WO2023050670A1

WO2023050670A1 - False information detection method and system, computer device, and readable storage medium

Info

Publication number: WO2023050670A1
Application number: PCT/CN2022/074411
Authority: WO
Inventors: 舒畅; 陈又新
Original assignee: 平安科技（深圳）有限公司
Priority date: 2021-09-30
Filing date: 2022-01-27
Publication date: 2023-04-06
Also published as: CN113869431A

Abstract

The present application discloses a false information detection method, comprising: obtaining data to be detected comprising source information to be detected and reply information to be detected of said source information at the current time; performing vectorization processing on said data to obtain a source feature vector to be detected corresponding to said source information and a reply feature vector to be detected corresponding to said reply information; performing encoding processing on the feature vector of said data by means of the encoding layer of a pre-trained false information classification model, so as to obtain the feature code of said data; performing classification pre-determination on the feature code of said data by means of a trained deep reinforcement learning model, so as to determine whether said data needs to be classified; and if yes, classifying said data by means of the classification layer of the false information classification model according to a source feature code to be detected and a reply feature code to be detected, so as to obtain a first classification result corresponding to said data. The present application can perform real-time detection on said data.

Description

False information detection method, system, computer equipment and readable storage medium

This application claims the priority of the Chinese patent application submitted to the China Patent Office on September 30, 2021 with the application number 202111156357.5 and the title of the invention is "False Information Detection Method, System, Computer Equipment, and Readable Storage Medium", the entire content of which Incorporated in this application by reference.

technical field

The embodiments of the present application relate to the technical field of data processing, and in particular to a false information detection method, system, computer equipment, and readable storage medium.

Background technique

With the rapid development of the Internet and the self-media industry, people have to receive and send countless information every day, entering the era of information explosion, which affects people's lives all the time. However, the inventors have found that, just like people in traditional oral communication, the information delivered by the Internet is not completely true and credible. The overwhelming information always contains some false information that misleads people's cognition, thinking and behavior. This is Internet rumors, that is, false information.

Network platforms such as Twitter, WeChat, Weibo, and Tieba are full of false information, followed by a large number of forwarding and replying. Although there is an algorithm for identifying false information at present, it does not meet the timeliness requirements and cannot quickly identify false information.

Contents of the invention

In view of this, the purpose of the embodiment of the present application is to provide a

The false information detection method, system, computer equipment and readable storage medium are used to solve the problem of insufficient real-time detection of false information.

In order to achieve the above purpose, an embodiment of the present application provides a false information detection method, including:

Acquiring data to be detected, wherein the data to be detected includes detection source information and response information to be detected corresponding to the source information to be detected at the current moment;

Performing vectorization processing on the data to be detected to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected;

Through the encoding layer in the pre-trained false information classification model, the source feature vector to be detected and the reply feature vector to be detected are encoded, and the source feature code to be detected corresponding to the source information to be detected and the source feature code to be detected are obtained. The feature code of the response to be detected corresponding to the response to be detected;

Perform classification pre-judgment on the source feature code to be detected and the reply feature code to be detected through the trained deep reinforcement learning model to determine whether the data to be detected needs to be classified; and

If it is determined that the data to be detected needs to be classified, the classification layer of the false information classification model is used to classify the data to be detected according to the source feature code to be detected and the reply feature code to be detected, A first classification result corresponding to the data to be detected is obtained.

In order to achieve the above purpose, an embodiment of the present application provides a false information detection system, including:

An acquisition module, configured to acquire data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment;

A vectorization module, configured to perform vectorization processing on the data to be detected, to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected;

An encoding module, configured to encode the feature vector of the source to be detected and the feature vector of the reply to be detected through the encoding layer in the pre-trained false information classification model, to obtain the source to be detected corresponding to the information of the source to be detected A feature code and a feature code of the reply to be detected corresponding to the reply information to be detected;

The judging module is used to classify and pre-judge the source feature code to be detected and the reply feature code to be detected through the trained deep reinforcement learning model, so as to determine whether the data to be detected needs to be classified; and

A classification module, configured to classify the data to be detected according to the source feature code to be detected and the reply feature code to be detected through the classification layer of the false information classification model if it is determined that the data to be detected needs to be classified. The detection data is classified to obtain a first classification result corresponding to the data to be detected.

