A false news detection method and device based on a neural network model and a medium

Abstract

The application discloses a false news detection method and device based on a neural network model and a medium, and aims to solve the technical problem that the prior art cannot detect false news in time. The method comprises the following steps: generating a news data set and preprocessing the news data set to obtain a news data set in a unified format; performing feature extraction on news titles and news texts through a term frequency-inverse document frequency algorithm; performing denoising on a feature vector through an elbow rule to obtain a plurality of feature vectors and calculate the cosine similarity between the plurality of feature vectors; dividing the plurality of feature vectors into a plurality of feature vector sets and inputting the feature vectors into a pre-trained neural network model to output a detection result of whether the feature vectors corresponding to the news titles and the news texts are consistent and determine whether the news is false; keeping the detection results of other news corresponding to the feature vectors in the feature vector set where the feature vectors corresponding to the news titles and the news texts are located consistent with the detection result of the news, and realizing the detection of false news.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent detection, in particular to a false news detection method and device based on a neural network model and a medium. BACKGROUND

[0002] In recent years, with the rapid development of network information technology and the iterative update of mobile terminals such as mobile phones and tablets, people tend to obtain the latest news from social media and news software. Such an acquisition method is more convenient and fast, and can be widely spread by forwarding at hand. However, due to the low cost and wide spread of news acquisition through social media and news software, the authenticity of news messages is questioned, and some news may be false every day. These false news messages often have huge traffic and attract people's attention, but will have a negative impact on society and the public and cause unnecessary trouble. SUMMARY

[0003] The embodiments of the present application provide a false news detection method, device and medium based on a neural network model, to solve the technical problem that the prior art cannot automatically determine whether the news title and the news body are consistent, and timely detect false news.

[0004] In one aspect, the embodiments of the present application provide a false news detection method based on a neural network model, comprising:

[0005] Obtain a plurality of news data to generate a corresponding news data set, and preprocess the news data in the news data set to obtain a news data set in a unified format;

[0006] Perform feature extraction on the news title and the news body in the news data set in the unified format through a term frequency-inverse document frequency algorithm;

[0007] Perform denoising on the feature vectors corresponding to the news title and the news body of the news data set by elbow rule to obtain a plurality of feature vectors corresponding to the news title and the news body of the news data set, and calculate the cosine similarity between the plurality of feature vectors;

[0008] According to the cosine similarity between the plurality of feature vectors, divide the plurality of feature vectors into a plurality of feature vector sets, and input the feature vectors corresponding to the news title and the news body in the plurality of feature vector sets into a pre-trained neural network model respectively; the neural network model is obtained based on the combination of a convolutional neural network and a deep neural network;

[0009] output, by the pre-trained neural network model, a detection result of whether the feature vectors corresponding to the news title and the news body match, and determine, according to the detection result, whether the news corresponding to the news title and the news body is fake news;

[0010] maintain the detection result corresponding to the other news in the feature vector set in which the feature vectors corresponding to the news title and the news body are located consistent with the detection result corresponding to the news, to realize detection of fake news.

[0011] In an implementation manner of the present application, the method further comprises:

[0012] obtain a plurality of news data from a network or an application by a preset manner to generate a news data set corresponding to the plurality of news data, and determine a news title and a news body in the news data set;

[0013] delete irrelevant words in the news data set for determining whether the news title and the news body match, to realize cleaning of the news title and the news body in the news data set; the irrelevant words at least include one or more of the following: punctuation, stop words and mood words;

[0014] obtain the cleaned news title and the news body, and perform format conversion on the cleaned news title and the news body to obtain a news data set in a unified format.

[0015] In an implementation manner of the present application, before the method further comprises:

[0016] input a sample news pre-determined as fake news into a convolutional neural network for training;

[0017] extract attribute features in the sample news by a convolution layer of the convolutional neural network; the attribute features include a news title and a news body;

[0018] output, by a full connection layer of the convolutional neural network, a detection result of whether the sample news is fake news, under the condition that the attribute features are pooled by a pooling layer, until the detection result of whether the sample news is fake news matches a result pre-determined that the sample news is fake news, to obtain a neural network model.

