A false news detection method and system based on user information and interaction information

Abstract

The application discloses a false news detection method and system based on user information and interactive information, inputs user information into a pre-trained multilayer perceptron to obtain a user pre-score, extracts features from text data of to-be-detected news to obtain text features, constructs a user interaction graph and combines the user pre-score to generate user features, constructs a user post graph and combines the user pre-score to generate user post features, detects the to-be-detected news according to the text features, the user features and the user post features, and obtains a detection result. The system comprises a social robot module, a text content extraction module, a user interaction module, a user post information aggregation module and a false news detection module. By using the application, the influence of user information and user interactive information is considered, and the accuracy of false news detection is improved. The application can be widely applied to the field of data analysis.

Description

Technical Field

[0001] This invention relates to the field of data analysis, and in particular to a method and system for detecting fake news based on user information and interaction information. Background Technology

[0002] In recent years, with the rapid development of internet technology, the use of online social media platforms has increased rapidly. These platforms have become an important source of information, providing the public with a new channel for accessing information. Online social media platforms have become convenient online platforms for users to obtain information, express opinions, and communicate with each other. As more and more people are keen to participate in discussions of trending topics and express their opinions and insights on social media, a large amount of information is contained within these platforms. However, this information is also rife with various rumors and fake news. The spread of false information on social media is equally simple and rapid. Furthermore, because false information can gain powerful penetration through widespread dissemination in society, it can cause enormous harm, public panic, and significant economic losses. Therefore, in response to the potential panic and threats caused by fake news on these social media platforms, there is an urgent need to develop a method to quickly and effectively identify fake news on social media.

[0003] Existing methods for detecting fake news are mainly single-mode approaches, focusing on extracting textual features or incorporating image data using multimodal methods and applying computer vision techniques to utilize relevant news image information. However, they do not take advantage of specific background information of users on social platforms (such as account time, online time, number of followers, number of followers, etc.) and rich interactive information directly from users (such as forwarding, commenting, following, etc.). They ignore the importance of publisher and user information for fake news detection, which leads to poor detection accuracy. Summary of the Invention

[0004] In view of this, in order to solve the technical problem that fake news detection methods do not consider the influence of user information and user interaction information, thus leading to low detection accuracy, firstly, this invention proposes a fake news detection method based on user information and interaction information, the method comprising the following steps:

[0005] User information is input into a pre-trained multilayer perceptron to obtain a user pre-score;

[0006] Feature extraction is performed on the text data of the news to be tested to obtain text features;

[0007] Construct a user interaction graph and combine it with the user pre-score to generate user features;

[0008] Construct a user posting graph and combine it with the user pre-score to generate user posting features;

[0009] Based on the text features, user features, and user posting features, the news to be tested is detected, and the detection results are obtained.

[0010] In some embodiments, the step of inputting user information into a pre-trained multilayer perceptron to obtain a user pre-score specifically includes:

[0011] A pre-trained multilayer perceptron is obtained by training a multilayer perceptron based on a social robot dataset.

[0012] User information is converted into information features and used as input to a pre-trained multilayer perceptron to obtain user pre-scores.

[0013] This optimization step utilizes a pre-trained social robot detection model to leverage user information and pre-determine user scores, which facilitates subsequent steps.

[0014] In some embodiments, the step of extracting features from the text data of the news to be tested to obtain text features specifically includes:

[0015] The text data of the news to be tested is cleaned and segmented to obtain preprocessed text data;

[0016] The preprocessed text data is used as input to the representation transformation model for feature extraction to obtain text features.

[0017] This optimized process involves simple preprocessing and feature extraction of text data, which can improve detection speed while saving computational resources.

[0018] In some embodiments, the step of constructing a user interaction graph and combining it with the user pre-score to generate user features specifically includes:

[0019] Construct a user interaction graph based on the interaction information between users;

[0020] The interaction information between the nodes is captured based on the graph attention mechanism, and the attention of the nodes is adjusted by combining the user pre-score to obtain user features.

[0021] The process of generating the user features is as follows:

[0022]

[0023]

[0024] Among them, U (0) The initial characteristics of nodes in the user interaction graph. This indicates the user's pre-scoring score. GAT(·) represents user characteristics and GAT(·) represents graph attention network.

[0025] This optimization step utilizes user interaction graphs and graph attention mechanisms to capture the importance of interactions between user nodes in order to obtain user characteristics.