To achieve the above purpose, an embodiment of the present application provides a computer device, the computer device includes a memory and a processor, the memory stores computer-readable instructions that can run on the processor, and the processor The following steps are also performed when the computer readable instructions are executed:

Acquiring data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment;

In order to achieve the above object, an embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by at least one processor to causing the at least one processor to perform the following steps:

The false information detection method, system, computer equipment, and readable storage medium provided in the embodiments of the present application obtain the source information to be detected and the corresponding reply information to be detected at the current moment, and perform the detection on the source information to be detected and the corresponding reply information to be detected. Vectorization processing and encoding processing, after encoding and inputting the source information to be detected and its corresponding reply information to be detected into the pre-trained deep reinforcement learning model, whether to classify the source information to be detected and its corresponding reply information to be detected When performing classification processing, classify the source information to be detected and its corresponding reply information to be detected through the false information classification model to obtain the corresponding classification information, so as to realize the classification of the source information to be detected and its corresponding reply information to be detected Real-time detection, if it is determined that the data to be detected needs to be classified and processed, then the feature code of the data to be detected is classified and judged, and the reply information to be detected in the data to be detected is not obtained again for classification processing, which improves the efficiency of false information detection.

Description of drawings

FIG. 1 is a flow chart of Embodiment 1 of the false information detection method of the present application.

FIG. 2 is a schematic diagram of the modules of Embodiment 2 of the false information detection system of the present application.

FIG. 3 is a schematic diagram of the hardware structure of Embodiment 3 of the computer device of the present application.

Detailed ways

In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application, not to limit the present application. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the scope of protection of this application.

Embodiment one

Referring to FIG. 1 , it shows a flow chart of steps of a false information detection method according to Embodiment 1 of the present application. It can be understood that the flowchart in this method embodiment is not used to limit the sequence of execution steps. An exemplary description is given below taking the computer device 2 as the execution subject. details as follows.

Step S100. Obtain data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment.

Specifically, in order to achieve real-time detection of false information, real-time monitoring and acquisition of data to be detected, the source information to be detected is the source information that needs to be detected for false information, and the reply information to be detected is information that replies to the source information. Since the source information to be detected and the reply information to be detected are obtained in real time, the source information to be detected and the reply information to be detected have a time sequence.

Step S102 , performing vectorization processing on the data to be detected to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected.

Specifically, the data to be detected is vectorized through the preset vectorization model to obtain the source feature vector to be detected corresponding to the source information to be detected and the reply feature vector to be detected corresponding to the reply information to be detected. The vectorization model is visual (Bert +ImageNet) model.

Exemplarily, the step S102 includes:

Step S1021. Obtain first text data in the source information to be detected, and perform vectorization processing on the first text data through a first vectorization model to obtain a first feature vector. Step S1022. Obtain the first image data in the source information to be detected, and perform vectorization processing on the first image data through a second vectorization model to obtain a second feature vector. Step S1023, splicing the first feature vector and the second feature vector to obtain a source feature vector to be detected corresponding to the source information to be detected.

Specifically, when the source information to be detected includes text data and chart data, vectorization processing is performed through the first vectorization model and the second vectorization model, wherein the first vectorization model is a visual Bert model, and the second vectorization model is a visual Bert model. The vectorized model is a visual ImageNet model. The visual ImageNet model is a model for vectorizing images. This model is obtained after vectorization training on the ImageNet image library, and the accuracy of the model is more accurate. The processed first feature vector and the second feature vector are then concatenated to obtain the source feature vector to be detected corresponding to the source information to be detected.

Exemplarily, the step S102 includes:

Step S102A. Obtain second text data in the reply information to be detected, and perform vectorization processing on the second text data through the first vectorization model to obtain a third feature vector. Step S102B. Obtain the second picture data in the reply information to be detected, and perform vectorization processing on the second picture data through the second vectorization model to obtain a fourth feature vector. Step S102C, concatenating the third feature vector and the fourth feature vector to obtain a feature vector of a reply to be detected corresponding to the reply information to be detected.

Specifically, when the reply information to be detected includes text data and graph data, vectorization processing is performed through the first vectorization model and the second vectorization model, wherein the first vectorization model is a visual Bert model, and the second vectorization model is a visual Bert model. The vectorized model is the visual ImageNet model. The processed third feature vector and the fourth feature vector are then concatenated to obtain the feature vector of the reply to be detected corresponding to the reply information to be detected.

Step S104: Encoding the source feature vector to be detected and the reply feature vector to be detected through the encoding layer in the pre-trained false information classification model to obtain the source feature code to be detected corresponding to the source information to be detected and the feature code of the reply to be detected corresponding to the reply to be detected information.

Specifically, the false information classification model includes an encoding layer. The encoding layer is a structure of an LSTM (Long Short-Term Memory, long-short-term memory) neural network model, which can encode the source feature vector to be detected and the reply feature vector to be detected, and obtain each The state vector States _t of a time step = LSTM(feature _t ), each reply message is a time step t, and the state vector is used to describe the current environment. The input of the LSTM model is each reply text, and the output of the LSTM is the vector feature of the text, which represents the information of the text, where t=0 indicates the source news/twitter/microblog/post. Since the LSTM model has a memory function, the data to be detected at the current time and before the current time are stored in the LSTM model. If the reply information to be detected at the latest time is obtained, it only needs to encode the reply information to be detected at the latest time.