[0019] In an implementation manner of the present application, after the neural network model is obtained, the method further comprises:

[0020] adjusting, by a deep neural network, weights corresponding to each attribute feature in the sample news to obtain target attribute feature weights in the neural network model;

[0021] optimizing the neural network model based on the target attribute feature weights to complete training of the neural network model.

[0022] In an implementation manner of the present application, the feature extraction is performed on the news title and the news body in the news data set in the unified format by using the term frequency-inverse document frequency algorithm, and specifically includes:

[0023] The news title and the corresponding news body in the news data set in the unified format are determined, and the news title and the corresponding news body are encoded by using the term frequency-inverse document frequency algorithm to generate an encoding list corresponding to the news title and an encoding list corresponding to the news body;

[0024] The feature extraction is performed on the news title according to the encoding list corresponding to the news title, and the feature extraction is performed on the news body according to the encoding list corresponding to the news body.

[0025] In an implementation manner of the present application, the calculation of the cosine similarity between the plurality of feature vectors specifically includes:

[0026] The feature vector corresponding to the news title and the feature vector of the news body corresponding to the news title are obtained from the plurality of feature vectors corresponding to the news title and the news body in the news data set;

[0027] The cosine similarity between the feature vector corresponding to the news title and the feature vector of the news body corresponding to the news title and the feature vector corresponding to other news titles and the feature vector corresponding to other news bodies in the news data set is calculated based on a cosine similarity calculation formula.

[0028] In an implementation manner of the present application, the division of the plurality of feature vectors into a plurality of feature vector sets according to the cosine similarity between the plurality of feature vectors specifically includes:

[0029] The calculated cosine similarity is compared with a preset similarity threshold, and other news with a cosine similarity greater than the preset similarity threshold is determined;

[0030] The determined other news and the news corresponding to the news title and the news body are stored in the same feature vector set.

[0031] In an implementation form of the application, the detection result of other news corresponding to the feature vector set in which the news title and news body corresponding feature vectors are located is consistent with the detection result of the news, so as to realize the detection of fake news, and specifically includes:

[0032] If the detection result of the news is fake news, it is determined that other news in the feature vector set in which the news title and news body corresponding feature vectors are located is fake news.

[0033] If the detection result of the news is not fake news, it is determined that other news in the feature vector set in which the news title and news body corresponding feature vectors are located is not fake news, so as to realize the detection of fake news in the news data set.

[0034] On the other hand, the embodiments of the application also provide a fake news detection device based on a neural network model, and the device comprises:

[0035] At least one processor;

[0036] and a memory in communication connection with the at least one processor;

[0037] Wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the above-mentioned fake news detection method based on a neural network model.

[0038] On the other hand, the embodiments of the application also provide a non-volatile computer storage medium, which stores computer executable instructions, and the computer executable instructions are configured to:

[0039] The above-mentioned fake news detection method based on a neural network model.

[0040] The embodiments of the application provide a fake news detection method, device and medium based on a neural network model, which at least include the following beneficial effects:

[0041] The irrelevant data in the news data set is deleted through preprocessing of the news data set, so as to subsequent data processing; the cosine similarity between the plurality of feature vectors is calculated by extracting features from the news title and the news body in the news data set and denoising the extracted features, so that the plurality of feature vectors can be divided into a plurality of feature vector sets according to the calculated cosine similarity; the detection result of whether the news title and the news body are consistent can be output by inputting the feature vectors corresponding to the news title and the news body in the plurality of feature vector sets into the neural network model trained in advance, and whether the news corresponding to the news title and the news body is false news can be determined according to the detection result; by setting the detection results corresponding to other news in the feature vector set where the feature vectors corresponding to the news title and the news body are located to be consistent with the detection result corresponding to the news, whether the news title and the corresponding news body in the same feature vector set are false information can be detected by detecting whether the feature vectors corresponding to the news title and the news body in the feature vector set are false information, and the detection efficiency of false news is improved. BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and serve to explain the principles of the application. In the drawings:

[0043] Figure 1 A flowchart of a false news detection method based on a neural network model provided by an embodiment of the application;

[0044] Figure 2 An internal structure diagram of a false news detection device based on a neural network model provided by an embodiment of the application. DETAILED DESCRIPTION

[0045] To make the objectives, technical solutions and advantages of the application clearer, the technical solutions of the application will be described below in conjunction with specific embodiments of the application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the application.

