The user data is encoded to obtain the encoded data of the user data. The encoded data includes time. Spatial hidden state and feature space hidden state, based on the two-way attention mechanism, calculate the time-space hidden state and feature space hidden state to obtain a two-way attention matrix, based on the two-way attention matrix and encoded data, generate data to be decoded, and to be decoded The data is decoded to obtain the decoded data of the user data, and the dangerous attributes of the user data are determined based on the decoded data. The two-way attention mechanism is designed to be embedded in the encoding-decoding structure to mine the dangerous attributes of the user data and hide the time and space. The hidden state of state and feature space integrates and represents attention, which improves the accuracy of judging the dangerous attributes of user data.

Please refer to FIG. 3, which shows a schematic flowchart of a method for determining a dangerous attribute of user data provided by another embodiment of the present application. The following will elaborate on the process shown in FIG. 3, and the method for determining the dangerous attributes of the user data may specifically include the following steps:

Step S201: Obtain user data, where the user data includes time data and characteristic data that have a corresponding relationship.

Step S202: Encode the user data to obtain encoded data of the user data, where the encoded data includes a time-space hidden state and a feature-space hidden state.

For the specific description of step S201 to step S202, please refer to step S101 to step S102, which will not be repeated here.

Step S203: Based on the two-way attention mechanism, calculate the temporal and spatial hidden state to obtain the first fully connected layer.

In this embodiment, after the time and space hidden state is obtained, the time and space hidden state can be calculated based on the two-way attention mechanism to obtain the first fully connected layer, as shown in B in FIG. 2.

Please refer to FIG. 4, which shows a schematic flowchart of step S203 of the method for determining the dangerous attribute of user data shown in FIG. 3 of the present application. The following will elaborate on the process shown in FIG. 4, and the method may specifically include the following steps:

Step S2031: Obtain the first weight coefficient matrix.

In this embodiment, after obtaining the temporal and spatial hidden state, the first weight coefficient matrix can be obtained based on the two-way attention mechanism. Among them, the first weight coefficient matrix can be learned during the training process. In some embodiments, the first weight coefficient matrix may be obtained in the training process and then stored locally in the electronic device, and after obtaining the temporal and spatial hidden state, the first weight coefficient matrix may be directly obtained locally from the electronic device. In another embodiment, the first weight coefficient matrix may be obtained during the training process and stored in a server connected to the electronic device in communication, and after obtaining the temporal and spatial hidden state, the first weight coefficient matrix may be obtained from the server through a wireless network or a data network. A matrix of weight coefficients.

Step S2032: Calculate the first weight coefficient matrix and the temporal and spatial hidden state to obtain a first fully connected layer.

In this embodiment, after obtaining the first weight coefficient matrix, the first weight coefficient matrix and the temporal and spatial hidden state can be calculated to obtain the first fully connected layer. In some embodiments, it can be based on

Calculate the first weight coefficient matrix and the time-space hidden state to obtain the first fully connected layer α _t , where W _α is the first weight coefficient matrix,

It is the hidden state of time and space. In some embodiments, the first weight coefficient matrix is

Step S204: Based on the two-way attention mechanism, calculate the hidden state of the feature space to obtain a second fully connected layer.

In this embodiment, after obtaining the hidden state of the feature space, the hidden state of the feature space can be calculated based on the bidirectional attention mechanism to obtain the second fully connected layer, as shown in C in FIG. 2.

Please refer to FIG. 5, which shows a schematic flowchart of step S204 of the method for determining the dangerous attribute of user data shown in FIG. 3 of the present application. The following will elaborate on the process shown in FIG. 5, and the method may specifically include the following steps:

Step S2041: Obtain a second weight coefficient matrix.

In this embodiment, after obtaining the hidden state of the feature space, the second weight coefficient matrix can be obtained based on the two-way attention mechanism. Wherein, the second weight coefficient matrix can be learned during the training process. In some embodiments, the second weighting coefficient matrix may be obtained in the training process and then stored locally in the electronic device, and after obtaining the hidden state of the feature space, the second weighting coefficient matrix may be directly obtained locally from the electronic device. In another embodiment, the second weight coefficient matrix may be obtained during the training process and then stored in a server that is communicatively connected to the electronic device, and after obtaining the hidden state of the feature space, the second weight coefficient matrix may be obtained from the server through a wireless network or a data network. Two-weight coefficient matrix.