Step S106 , perform classification pre-judgment on the to-be-detected source feature code and the to-be-detected reply feature code through the trained deep reinforcement learning model, so as to determine whether to classify the to-be-detected data. The false information classification model is a neural network model.

Specifically, the deep reinforcement learning model is the Dueling-DQN (Deep Q Network) network model, which can predict the input source feature codes to be detected and the reply feature codes to be detected, and predict whether the input data to be detected at the current moment can be followed up. classification processing. Through the deep reinforcement learning model, the steps of subsequent data processing can be reduced, and the efficiency of false information identification can be improved. When the deep reinforcement learning model predicts that classification processing can be performed at the current moment, the false information classification model is started to classify the data to be detected. The Dueling-DQN network model includes state (state), reward (Reward), and behavior (Action). Q is Q(s, a) that is, in the state of s at a certain moment (s∈S), take action a(a ∈A) The expectation that the action can gain benefits, the environment will feedback the corresponding reward r according to the action of the agent, if the agent judges that the current state is enough for the classifier to classify with high confidence, then stop reading the reply and let the classifier do the classification; Otherwise continue reading the next reply. A certain amount of randomness is added to the action selection, that is, an action k=random(K) is randomly selected with a probability of ∈=1%.

Step S108, if it is determined that the data to be detected needs to be classified, the data to be detected is classified according to the source feature code to be detected and the reply feature code to be detected through the classification layer of the false information classification model Perform classification to obtain a first classification result corresponding to the data to be detected.

Specifically, the false information classification model is provided with a Classifier classifier, that is, a softmax loss function, which performs classification processing on the data to be detected to obtain a corresponding first classification result. The first classification result is used to indicate whether the data to be detected is false information. The classifier may be a binary classification classifier, and the first classification result output is 0 or 1, 0 indicating that it is not false information, and 1 indicating that it is false information. For false information and false information in news, tweets, Weibo, posts, WeChat articles, and group chats, it is possible to judge whether it is false information and false information in a timely and rapid manner before the spread of reply information increases, and allow users to act as soon as possible. Public opinion control measures such as deletion and blocking have been introduced.

Exemplarily, after the step S108, it also includes:

If it is determined that there is no need to classify the data to be detected, return to the step of obtaining the data to be detected.

Specifically, if it is judged that the classification process cannot be performed, the reply information to be detected at the next moment is obtained, and step S102 to step S106 are repeated to perform vectorization and encoding processing on the reply information to be detected at the next moment, and then input into the deep Strengthen the pre-judgment in the school model until it is determined that the data to be tested needs to be classified. When the deep enhanced school model performs classification pre-judgment, it is necessary to predict all the data to be tested. That is, the data to be detected includes the source information to be detected, the reply information to be detected before the current moment, the reply information to be detected at the current moment, and the reply information to be detected at the next moment; if it is still determined that the data to be detected does not need to be classified, Then obtain the reply data to be detected at the next moment.

Exemplarily, the training steps of the deep reinforcement learning model include:

Obtain multiple training sample sets, each training sample set includes the first feature code of the sample source information and the second feature code of the sample reply information corresponding to the sample source information at different time step values, wherein the sample source information corresponds to The time step value of is smaller than the time step value of the corresponding sample reply information. The first feature encoding and the second feature encoding in the training sample set are obtained by vectorizing and re-encoding the sample source information and the sample reply information. The second feature code corresponding to each time step value includes the first feature code corresponding to the sample source information and the feature codes of all sample reply information at the current moment, so that the enhanced model can be classified according to the first feature code or the second feature code preprocessing.

Input the first feature code and each second feature code in each training sample set into the preset reinforcement model in sequence according to the size of the time step, and judge whether to stop inputting the second feature code into the in the enhanced model. In order to better detect the sample source information and the sample reply information corresponding to the sample source information, the feature encoding corresponding to each time step value is classified. The enhanced model pre-judges the feature encoding of each time step. Based on the Q value judgment in the enhanced model Dueling-DQN, the larger the Q value, the greater the possibility of classification processing. When the Q value is greater than the preset threshold When , it means that the classification process is carried out. If the Q value calculated by the current enhanced model is less than the preset threshold, continue to obtain the feature code of the sample reply information, generate the next second feature code, and input it into the enhanced model for judgment.