[0046] The embodiment of the application provides a false news detection method, device and medium based on a neural network model. The method comprises the following steps: preprocessing news data sets, deleting irrelevant data in the news data sets, and then processing subsequent data; extracting features of news titles and news texts in the news data sets, denoising the extracted features, obtaining a plurality of processed feature vectors, and then calculating cosine similarities between the plurality of feature vectors, so as to divide the plurality of feature vectors into a plurality of feature vector sets according to the calculated cosine similarities; inputting feature vectors corresponding to the news titles and the news texts in the plurality of feature vector sets into a pre-trained neural network model, outputting a detection result of whether the news titles and the news texts are consistent, and determining whether the news titles and the news texts corresponding to the news are false news according to the detection result; and setting detection results corresponding to other news in the feature vector set in which the feature vectors corresponding to the news titles and the news texts are located to be consistent with the detection result corresponding to the news, so as to improve the detection efficiency of the false news. The technical problem that the prior art cannot automatically determine whether the news titles and the news texts are consistent and timely detect false news is solved.

[0047] Figure 1 A flowchart of a false news detection method based on a neural network model is provided in the embodiment of the application. As shown in the figure, Figure 1 The false news detection method based on the neural network model provided in the embodiment of the application can mainly comprise the following steps:

[0048] 101. Obtain a plurality of news data to generate a corresponding news data set, and preprocess news data in the news data set to obtain a news data set in a unified format.

[0049] The server obtains a plurality of news data from social media or news software, stores the plurality of news data in the same set to generate a corresponding news data set, and then needs to preprocess the news data in the news data set, so as to obtain a news data set in a unified format.

[0050] Specifically, the server obtains a plurality of news data from the network or the application through a preset mode, generates a news data set corresponding to the plurality of news data, determines news titles and news texts in the news data set, and deletes irrelevant words in the news data set to determine whether the news titles and the news texts are consistent, so as to clean the news titles and the news texts in the news data set. It should be noted that the irrelevant words in the embodiment of the application at least include one or more of the following: punctuation marks, stop words and mood words.

[0051] The server obtains the cleaned news title and news body, and converts the format of the cleaned news title and news body, so as to obtain a news data set in a unified format, so as to process the news data in the news data set subsequently.

[0052] 102. Feature extraction is performed on the news title and news body in the news data set in the unified format by using the term frequency-inverse document frequency algorithm.

[0053] The server performs feature extraction on the news title and news body in the news data set in the unified format by using the term frequency-inverse document frequency algorithm. However, not all the extracted features are related to the judgment of whether the news is fake news, so subsequent processing of the extracted features is required.

[0054] Specifically, the server needs to determine the news title and the news body corresponding to the news title from the news data set in the unified format, and encode the news title and the news body corresponding to the news title by using the term frequency-inverse document frequency algorithm, so as to generate a code list corresponding to the news title and a code list corresponding to the news body, and then perform feature extraction on the news title according to the code list corresponding to the news title, and perform feature extraction on the news body according to the code list corresponding to the news body.

[0055] 103. The elbow method is used to denoise the feature vectors corresponding to the news title and the news body of the news data set, so as to obtain a plurality of feature vectors corresponding to the news title and the news body of the news data set, and calculate the cosine similarity between the plurality of feature vectors.

[0056] The server deletes the irrelevant feature vectors in the extracted feature vectors by using the elbow method to denoise the feature vectors corresponding to the news title and the news body of the news data set, so as to obtain a plurality of feature vectors corresponding to the news title and the news body of the news data set, and then calculate the cosine similarity between the plurality of feature vectors.

[0057] Specifically, when calculating the cosine similarity between the plurality of feature vectors, the server obtains the feature vector corresponding to the news title and the feature vector corresponding to the news body corresponding to the news title from the plurality of feature vectors corresponding to the news title and the news body of the news data set, and then calculates the cosine similarity between the feature vector corresponding to the news title and the feature vector corresponding to the news body corresponding to the news title and the feature vectors corresponding to other news titles and the feature vectors corresponding to other news bodies in the news data set based on the cosine similarity calculation formula.

[0058] 104. Based on the cosine similarity between several feature vectors, divide the several feature vectors into multiple feature vector sets, and input the feature vectors corresponding to the news headlines and news texts in the multiple feature vector sets into the pre-trained neural network model.