Step S2042: Calculate the second weight coefficient matrix and the hidden state of the feature space to obtain a second fully connected layer.

In this embodiment, after the second weight coefficient matrix is obtained, the second weight coefficient matrix and the hidden state of the feature space can be calculated to obtain the second fully connected layer. In some embodiments, the second weight coefficient matrix and the hidden state of the feature space may be calculated ^{based on β m} =W _β h ^m ^{to obtain the second fully connected layer β m} , where W _β is the second weight coefficient matrix, h ^m is the hidden state of the feature space. In some embodiments, the second weight coefficient matrix is

Step S205: Perform calculations on the first fully connected layer and the second fully connected layer to obtain the two-way attention matrix.

In this embodiment, after obtaining the first fully connected layer and the second fully connected layer, the first fully connected layer and the second fully connected layer can be calculated to obtain the two-way attention matrix, as shown in D in Figure 2 Show.

Please refer to FIG. 6. FIG. 6 shows a schematic flowchart of step S205 of the method for determining the dangerous attribute of user data shown in FIG. 3 of the present application. The process shown in FIG. 6 will be described in detail below, and the method may specifically include the following steps:

Step S2051: Obtain a weight coefficient vector.

In this embodiment, after obtaining the first fully connected layer and the second fully connected layer, the weight coefficient vector can be obtained based on the two-way attention mechanism. Among them, the weight coefficient vector can be learned during the training process. In some embodiments, the weight coefficient vector may be obtained in the training process and then stored locally in the electronic device, and after obtaining the first fully connected layer and the second fully connected layer, the weight coefficient can be obtained directly from the electronic device. vector. In another embodiment, the weight coefficient vector can be obtained in the training process and then stored in the server connected to the electronic device communication, and after obtaining the first fully connected layer and the second fully connected layer, it can be passed through a wireless network or a data network. Obtain the weight coefficient vector from the server.

Step S2052: Calculate the weight coefficient vector, the first fully connected layer, and the second fully connected layer to obtain the bidirectional attention matrix.

In this embodiment, after the weight coefficient vector is obtained, the weight coefficient vector, the first fully connected layer, and the second fully connected layer can be calculated to obtain the bidirectional attention matrix. In some embodiments, the weight coefficient vector, the first fully connected layer, and the second fully connected layer may be calculated _{based on r t} ^m =W _r tanh(α _t +β ^m _{) to obtain the bidirectional attention matrix r t} ^m , Among them, W _r is the weight coefficient vector, α _t is the first fully connected layer, and β ^m is the second fully connected layer. In some embodiments, the weight coefficient vector is

Among them, in this embodiment, since the two-way attention matrix r _t ^m integrates the first fully connected layer α _t and the second fully connected layer β ^m at the same time, the two-way attention matrix r _t ^{m has} both time, space, and time. The distribution information of the feature space, that is, the information reflected by the _{bidirectional attention matrix r t} ^{m is more accurate.}

Step S206: Process the bidirectional attention matrix based on the softmax function to obtain a probability matrix.

In this embodiment, after obtaining the bidirectional attention matrix, the bidirectional attention matrix can be processed based on the softmax function to obtain the probability matrix. Specifically, in order to ensure that the sum of all attention weights in the same channel is 1, Then the two-way attention matrix can be processed by the softmax function to obtain the probability matrix. In some embodiments, it can be based on

Calculate the bidirectional attention matrix to obtain the probability matrix

Among them, r _t ^m is a two-way attention matrix.

Step S207: Generate the data to be decoded based on the probability matrix and the encoded data.

In some embodiments, after obtaining the probability matrix and the encoded data, it can be based on

Calculate the probability matrix and encoded data to obtain the data to be decoded

in,

Is probabilistic data,

Is the encoded data.

Step S208: Decode the data to be decoded, obtain decoded data of the user data, and determine the dangerous attribute of the user data based on the decoded data.

For the specific description of step S208, please refer to step S105, which will not be repeated here.