If it is determined to stop inputting the second feature code into the enhanced model, then input the second feature code input into the enhanced model at the last moment into the false information classification model, so as to pass the false information classification The model outputs the second classification result of each training sample set. If the enhanced model judges to stop inputting the second feature code into the enhanced model, it means that the enhanced model judges that the current second feature code can be classified into false information, and inputs it into the false information classification model for classification processing. Wherein, if the enhanced model can determine that it is not necessary to read the second feature code of the sample reply information through the first feature code of the sample source information, then input the first feature code into the false information classification model to perform false information classification prediction, Get the second classification result.

Judging whether the second classification result is the same as the real classification result of each training sample set. The real classification results of the training sample set are obtained in advance and associated with the training sample set.

If not, update the reward and punishment value of the enhanced model to obtain a first updated reward and punishment value, and calculate the loss function of the enhanced model according to the first updated reward and punishment value to obtain a first updated function. The model parameters of the reinforcement model are updated according to the first update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.

Specifically, when the prediction result of the false information classification model is inaccurate, a penalty r _t = -100 is given to obtain the first updated reward and punishment value, and the loss function is calculated according to the first updated reward and punishment value

Perform backpropagation to update the network weights.

Exemplarily, after the judging whether the second classification result is the same as the true classification result of each training sample set, the method further includes:

If they are the same, update the reward and punishment value of the enhanced model to obtain a second updated reward and punishment value, and calculate the loss function of the enhanced model according to the second updated reward and punishment value to obtain a second update function; according to the second update The function updates the model parameters of the reinforcement model until the preset conditions are met, and a trained deep reinforcement learning model is obtained. If the output result of the current false information classifier Classifier is consistent with the actual label, a reward r _t =1+log(M) will be given to obtain the second updated reward and punishment value, where M represents the cumulative number of times the Agent has successfully obtained rewards (r>0) .

Exemplarily, after determining whether to stop inputting the second feature code into the enhanced model through the enhanced model, the method further includes:

If it is determined to continue to input the second feature code into the enhanced model, update the reward and punishment value of the enhanced model to obtain a third updated reward and punishment value, and calculate the loss of the enhanced model according to the third updated reward and punishment value function to obtain a third update function; update the model parameters of the enhanced model according to the third update function to obtain an updated enhanced model; encode the first feature code and each second feature code in each training sample set Input into the updated enhanced model in sequence according to the time step value, and judge whether to stop inputting the second feature code into the enhanced model through the updated enhanced model, until it is determined to stop inputting the second feature code to the enhanced model. If you choose to continue to read the reply, give a small penalty r _t =0.05 to get the third update reward and punishment value, so as to limit the Agent to keep increasing the reply.

Exemplarily, in order to fully understand the process of false information detection, the following examples are used to describe again:

The whole detection process is divided into two modules, one is the classification module (false information classification model), and the other is the control module (deep reinforcement learning model). For a source news/twitter/microblog/post (source information to be detected), several reply messages to be detected will be generated in chronological order, and each time a reply message to be detected is generated, the LSTM of the false information classification module will be used for detection The reply information is encoded, and after encoding, the encoded information will be input to the control module. The control module is a deep reinforcement learning model, which will judge the action of the input encoded information, and judge whether to stop or continue to obtain the next reply information to be detected. If it is judged that the action is to stop, the false information classification module will be triggered to classify the current status information (feature code) to determine whether it is false information; if the action is judged to be continued, the classification module will not be triggered to classify, thereby allowing the next The reply information is input into LSTM for encoding, and the latest encoded information is input to the control module to judge the action again, and so on. Because LSTM is a network with a cyclic neural network structure, it has the ability to encode historical information as a whole. When it is judged to continue, it only needs to obtain the next reply information to be detected. The deep reinforcement learning model will judge the action on the feature encoding of the overall information every time.

When training the deep reinforcement learning model, once the action is judged to be stopped, the classifier will be triggered to start classification, so the classification result of the classifier will be divided into right and wrong. If the classification is correct, the control module will be rewarded. If the classification is wrong, the control module is penalized. If the action is always judged as continuing, the control model will also get a penalty, but this penalty will be very small.

Embodiment two

Please continue to refer to FIG. 2 , which shows a schematic diagram of the program modules of Embodiment 2 of the false information detection system of the present application. In this embodiment, the false information detection system 20 may include or be divided into one or more program modules, and one or more program modules are stored in a storage medium and executed by one or more processors to complete In this application, the above false information detection method can be realized. The program module referred to in the embodiment of this application refers to a series of computer program instruction segments capable of completing specific functions, which is more suitable for describing the execution process of the false information detection system 20 in the storage medium than the program itself. The following description will specifically introduce the functions of each program module of the present embodiment:

The acquisition module 200 is configured to acquire data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment.

The vectorization module 202 is configured to perform vectorization processing on the data to be detected to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected.