[0059] Based on the cosine similarity between several feature vectors, the server can find similar news headlines and news articles. It can then store the feature vectors corresponding to the similar news headlines and news articles in the same feature vector set, and input the news headlines and corresponding news articles from multiple feature vector sets into a pre-trained neural network model. The neural network model can then determine whether the news headlines and corresponding news articles match.

[0060] Specifically, the server compares the calculated cosine similarity with a preset similarity threshold, and determines other news items whose cosine similarity is greater than the preset similarity threshold based on the comparison results. Then, the determined other news items are stored in the same feature vector set as the news items corresponding to the news titles and news articles.

[0061] In one embodiment of this application, before inputting the feature vectors corresponding to news titles and news texts from multiple feature vector sets into a pre-trained neural network model, the server inputs sample news pre-determined as fake news into the convolutional neural network for training, and extracts attribute features from the sample news through the convolutional layers of the convolutional neural network. It should be noted that the attribute features in this embodiment include: news titles and news texts.

[0062] After the attribute features are pooled by the pooling layer, the server outputs the detection result of whether the sample news is fake news through the fully connected layer of the convolutional neural network, until the output detection result of whether the sample news is fake news matches the pre-determined result of whether the sample news is fake news, and the neural network model is obtained.

[0063] In one embodiment of this application, after the server obtains the neural network model, it adjusts the weights corresponding to each attribute feature in the sample news through a deep neural network, thereby obtaining the target attribute feature weights in the neural network model. Then, based on the target attribute feature weights, the server optimizes the neural network model and completes the training of the neural network model.

[0064] 105. Using a pre-trained neural network model, output the detection results of whether the feature vectors corresponding to the news headline and the news body match, and determine whether the news headline and the news body are fake news based on the detection results.

[0065] After the server inputs the feature vectors corresponding to the news title and the news body into a pre-trained neural network model, the pre-trained neural network model will output the detection result of whether the feature vectors corresponding to the news title and the news body match. Then, the server can determine whether the news corresponding to the news title and the news body is fake news based on the detection result.

[0066] 106. Ensure that the detection results of other news items in the feature vector set corresponding to the news headlines and news texts are consistent with the detection results of the news articles, thereby achieving the detection of fake news.

[0067] After determining whether the news corresponding to the news title and the news body is fake news, the server sets the detection results of other news in the feature vector set where the feature vectors corresponding to the news title and the news body are located to be consistent with the detection results of the news, thereby realizing the detection of fake news and improving the detection efficiency of fake news.

[0068] Specifically, when the server determines that the detection result for a news article is fake news, it can also determine that other news articles in the feature vector set containing the feature vectors corresponding to the news title and the news body are also fake news. Conversely, when the server determines that the detection result for a news article is not fake news, it can also determine that other news articles in the feature vector set containing the feature vectors corresponding to the news title and the news body are not fake news. Thus, by detecting whether the feature vectors corresponding to a news title and the news body in the same feature vector set match, the server can detect whether other news articles in the news dataset are fake news, saving resources that would otherwise be used to detect other news articles in the same feature vector set.

[0069] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a fake news detection device based on a neural network model, the structure of which is as follows: Figure 2 As shown.

[0070] Figure 2 This is a schematic diagram of the internal structure of a fake news detection device based on a neural network model, provided as an embodiment of this application. Figure 2 As shown, the device includes:

[0071] At least one processor;

[0072] And, a memory that is communicatively connected to at least one processor;

[0073] The memory stores instructions that can be executed by at least one processor, and the instructions, when executed by at least one processor, enable at least one processor to:

[0074] Obtaining a plurality of news data to generate a corresponding news data set, and preprocessing news data in the news data set to obtain a news data set in a unified format;

[0075] Extracting features of news titles and news texts in the news data set in the unified format by a term frequency-inverse document frequency algorithm;

[0076] Denoising feature vectors corresponding to the news titles and the news texts of the news data set by an elbow rule to obtain a plurality of feature vectors corresponding to the news titles and the news texts of the news data set, and calculating cosine similarities between the plurality of feature vectors;

[0077] According to the cosine similarities between the plurality of feature vectors, dividing the plurality of feature vectors into a plurality of feature vector sets, and inputting feature vectors corresponding to the news titles and the news texts in the plurality of feature vector sets into a pre-trained neural network model respectively; the neural network model is obtained based on a combination of a convolutional neural network and a deep neural network;

[0078] Outputting, by the pre-trained neural network model, a detection result of whether the feature vectors corresponding to the news titles and the news texts are consistent, and determining, according to the detection result, whether news corresponding to the news titles and the news texts is fake news;

[0079] Keeping detection results corresponding to other news in a feature vector set in which the feature vectors corresponding to the news titles and the news texts are located consistent with the detection result corresponding to the news, to realize detection of fake news.