According to another embodiment of the present application, a method for determining a dangerous attribute of user data is obtained. The user data includes time data and characteristic data that have a corresponding relationship. The user data is encoded to obtain the encoded data of the user data. The encoded data includes time. Space hidden state and feature space hidden state, based on the two-way attention mechanism, calculate the hidden state of time and space, obtain the first fully connected layer, based on the two-way attention mechanism, calculate the hidden state of feature space, obtain the second fully connected layer , Calculate the first fully connected layer and the second fully connected layer to obtain a two-way attention matrix, process the two-way attention matrix based on the softmax function to obtain a probability matrix, and generate the data to be decoded based on the probability matrix and the encoded data. The decoded data is decoded to obtain the decoded data of the user data, and the dangerous attributes of the user data are determined based on the decoded data. Compared with the method for determining the dangerous attributes of user data shown in FIG. 1, this embodiment also calculates the hidden state of time and space and the hidden state of feature space respectively to obtain two different fully connected layers based on two different fully connected layers. The connection layer obtains a two-way attention matrix to improve the accuracy of judging the dangerous attributes of the user data.

Please refer to FIG. 7, which shows a schematic flowchart of a method for determining a dangerous attribute of user data provided by another embodiment of the present application. The process shown in FIG. 7 will be described in detail below. The method for determining the dangerous attributes of the user data may specifically include the following steps:

Step S301: Obtain user data, where the user data includes time data and characteristic data that have a corresponding relationship.

Step S302: Encode the user data to obtain encoded data of the user data, where the encoded data includes a time-space hidden state and a feature-space hidden state.

Step S303: Based on the two-way attention mechanism, calculate the time-space hidden state and the feature space hidden state to obtain a two-way attention matrix.

Step S304: Generate data to be decoded based on the two-way attention matrix and the encoded data.

Step S305: Decode the data to be decoded to obtain decoded data of the user data.

For the specific description of step S301 to step S305, please refer to step S101 to step S105, which will not be repeated here.

Step S306: When the decoded data is the first data, it is determined that the dangerous attribute of the user data is dangerous.

In some embodiments, the first data and the second data may be set in advance, where the first data may be used to characterize the risk attribute of the user data corresponding to the decoded data as dangerous, and the second data may be used to characterize the interface data The dangerous attribute of the corresponding user data is not dangerous. For example, the first data may be "1" and the second data may be "0".

In some embodiments, after the decoded data is obtained, the decoded data can be compared with the first data and the second data, respectively, to determine whether the decoded data is the first data or the second data. Wherein, when the comparison result characterizes that the decoded data is the first data, for example, when it is determined that the decoded data is "1", it can be determined that the user attribute of the user data corresponding to the decoded data is dangerous.

Step S307: When the information request corresponding to the user data is received, the information request is rejected.

In some embodiments, when it is determined that the attribute information of the user data is dangerous, it characterizes that the operation performed by the user is a dangerous operation or the user's behavior is a dangerous behavior, and then when the information request corresponding to the user data is received, refuse This information is requested to avoid dangerous operations or dangerous behaviors.

Step S308: Send an alarm prompt message, and add the user data to the blacklist.

In some embodiments, when it is determined that the attribute information of the user data is dangerous, it characterizes that the operation performed by the user is a dangerous operation or the behavior of the user is a dangerous behavior, and then an alarm prompt message may be issued to prompt the user to perform actions on the user data. Corresponding defensive operations. Among them, the alarm prompt information can include voice alarm prompt information, text alarm prompt information, picture alarm prompt information, etc. In addition, the alarm prompt information can be directly output on the electronic device, or the alarm information can be sent to the server for output through the electronic device. This is not limited.

In some implementations, when it is determined that the attribute information of the user data is dangerous, it characterizes that the operation performed by the user is a dangerous operation or the behavior of the user is a dangerous behavior, the user data can be added to the blacklist to directly reject Any operation of the user data to avoid dangerous operations or dangerous behaviors. In some embodiments, when it is determined that the attribute information of the user data is dangerous, the number of times the attribute information of the user data is dangerous can be obtained based on the historical information of the user data, and the attribute information of the user data is the number of times dangerous. When the specified number of times (such as 3 times) is reached, the user data is added to the blacklist, and when the attribute information of the user data is dangerous and the specified number of times is not reached, an alarm message is issued.

Step S309: When the decoded data is the second data, it is determined that the dangerous attribute of the user data is not dangerous.

In some embodiments, after the decoded data is obtained, the decoded data can be compared with the first data and the second data, respectively, to determine whether the decoded data is the first data or the second data. Wherein, when the comparison result indicates that the decoded data is the second data, for example, when it is determined that the decoded data is "0", it can be determined that the user attribute of the user data corresponding to the decoded data is not dangerous.