Exemplarily, the vectorization module 202 is also used for:

Acquiring the first text data in the source information to be detected, and performing vectorization processing on the first text data through a first vectorization model to obtain a first feature vector; obtaining the first picture in the source information to be detected Data, the first image data is vectorized through the second vectorization model to obtain a second feature vector; the first feature vector and the second feature vector are spliced to obtain the source information to be detected The corresponding source feature vector to be detected.

Exemplarily, the vectorization module 202 is also used for:

Obtaining the second text data in the reply information to be detected, and performing vectorization processing on the second text data through the first vectorization model to obtain a third feature vector; obtaining the first text data in the reply information to be detected Two picture data, performing vectorization processing on the second picture data through the second vectorization model to obtain a fourth feature vector; splicing the third feature vector and the fourth feature vector to obtain the described The feature vector of the response to be detected corresponding to the reply information to be detected.

The encoding module 204 is configured to encode the source feature vector to be detected and the reply feature vector to be detected through the encoding layer in the pre-trained false information classification model to obtain the source information to be detected corresponding to the source information to be detected The source feature code and the feature code of the reply to be detected corresponding to the reply information to be detected.

The judging module 206 is configured to perform classification pre-judgment on the to-be-detected source feature code and the to-be-detected reply feature code through the trained deep reinforcement learning model, so as to determine whether to classify the to-be-detected data.

The classification module 208 is configured to classify the data to be detected according to the source feature code to be detected and the reply feature code to be detected through the classification layer of the false information classification model if it is determined that the data to be detected needs to be classified. The data to be detected is classified to obtain a first classification result corresponding to the data to be detected.

Exemplarily, the classification module 208 is also used for:

Embodiment three

Referring to FIG. 3 , it is a schematic diagram of a hardware architecture of a computer device according to Embodiment 3 of the present application. In this embodiment, the computer device 2 is a device capable of automatically performing numerical calculation and/or information processing according to preset or stored instructions. The computer device 2 may be a rack server, a blade server, a tower server or a cabinet server (including an independent server, or a server cluster composed of multiple servers) and the like. The server can be an independent server, or it can provide cloud services, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, content delivery network (Content Delivery Network, CDN), and cloud servers for basic cloud computing services such as big data and artificial intelligence platforms. As shown in FIG. 3 , the computer device 2 at least includes, but is not limited to, a memory 21 , a processor 22 , a network interface 23 , and a false information detection system 20 that can communicate with each other through a system bus. in:

In this embodiment, the memory 21 includes at least one type of computer-readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory ( RAM), static random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 21 may be an internal storage unit of the computer device 2 , such as a hard disk or memory of the computer device 2 . In other embodiments, the memory 21 can also be an external storage device of the computer device 2, such as a plug-in hard disk equipped on the computer device 2, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, flash memory card (Flash Card), etc. Of course, the storage 21 may also include both the internal storage unit of the computer device 2 and its external storage device. In this embodiment, the memory 21 is generally used to store the operating system and various application software installed in the computer device 2, such as the program code of the false information detection system 20 in the second embodiment. In addition, the memory 21 can also be used to temporarily store various types of data that have been output or will be output.

In some embodiments, the processor 22 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 22 is generally used to control the overall operation of the computer device 2 . In this embodiment, the processor 22 is used to run the program codes stored in the memory 21 or process data, for example, to run the false information detection system 20, so as to realize the false information detection method in the first embodiment.

The network interface 23 may include a wireless network interface or a wired network interface, and the network interface 23 is generally used to establish a communication connection between the server 2 and other electronic devices. For example, the network interface 23 is used to connect the server 2 with an external terminal through a network, and establish a data transmission channel and a communication connection between the server 2 and an external terminal. The network can be an enterprise intranet (Intranet), Internet (Internet), Global System of Mobile communication (Global System of Mobile communication, GSM), wideband code division multiple access (Wideband Code Division Multiple Access, WCDMA), 4G network, 5G Internet, Bluetooth (Bluetooth), Wi-Fi and other wireless or wired networks.

It should be noted that FIG. 3 only shows computer device 2 having components 20-23, but it should be understood that implementation of all of the illustrated components is not required and that more or fewer components may instead be implemented.

In this embodiment, the false information detection system 20 stored in the memory 21 can also be divided into one or more program modules, and the one or more program modules are stored in the memory 21 and are controlled by one or more program modules. Executed by multiple processors (the processor 22 in this embodiment) to complete the application.

For example, FIG. 2 shows a schematic diagram of program modules for realizing the second embodiment of the false information detection system 20. In this embodiment, the false information detection system 20 can be divided into the acquisition module 200, the vectorization module 202 , the encoding module 204 , the judging module 206 and the classifying module 208 . Wherein, the program module referred to in this application refers to a series of computer program instruction segments capable of completing specific functions, which is more suitable than a program to describe the execution process of the false information detection system 20 in the computer device 2 . The specific functions of the program modules 200-208 have been described in detail in the second embodiment, and will not be repeated here.