[0080] The embodiment of the application also provides a non-volatile computer storage medium, which stores computer executable instructions, and the computer executable instructions are configured to:

[0081] Obtaining a plurality of news data to generate a corresponding news data set, and preprocessing news data in the news data set to obtain a news data set in a unified format;

[0082] Extracting features of news titles and news texts in the news data set in the unified format by a term frequency-inverse document frequency algorithm;

[0083] Denoising feature vectors corresponding to the news titles and the news texts of the news data set by an elbow rule to obtain a plurality of feature vectors corresponding to the news titles and the news texts of the news data set, and calculating cosine similarities between the plurality of feature vectors;

[0084] The plurality of feature vectors are divided into a plurality of feature vector sets according to cosine similarities between the plurality of feature vectors, and feature vectors corresponding to news titles and news bodies in the plurality of feature vector sets are respectively input into a neural network model pre-trained.

[0085] A detection result of whether the feature vectors corresponding to the news title and the news body are consistent is output through the neural network model pre-trained, and whether the news corresponding to the news title and the news body is false news is determined according to the detection result.

[0086] The detection result corresponding to other news in the feature vector set where the feature vectors corresponding to the news title and the news body are located is consistent with the detection result corresponding to the news, so as to realize detection of false news.

[0087] Each of the embodiments in the present application is described in a progressive manner, and the same and similar parts of each of the embodiments can be referred to each other. Each of the embodiments mainly describes the difference from other embodiments. In particular, the device and medium embodiments are described simply because they are basically similar to the method embodiments, and the relevant parts can be referred to the part of the method embodiments.

[0088] The device and medium provided by the embodiments of the present application are one-to-one corresponding to the method, so the device and medium also have the similar beneficial technical effects as the method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the device and medium will not be described here.

[0089] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can be in the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can be in the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0090] The present application is described with reference to flowcharts and / or block diagrams according to the method, device (system), and computer program product of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device that implements the functions described in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 an apparatus with the functionality to achieve the specified processes or functions in the flow

[0091] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 one or more processes and / or blocks Figure 1 an apparatus with the functionality to achieve the specified processes or functions in the flow

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 one or more processes and / or blocks Figure 1 an apparatus with the functionality to achieve the specified processes or functions in the flow

[0093] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0094] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) about which the computer stores information about an operating system, application software, and / or the like. Memory is an example of computer readable media.

[0095] Computer readable media includes permanent and non-permanent, moveable and non- moveable media that can store information by any method or technology. Information can be computer readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital versatile discs (DVDs) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can store information accessible by a computing device. According to the definition herein, computer readable media does not include transitory media, such as modulated data signals and carrier waves.

[0096] It should also be noted that the terms "comprising," "including," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0097] The above description is merely illustrative of the application, and not restrictive. Various modifications and changes can become apparent to those skilled in the art. Incorporating any modification, equivalent substitution, improvement, etc. within the spirit and principle of the application, shall be included in the scope of the claims of the application.