Step S310: When receiving the information request corresponding to the user data, respond to the information request.

In some embodiments, when it is determined that the attribute information of the user data is not dangerous, it is characterized that the operation performed by the user is not dangerous or the behavior of the user is not dangerous, then the information request corresponding to the user data is received At the time, respond to the information request to respond to the user's request normally to meet the user's needs.

According to another embodiment of the present application, a method for determining a dangerous attribute of user data is obtained. The user data includes time data and characteristic data that have a corresponding relationship. The user data is encoded to obtain the encoded data of the user data. The encoded data includes Time and space hidden state and feature space hidden state, based on the two-way attention mechanism, calculate the time and space hidden state and feature space hidden state to obtain a two-way attention matrix, based on the two-way attention matrix and encoded data, generate the data to be decoded, and treat The decoded data is decoded to obtain the decoded data of the user data. When the decoded data is the first data, the dangerous attribute of the user data is determined to be dangerous. When the information request corresponding to the user data is received, the information request is rejected, an alarm is issued, and The user data is added to the blacklist, and when the decoded data is the second data, it is determined that the user data has a dangerous attribute as not dangerous, and when the decoded data is the second data, the information request is responded to. Compared with the method for determining the dangerous attribute of user data shown in FIG. 1, this embodiment also rejects the information request corresponding to the user data when the dangerous attribute of the user data is determined to be dangerous, and when the dangerous data of the user data is not dangerous Respond to the information request corresponding to the user data to improve the accuracy of determining the dangerous attributes of the user data.

Please refer to FIG. 8. FIG. 8 shows a block diagram of a device 200 for determining a dangerous attribute of user data provided by an embodiment of the present application. The following will elaborate on the block diagram shown in FIG. 8, the device 200 for determining the dangerous attributes of user data includes: a user data acquisition module 210, an encoded data acquisition module 220, a bidirectional attention matrix acquisition module 230, and a to-be-decoded data generation module 240 And the dangerous attribute determination module 250, in which:

The user data acquisition module 210 is configured to acquire user data, and the user data includes time data and characteristic data that have a corresponding relationship.

The coded data obtaining module 220 is configured to code the user data to obtain coded data of the user data, and the coded data includes a time-space hidden state and a feature-space hidden state.

The bidirectional attention matrix obtaining module 230 is configured to calculate the temporal and spatial hidden state and the feature space hidden state based on the bidirectional attention mechanism to obtain a bidirectional attention matrix.

Further, the bidirectional attention matrix obtaining module 230 includes: a first fully connected layer obtaining submodule, a second fully connected layer obtaining submodule, and a bidirectional attention matrix obtaining submodule, wherein:

The first fully connected layer obtaining sub-module is used to calculate the temporal and spatial hidden state based on the two-way attention mechanism to obtain the first fully connected layer.

Further, the first fully connected layer obtaining submodule includes: a first weight coefficient matrix obtaining unit and a first fully connected layer obtaining unit, wherein:

The first weight coefficient matrix obtaining unit is configured to obtain the first weight coefficient matrix.

The first fully connected layer obtaining unit is configured to calculate the first weight coefficient matrix and the temporal and spatial hidden state to obtain the first fully connected layer.

Further, the first fully connected layer obtaining unit includes: a first fully connected layer obtaining subunit, wherein:

The first fully connected layer obtains subunits for use based on

Calculate the first weight coefficient matrix and the temporal and spatial hidden state to obtain the first fully connected layer α _t , where W _α is the first weight coefficient matrix,

It is the hidden state of time and space.

The second fully connected layer obtaining sub-module is used to calculate the hidden state of the feature space based on the two-way attention mechanism to obtain the second fully connected layer.

Further, the second fully connected layer obtaining submodule includes: a second weight coefficient matrix obtaining unit and a second fully connected layer obtaining unit, wherein:

The second weight coefficient matrix obtaining unit is used to obtain the second weight coefficient matrix.

The second fully connected layer obtaining unit is configured to calculate the second weight coefficient matrix and the hidden state of the feature space to obtain a second fully connected layer.

Further, the second fully connected layer obtaining unit includes: a second fully connected layer obtaining subunit, wherein:

The second fully connected layer obtaining subunit is used to calculate the second weight coefficient matrix and the hidden state of the feature space ^{based on β m} =W _β h ^m ^{to obtain the second fully connected layer β m} , where W _β Is the second weight coefficient matrix, h ^m is the hidden state of the feature space.