Embodiment four

This embodiment also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), only Read memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, server, App application store, etc., on which computer programs are stored, The corresponding functions are realized when the program is executed by the processor. The computer-readable storage medium may be non-volatile or volatile. The computer-readable storage medium of this embodiment is used for a computer program, and when executed by a processor, the following steps are implemented:

The serial numbers of the above embodiments of the present application are for description only, and do not represent the advantages and disadvantages of the embodiments.

Through the description of the above embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus a necessary general-purpose hardware platform, and of course also by hardware, but in many cases the former is better implementation.

The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structures or equivalent process transformations made by using the description of the application and the accompanying drawings are directly or indirectly used in other related technical fields. , are all included in the patent protection scope of the present application in the same way.

Claims

A false information detection method, including:

Acquiring data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment;

Performing vectorization processing on the data to be detected to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected;

Through the encoding layer in the pre-trained false information classification model, the source feature vector to be detected and the reply feature vector to be detected are encoded, and the source feature code to be detected corresponding to the source information to be detected and the source feature code to be detected are obtained. The feature code of the response to be detected corresponding to the response to be detected;

Perform classification pre-judgment on the source feature code to be detected and the reply feature code to be detected through the trained deep reinforcement learning model to determine whether the data to be detected needs to be classified; and

If it is determined that the data to be detected needs to be classified, the classification layer of the false information classification model is used to classify the data to be detected according to the source feature code to be detected and the reply feature code to be detected, A first classification result corresponding to the data to be detected is obtained.
The false information detection method according to claim 1, wherein, the trained deep reinforcement learning model performs classification pre-judgment on the source feature codes to be detected and the reply feature codes to be detected, so as to determine whether the After the data to be detected is classified and processed, it also includes:

If it is determined that there is no need to classify the data to be detected, return to the step of obtaining the data to be detected.
The method for detecting false information according to claim 1, wherein the vectorization processing is performed on the data to be detected to obtain the source feature vector to be detected corresponding to the source information to be detected and the eigenvector corresponding to the reply information to be detected Reply feature vectors to be detected include:

Acquiring the first text data and the first image data in the source information to be detected;

performing vectorization processing on the first text data through a first vectorization model to obtain a first feature vector; and performing vectorization processing on the first image data through a second vectorization model to obtain a second feature vector;

splicing the first feature vector and the second feature vector to obtain a source feature vector to be detected corresponding to the source information to be detected;

Acquiring the second text data and the second image data in the reply information to be detected;

performing vectorization processing on the second text data through the first vectorization model to obtain a third feature vector; and performing vectorization processing on the second picture data through the second vectorization model to obtain a third feature vector Four eigenvectors; and

The third feature vector and the fourth feature vector are spliced to obtain a feature vector of a reply to be detected corresponding to the reply information to be detected.
The false information detection method according to claim 3, wherein the false information classification model is a neural network model.
The false information detection method according to claim 1, wherein, the training step of the deep reinforcement learning model comprises:

Obtain multiple training sample sets, each training sample set includes the first feature code of the sample source information and the second feature code of the sample reply information corresponding to the sample source information at different time step values, wherein the sample source information corresponds to The time step value of is less than the time step value of the corresponding sample reply information;

Input the first feature code and each second feature code in each training sample set into the preset reinforcement model in sequence according to the size of the time step, and judge whether to stop inputting the second feature code into the In the reinforcement model;

If it is determined to stop inputting the second feature code into the enhanced model, then input the second feature code input into the enhanced model at the last moment into the false information classification model, so as to pass the false information classification The model outputs the second classification result of each training sample set;

judging whether the second classification result is the same as the true classification result of each training sample set;

If not, update the reward and punishment value of the enhanced model to obtain a first updated reward and punishment value, and calculate the loss function of the enhanced model according to the first updated reward and punishment value to obtain a first update function; and

The model parameters of the reinforcement model are updated according to the first update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.
The method for detecting false information according to claim 5, wherein, after determining whether the second classification result is the same as the true classification result of each training sample set, the method further comprises:

If they are the same, update the reward and punishment value of the enhanced model to obtain a second updated reward and punishment value, and calculate the loss function of the enhanced model according to the second updated reward and punishment value to obtain a second update function; and

The model parameters of the reinforcement model are updated according to the second update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.
The false information detection method according to claim 5, wherein, after determining whether to stop inputting the second feature code into the enhanced model through the enhanced model, the method further comprises:

If it is determined to continue to input the second feature code into the enhanced model, update the reward and punishment value of the enhanced model to obtain a third updated reward and punishment value, and calculate the loss of the enhanced model according to the third updated reward and punishment value function to get the third update function;

updating model parameters of the enhanced model according to the third update function to obtain an updated enhanced model; and

Input the first feature code and each second feature code in each training sample set into the updated enhanced model in sequence according to the size of the time step, and judge whether to stop inputting the second feature code through the updated enhanced model into the enhanced model until it is determined to stop inputting the second feature code into the enhanced model.
A false information detection system, including:

An acquisition module, configured to acquire data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment;

A vectorization module, configured to perform vectorization processing on the data to be detected, to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected;

An encoding module, configured to encode the feature vector of the source to be detected and the feature vector of the reply to be detected through the encoding layer in the pre-trained false information classification model, to obtain the source to be detected corresponding to the information of the source to be detected A feature code and a feature code of the reply to be detected corresponding to the reply information to be detected;

The judging module is used to classify and pre-judge the source feature code to be detected and the reply feature code to be detected through the trained deep reinforcement learning model, so as to determine whether the data to be detected needs to be classified; and

A classification module, configured to classify the data to be detected according to the source feature code to be detected and the reply feature code to be detected through the classification layer of the false information classification model if it is determined that the data to be detected needs to be classified. The detection data is classified to obtain a first classification result corresponding to the data to be detected.
A computer device, wherein the computer device includes a memory and a processor, the memory stores computer-readable instructions operable on the processor, and the processor also executes the computer-readable instructions Perform the following steps:

Acquiring data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment;

Performing vectorization processing on the data to be detected to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected;

Through the encoding layer in the pre-trained false information classification model, the source feature vector to be detected and the reply feature vector to be detected are encoded, and the source feature code to be detected corresponding to the source information to be detected and the source feature code to be detected are obtained. The feature code of the response to be detected corresponding to the response to be detected;

Perform classification pre-judgment on the source feature code to be detected and the reply feature code to be detected by the trained deep reinforcement learning model to determine whether the data to be detected needs to be classified; and

If it is determined that the data to be detected needs to be classified, the classification layer of the false information classification model is used to classify the data to be detected according to the source feature code to be detected and the reply feature code to be detected, A first classification result corresponding to the data to be detected is obtained.
The computer device according to claim 9, wherein said processor further performs the following steps when executing said computer readable instructions:

If it is determined that there is no need to classify the data to be detected, return to the step of obtaining the data to be detected.
The computer device according to claim 9, wherein said processor further performs the following steps when executing said computer readable instructions:

Acquiring the first text data and the first image data in the source information to be detected;

performing vectorization processing on the first text data through a first vectorization model to obtain a first feature vector; and performing vectorization processing on the first image data through a second vectorization model to obtain a second feature vector;

splicing the first feature vector and the second feature vector to obtain a source feature vector to be detected corresponding to the source information to be detected;

Acquiring the second text data and the second image data in the reply information to be detected;

performing vectorization processing on the second text data through the first vectorization model to obtain a third feature vector; and performing vectorization processing on the second picture data through the second vectorization model to obtain a third feature vector Four eigenvectors; and

The third feature vector and the fourth feature vector are spliced to obtain a feature vector of a reply to be detected corresponding to the reply information to be detected.
The computer device according to claim 11, wherein said processor further performs the following steps when executing said computer readable instructions:

Obtain multiple training sample sets, each training sample set includes the first feature code of the sample source information and the second feature code of the sample reply information corresponding to the sample source information at different time step values, wherein the sample source information corresponds to The time step value of is less than the time step value of the corresponding sample reply information;

Input the first feature code and each second feature code in each training sample set into the preset reinforcement model in sequence according to the size of the time step, and judge whether to stop inputting the second feature code into the In the reinforcement model;

If it is determined to stop inputting the second feature code into the enhanced model, then input the second feature code input into the enhanced model at the last moment into the false information classification model, so as to pass the false information classification The model outputs the second classification result of each training sample set;

judging whether the second classification result is the same as the true classification result of each training sample set;

If not, update the reward and punishment value of the enhanced model to obtain a first updated reward and punishment value, and calculate the loss function of the enhanced model according to the first updated reward and punishment value to obtain a first update function; and

The model parameters of the reinforcement model are updated according to the first update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.
The computer device according to claim 12, wherein said processor further performs the following steps when executing said computer readable instructions:

If they are the same, update the reward and punishment value of the enhanced model to obtain a second updated reward and punishment value, and calculate the loss function of the enhanced model according to the second updated reward and punishment value to obtain a second update function; and

The model parameters of the reinforcement model are updated according to the second update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.
The computer device according to claim 13, wherein said processor, when executing said computer readable instructions, further performs the following steps:

If it is determined to continue to input the second feature code into the enhanced model, update the reward and punishment value of the enhanced model to obtain a third updated reward and punishment value, and calculate the loss of the enhanced model according to the third updated reward and punishment value function to get the third update function;

updating model parameters of the enhanced model according to the third update function to obtain an updated enhanced model; and

Input the first feature code and each second feature code in each training sample set into the updated enhanced model in sequence according to the size of the time step, and judge whether to stop inputting the second feature code through the updated enhanced model into the enhanced model until it is determined to stop inputting the second feature code into the enhanced model.
A computer-readable storage medium, wherein computer-readable instructions are stored in the computer-readable storage medium, and the computer-readable instructions can be executed by at least one processor, so that the at least one processor performs the following step:

Acquiring data to be detected, wherein the data to be detected includes source information to be detected and reply information to be detected corresponding to the source information to be detected at the current moment;

Performing vectorization processing on the data to be detected to obtain a source feature vector to be detected corresponding to the source information to be detected and a reply feature vector to be detected corresponding to the reply information to be detected;

Through the encoding layer in the pre-trained false information classification model, the source feature vector to be detected and the reply feature vector to be detected are encoded, and the source feature code to be detected corresponding to the source information to be detected and the source feature code to be detected are obtained. The feature code of the response to be detected corresponding to the response to be detected;

Perform classification pre-judgment on the source feature code to be detected and the reply feature code to be detected through the trained deep reinforcement learning model to determine whether the data to be detected needs to be classified; and

If it is determined that the data to be detected needs to be classified, the classification layer of the false information classification model is used to classify the data to be detected according to the source feature code to be detected and the reply feature code to be detected, A first classification result corresponding to the data to be detected is obtained.
The computer-readable storage medium of claim 15, wherein the computer-readable instructions are executable by at least one processor, such that the at least one processor further performs the steps of:

If it is determined that there is no need to classify the data to be detected, return to the step of obtaining the data to be detected.
The computer-readable storage medium of claim 15, wherein the computer-readable instructions are executable by at least one processor, such that the at least one processor further performs the steps of:

Acquiring the first text data and the first image data in the source information to be detected;

performing vectorization processing on the first text data through a first vectorization model to obtain a first feature vector; and performing vectorization processing on the first image data through a second vectorization model to obtain a second feature vector;

splicing the first feature vector and the second feature vector to obtain a source feature vector to be detected corresponding to the source information to be detected;

Acquiring the second text data and the second image data in the reply information to be detected;

performing vectorization processing on the second text data through the first vectorization model to obtain a third feature vector; and performing vectorization processing on the second picture data through the second vectorization model to obtain a third feature vector Four eigenvectors; and

The third feature vector and the fourth feature vector are spliced to obtain a feature vector of a reply to be detected corresponding to the reply information to be detected.
The computer-readable storage medium of claim 17, wherein the computer-readable instructions are executable by at least one processor, such that the at least one processor further performs the steps of:

Obtain multiple training sample sets, each training sample set includes the first feature code of the sample source information and the second feature code of the sample reply information corresponding to the sample source information at different time step values, wherein the sample source information corresponds to The time step value of is less than the time step value of the corresponding sample reply information;

Input the first feature code and each second feature code in each training sample set into the preset reinforcement model in sequence according to the size of the time step, and judge whether to stop inputting the second feature code into the In the reinforcement model;

If it is determined to stop inputting the second feature code into the enhanced model, then input the second feature code input into the enhanced model at the last moment into the false information classification model, so as to pass the false information classification The model outputs the second classification result of each training sample set;

judging whether the second classification result is the same as the true classification result of each training sample set;

If not, update the reward and punishment value of the enhanced model to obtain a first updated reward and punishment value, and calculate the loss function of the enhanced model according to the first updated reward and punishment value to obtain a first update function; and

The model parameters of the reinforcement model are updated according to the first update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.
The computer-readable storage medium according to claim 18, wherein, if they are the same, update the reward and punishment value of the enhanced model to obtain a second updated reward and punishment value, and calculate the reward and punishment value of the enhanced model according to the second updated reward and punishment value A loss function to obtain a second update function; and

The model parameters of the reinforcement model are updated according to the second update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.
The computer-readable storage medium of claim 18, wherein the computer-readable instructions are executable by at least one processor, such that the at least one processor further performs the steps of:

If they are the same, update the reward and punishment value of the enhanced model to obtain a second updated reward and punishment value, and calculate the loss function of the enhanced model according to the second updated reward and punishment value to obtain a second update function; and

The model parameters of the reinforcement model are updated according to the second update function until a preset condition is met, and a trained deep reinforcement learning model is obtained.