Claims

1. A method for detecting fake news based on a neural network model, characterized in that, The method comprises: Obtaining a plurality of news data to generate a corresponding news data set, and preprocessing news data in the news data set to obtain a news data set in a unified format; Performing feature extraction on news titles and news texts in the news data set in a unified format through a term frequency-inverse document frequency algorithm; Performing denoising on feature vectors corresponding to the news titles and the news texts of the news data set through an elbow rule to obtain a plurality of feature vectors corresponding to the news titles and the news texts of the news data set, and calculating cosine similarity between the plurality of feature vectors; According to the cosine similarity between the plurality of feature vectors, the plurality of feature vectors are divided into a plurality of feature vector sets, and the feature vectors corresponding to the news titles and the news texts in the plurality of feature vector sets are respectively input into a pre-trained neural network model; the neural network model is obtained based on the combination of a convolutional neural network and a deep neural network; Through the pre-trained neural network model, a detection result of whether the feature vectors corresponding to the news titles and the news texts are consistent is output, and whether the news corresponding to the news titles and the news texts is false news is determined according to the detection result; The detection result of other news corresponding to the feature vector set in which the feature vectors corresponding to the news titles and the news texts are located is consistent with the detection result corresponding to the news, so as to realize the detection of false news; The calculation of the cosine similarity between the plurality of feature vectors specifically comprises: From the plurality of feature vectors corresponding to the news titles and the news texts of the news data set, the feature vector corresponding to the news title and the feature vector corresponding to the news text of the news title are obtained; Based on a cosine similarity calculation formula, the cosine similarity between the feature vector corresponding to the news title and the feature vector corresponding to the news text of the news title and the feature vectors corresponding to other news titles and the feature vectors corresponding to other news texts in the news data set is calculated; According to the cosine similarity between the plurality of feature vectors, the plurality of feature vectors are divided into a plurality of feature vector sets, specifically comprising: Comparing the calculated cosine similarity with a preset similarity threshold, and determining other news with a cosine similarity greater than the preset similarity threshold; The determined other news and the news corresponding to the news title and the news text are stored in the same feature vector set.

2. The method of claim 1, wherein the method further comprises: The obtaining of a plurality of news data to generate a corresponding news data set, and preprocessing news data in the news data set to obtain a news data set in a unified format specifically comprises: Obtaining a plurality of news data from a network or an application through a preset mode to generate a news data set corresponding to the plurality of news data, and determining news titles and news texts in the news data set; The irrelevant words in the news data set are deleted to determine whether the news title and the news body are consistent, so as to clean the news title and the news body in the news data set; the irrelevant words at least include one or more of the following: punctuation marks, stop words and mood words; The cleaned news title and news body are obtained, and the cleaned news title and news body are format-converted to obtain a news data set in a unified format.

3. The method of claim 1, wherein the method further comprises: Before the feature vectors corresponding to the news title and the news body in the plurality of feature vector sets are respectively input into the pre-trained neural network model, the method further comprises: inputting the sample news previously determined as false news into the convolutional neural network for training; extracting attribute features in the sample news through the convolutional layer of the convolutional neural network; the attribute features include: news title, news body; In the case that the attribute features are pooled through the pooling layer, the detection result of whether the sample news is false news is output through the fully connected layer of the convolutional neural network until the detection result of whether the sample news is false news matches the result previously determined that the sample news is false news, and a neural network model is obtained.

4. The method of claim 3, wherein the neural network model is trained based on a dataset including a plurality of news articles and a plurality of fake news articles. After the neural network model is obtained, the method further comprises: adjusting the weight of each attribute feature in the sample news through the deep neural network to obtain the target attribute feature weight in the neural network model; optimizing the neural network model based on the target attribute feature weight to complete the training of the neural network model.

5. The method of claim 1, wherein the method further comprises: The feature extraction of the news title and the news body in the news data set in the unified format through the term frequency-inverse document frequency algorithm specifically comprises: determining the news title and the corresponding news body in the news data set in the unified format, and encoding the news title and the corresponding news body through the term frequency-inverse document frequency algorithm to generate an encoding list corresponding to the news title and an encoding list corresponding to the news body; extracting features from the news title according to the encoding list corresponding to the news title, and extracting features from the news body according to the encoding list corresponding to the news body.

6. The method of claim 1, wherein the method further comprises: The detection result of other news in the feature vector set corresponding to the feature vectors of the news title and the news body is consistent with the detection result of the news, so as to realize the detection of false news, specifically comprising: if the detection result of the news is false news, it is determined that other news in the feature vector set corresponding to the feature vectors of the news title and the news body is false news; if the detection result of the news is not false news, it is determined that other news in the feature vector set corresponding to the feature vectors of the news title and the news body is not false news, and the detection of false news in the news data set is realized. 7.A fake news detection device based on a neural network model, characterized by, The device comprises: at least one processor; and a memory connected in communication with the at least one processor; The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of claim 1-6.

8. A non-transitory computer storage medium storing computer-executable instructions that, when executed, cause a computer to: The computer executable instructions are configured to: ​ The method of claim 1-6.

Images