The bidirectional attention matrix obtaining sub-module is used to calculate the first fully connected layer and the second fully connected layer to obtain the bidirectional attention matrix.

Further, the bidirectional attention matrix obtaining submodule includes: a weight coefficient vector obtaining unit and a bidirectional attention matrix obtaining unit, wherein:

The weight coefficient vector obtaining unit is used to obtain the weight coefficient vector.

The bidirectional attention matrix obtaining unit is configured to calculate the weight coefficient vector, the first fully connected layer, and the second fully connected layer to obtain the bidirectional attention matrix.

Further, the bidirectional attention matrix obtaining unit includes: a bidirectional attention matrix obtaining subunit, wherein:

The bidirectional attention matrix obtaining subunit is used to calculate the weight coefficient vector, the first fully connected layer, and the second fully connected layer _{based on r t} ^m =W _r tanh(α _t +β ^{m ),} Obtain the bidirectional attention matrix r _t ^m , where W _r is the weight coefficient vector, α _t is the first fully connected layer, and β ^m is the second fully connected layer.

The to-be-decoded data generation module 240 is configured to generate the to-be-decoded data based on the two-way attention matrix and the encoded data.

Further, the to-be-decoded data generation module 240 includes: a probability matrix obtaining sub-module and a to-be-decoded data generation sub-module, wherein:

The probability matrix obtaining sub-module is used to process the bidirectional attention matrix based on the softmax function to obtain the probability matrix.

Further, the probability matrix obtaining sub-module includes: a probability matrix obtaining unit, wherein:

Probability matrix acquisition unit, used based on

Calculate the bidirectional attention matrix to obtain the probability matrix

Among them, r _t ^m is a two-way attention matrix.

The to-be-decoded data generation sub-module is configured to generate the to-be-decoded data based on the probability matrix and the encoded data.

Further, the to-be-decoded data generation sub-module includes: a to-be-decoded data generation unit, wherein:

The data generating unit to be decoded is used to generate data based on

Calculate the probability matrix and the encoded data to obtain the data to be decoded

in,

Is probabilistic data,

Is the encoded data.

The dangerous attribute determination module 250 is configured to decode the data to be decoded, obtain decoded data of the user data, and determine the dangerous attribute of the user data based on the decoded data.

Further, the dangerous attribute determining module 250 includes: a decoding data obtaining sub-module, a first dangerous attribute determining sub-module, and a second dangerous attribute determining sub-module, wherein:

The decoded data obtaining sub-module is used to decode the data to be decoded to obtain decoded data of the user data.

The first risk attribute determination sub-module is configured to determine that the risk attribute of the user data is dangerous when the decoded data is the first data.

The second risk attribute determination sub-module is configured to determine that the risk attribute of the user data is not dangerous when the decoded data is the second data.

Further, the dangerous attribute determining module 250 further includes: a request response sub-module, wherein:

The request response submodule is configured to respond to the information request when the information request corresponding to the user data is received.

Further, the dangerous attribute determination module 250 further includes: a request rejection sub-module, wherein:

The request rejection sub-module is configured to reject the information request when the information request corresponding to the user data is received.

Further, the dangerous attribute determining module 250 further includes: a prompt message issuing submodule, wherein:

The prompt information issuing sub-module is used for issuing alarm prompt information and adding the user data to the blacklist.

Those skilled in the art can clearly understand that, for the convenience and conciseness of the description, the specific working process of the above described device and module can be referred to the corresponding process in the foregoing method embodiment, which will not be repeated here.

In the several embodiments provided in this application, the coupling between the modules may be electrical, mechanical or other forms of coupling.

In addition, the functional modules in the various embodiments of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or software functional modules.

Please refer to FIG. 9, which shows a structural block diagram of an electronic device 100 provided by an embodiment of the present application. The electronic device 100 may be an electronic device capable of running application programs, such as a smart phone, a tablet computer, or an e-book. The electronic device 100 in this application may include one or more of the following components: a processor 110, a memory 120, and one or more application programs, where one or more application programs may be stored in the memory 120 and configured to be composed of one Or multiple processors 110 execute, and one or more programs are configured to execute the method described in the foregoing method embodiment.

The processor 110 may include one or more processing cores. The processor 110 uses various interfaces and lines to connect various parts of the entire electronic device 100, and executes by running or executing instructions, programs, code sets, or instruction sets stored in the memory 120, and calling data stored in the memory 120. Various functions and processing data of the electronic device 100. Optionally, the processor 110 may adopt at least one of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). A kind of hardware form to realize. The processor 110 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), a modem, and the like. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is used for rendering and drawing the content to be displayed; the modem is used for processing wireless communication. It is understandable that the above-mentioned modem may not be integrated into the processor 110, but may be implemented by a communication chip alone.

The memory 120 may include random access memory (RAM) or read-only memory (Read-Only Memory). The memory 120 may be used to store instructions, programs, codes, code sets or instruction sets. The memory 120 may include a program storage area and a data storage area, where the program storage area may store instructions for implementing the operating system and instructions for implementing at least one function (such as touch function, sound playback function, image playback function, etc.) , Instructions used to implement the following various method embodiments, etc. The storage data area can also store data (such as phone book, audio and video data, chat record data) created by the electronic device 100 during use.

Please refer to FIG. 10, which shows a structural block diagram of a computer-readable storage medium provided by an embodiment of the present application. The computer-readable medium 300 stores program code, and the program code can be invoked by a processor to execute the method described in the foregoing method embodiment.

The computer-readable storage medium 300 may be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read Only Memory), EPROM, hard disk, or ROM. Optionally, the computer-readable storage medium 300 includes a non-transitory computer-readable storage medium. The computer-readable storage medium 300 has storage space for the program code 310 for executing any method steps in the above-mentioned methods. These program codes can be read from or written into one or more computer program products. The program code 310 may be compressed in a suitable form, for example.

In summary, the method, device, and electronic device for determining the dangerous attributes of user data provided in the embodiments of the present application acquire user data, and the user data includes time data and characteristic data that have a corresponding relationship. The user data is encoded to obtain user data. The coded data includes time-space hidden state and feature-space hidden state. Based on the two-way attention mechanism, the time-space hidden state and feature-space hidden state are calculated to obtain a two-way attention matrix, based on the two-way attention matrix and coding Data, generate the data to be decoded, decode the data to be decoded, obtain the decoded data of the user data, and determine the dangerous attributes of the user data based on the decoded data, so as to design a two-way attention mechanism and embed it into the encoding-decoding structure to mine user data It integrates the hidden state of time and space and the hidden state of feature space to represent attention, and improves the accuracy of judging the dangerous attributes of user data.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the application, not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: The technical solutions recorded in the foregoing embodiments are modified, or some of the technical features thereof are equivalently replaced; these modifications or replacements do not drive the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

A method for determining dangerous attributes of user data, characterized in that the method includes:

Acquiring user data, where the user data includes time data and characteristic data that have a corresponding relationship;

Encoding the user data to obtain encoded data of the user data, where the encoded data includes a time-space hidden state and a feature-space hidden state;

Based on a two-way attention mechanism, calculate the hidden state of the time space and the hidden state of the feature space to obtain a two-way attention matrix;

Generating data to be decoded based on the bidirectional attention matrix and the encoded data;

The data to be decoded is decoded to obtain the decoded data of the user data, and the dangerous attribute of the user data is determined based on the decoded data.
The method according to claim 1, wherein the calculation of the temporal and spatial hidden state and the feature space hidden state based on a two-way attention mechanism to obtain a two-way attention matrix comprises:

Based on the two-way attention mechanism, calculate the temporal and spatial hidden state to obtain the first fully connected layer;

Based on the two-way attention mechanism, calculate the hidden state of the feature space to obtain the second fully connected layer;

Perform calculations on the first fully connected layer and the second fully connected layer to obtain the bidirectional attention matrix.
The method according to claim 2, wherein the calculating the temporal and spatial hidden state based on the two-way attention mechanism to obtain the first fully connected layer comprises:

Obtain the first weight coefficient matrix;

The first weight coefficient matrix and the time-space hidden state are calculated to obtain the first fully connected layer.
The method according to claim 3, wherein the calculating the first weight coefficient matrix and the temporal and spatial hidden state to obtain the first fully connected layer comprises:

based on
Calculate the first weight coefficient matrix and the temporal and spatial hidden state to obtain the first fully connected layer α t , where W α is the first weight coefficient matrix,
It is the hidden state of time and space.
The method according to claim 4, wherein the calculating the hidden state of the feature space based on the bidirectional attention mechanism to obtain the second fully connected layer comprises:

Obtaining the second weight coefficient matrix;

The second weight coefficient matrix and the hidden state of the feature space are calculated to obtain a second fully connected layer.
The method according to claim 5, wherein the calculating the second weight coefficient matrix and the hidden state of the feature space to obtain a second fully connected layer comprises:

Calculate the second weight coefficient matrix and the hidden state of the feature space based on β m =W β h m to obtain a second fully connected layer β m , where W β is the second weight coefficient matrix, and h m is the feature Space hidden state.
The method according to claim 6, wherein the calculating the first fully connected layer and the second fully connected layer to obtain the two-way attention matrix comprises:

Obtain the weight coefficient vector;

The weight coefficient vector, the first fully connected layer, and the second fully connected layer are calculated to obtain the bidirectional attention matrix.
The method according to claim 7, wherein the calculating the weight coefficient vector, the first fully connected layer, and the second fully connected layer to obtain the two-way attention matrix comprises:

based on
The weight coefficient vector, the first fully connected layer, and the second fully connected layer are calculated to obtain the bidirectional attention matrix r t m , where W r is the weight coefficient vector, and α t is the first Fully connected layer, β m is the second fully connected layer.
The method according to claim 8, wherein the first weight coefficient matrix is
The second weight coefficient matrix is
The weight coefficient vector is
The method according to claim 8 or 9, wherein the generating data to be decoded based on the two-way attention matrix and the encoded data comprises:

Processing the bidirectional attention matrix based on the softmax function to obtain a probability matrix;

Based on the probability matrix and the encoded data, the data to be decoded is generated.
The method according to claim 10, wherein the processing the bidirectional attention matrix based on a softmax function to obtain a probability matrix comprises:

based on
Calculate the bidirectional attention matrix to obtain the probability matrix
Among them, r t m is a two-way attention matrix.
The method according to claim 11, wherein said generating said data to be decoded based on said probability matrix and said encoded data comprises:

based on
Calculate the probability matrix and the encoded data to obtain the data to be decoded
in,
Is probabilistic data,
Is the encoded data.
The method according to any one of claims 1-12, wherein the encoded data is
Among them, T represents the length of the sequence, M represents the length of the hidden state,
Identify the hidden state of encoded data with sequence length T and hidden state length M.
The method according to any one of claims 1-13, wherein the decoding of the to-be-decoded data to obtain the decoded data of the user data, and the determination of the user data based on the decoded data Dangerous attributes, including:

Decode the data to be decoded to obtain decoded data of the user data;

When the decoded data is the first data, it is determined that the dangerous attribute of the user data is dangerous;

When the decoded data is the second data, it is determined that the dangerous attribute of the user data is not dangerous.
The method according to claim 14, wherein, when the decoded data is the second data, after determining that the dangerous attribute of the user data is not dangerous, the method further comprises:

When receiving the information request corresponding to the user data, respond to the information request.
The method according to claim 14 or 15, wherein when the decoded data is the first data, after determining that the dangerous attribute of the user data is dangerous, the method further comprises:

When the information request corresponding to the user data is received, the information request is rejected.
The method according to any one of claims 14-16, wherein when the decoded data is the first data, after determining that the risk attribute of the user data is dangerous, the method further comprises:

A warning message is issued, and the user data is added to the blacklist.
A device for determining dangerous attributes of user data, characterized in that the device comprises:

The user data acquisition module is configured to acquire user data, where the user data includes time data and characteristic data that have a corresponding relationship;

The coded data obtaining module is configured to code the user data to obtain coded data of the user data, the coded data including time-space hidden state and feature-space hidden state;

A two-way attention matrix obtaining module is used to calculate the hidden state of the time space and the hidden state of the feature space based on the two-way attention mechanism to obtain a two-way attention matrix;

A data generating module to be decoded, configured to generate data to be decoded based on the two-way attention matrix and the encoded data;

The dangerous attribute determination module is configured to decode the data to be decoded, obtain decoded data of the user data, and determine the dangerous attribute of the user data based on the decoded data.
An electronic device, comprising a memory and a processor, the memory is coupled to the processor, the memory stores instructions, and the processor executes the instructions when the instructions are executed by the processor. The method of any one of 1-17 is required.
A computer-readable storage medium, wherein the computer-readable storage medium stores program code, and the program code can be called by a processor to execute the method according to any one of claims 1-17 .