[0026] In some embodiments, the step of constructing a user posting graph and combining it with the user pre-score to generate user posting features specifically includes:

[0027] Construct a user posting graph based on the posts published by users of the news to be tested;

[0028] Using the text features of the post as the source node features of the post node, the initial node features in the user posting graph are obtained;

[0029] Based on a graph convolutional neural network, the initial features of nodes in the user posting graph are aggregated and combined with the user pre-score to obtain user posting features.

[0030] The process of generating the user-published features is as follows:

[0031] P (1) =GCN(P)

[0032]

[0033] Where P represents the initial features of nodes in the user-published graph, and GCN(·) represents a graph convolutional neural network. This indicates the user's pre-score corresponding to the image posted by that user. This indicates the user-published characteristics.

[0034] This optimization step utilizes user-posted graphs and graph convolutional neural networks to aggregate the content features of user-posted posts in order to obtain user-posted features.

[0035] In some embodiments, the step of detecting the news to be tested based on the text features, the user features, and the user posting features to obtain the detection result specifically includes:

[0036] The text features, user features, and user posting features are concatenated to obtain the final representation;

[0037] The final representation is passed through a fully connected layer and a classification layer to complete the detection and classification of the news to be tested, and the detection result is obtained.

[0038] By combining text features, user representation features, and user posting features in this optimized step, the detection accuracy can be overcome by addressing the problem that existing detection methods do not utilize this rich information, which leads to low detection accuracy.

[0039] Secondly, this invention proposes a fake news detection system based on user information and interaction information, the system comprising:

[0040] The social robot module is used to input user information into a pre-trained multilayer perceptron to obtain a pre-score for the user;

[0041] The text content extraction module is used to extract features from the text data of the news to be tested, and obtain text features;

[0042] The user interaction module is used to construct a user interaction graph and, in conjunction with the user pre-score, generate user features.

[0043] The user post information aggregation module is used to construct a user posting graph and, in conjunction with the user pre-score, generate user posting features;

[0044] The fake news detection module is used to detect the news to be tested based on the text features, the user features, and the user posting features, and obtain the detection results.

[0045] This invention also proposes a fake news detection device based on user information and interaction information, comprising:

[0046] At least one processor;

[0047] At least one memory for storing at least one program;

[0048] When the at least one program is executed by the at least one processor, the at least one processor implements the fake news detection method based on user information and interaction information as described above.

[0049] Based on the above scheme, the present invention provides a method and system for detecting fake news based on user information and interaction information. It can effectively utilize user information and user interaction structure information, use a pre-trained detection model to utilize user information, make a pre-scoring judgment on users, and utilize the subsequent dissemination structure. Finally, it can judge the news to be tested on social networks to achieve better detection results. Attached Figure Description

[0050] Figure 1 This is a flowchart of the steps of a method for detecting fake news based on user information and interaction information according to the present invention;

[0051] Figure 2 This is a flowchart of the user pre-score generation steps in a specific embodiment of the present invention;

[0052] Figure 3 This is a flowchart of the text feature generation steps in a specific embodiment of the present invention;

[0053] Figure 4 This is a flowchart of the steps for generating user features according to a specific embodiment of the present invention;

[0054] Figure 5 This is a flowchart of the steps for generating user-published features according to a specific embodiment of the present invention;

[0055] Figure 6 This is a flowchart of the steps for generating the detection results according to a specific embodiment of the present invention;

[0056] Figure 7 This is a structural block diagram of a fake news detection system based on user information and interaction information according to the present invention. Detailed Implementation

[0057] This invention first pre-trains a social robot detection module using a social robot detection dataset, and pre-scores users using user feature data. Then, it obtains text features based on fake news text content through a text content extraction module, followed by a user post information aggregation module to obtain aggregated posting features. Finally, a user interaction graph module combines user information and pre-scores to obtain user features. Finally, the text features, user features, and user posting features are concatenated in the fake news detection module to form the final representation of the news to be detected, and a final judgment is made. This invention can better utilize user information and user interaction information, resulting in better detection performance.

[0058] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0059] It should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0060] It should be understood that the terms "system," "apparatus," "unit," and / or "module" used in this application are a method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0061] Unless the context explicitly indicates an exception, words such as "a," "an," "a kind," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list; a method or apparatus may also include other steps or elements. An element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, product, or apparatus that includes the element.

[0062] In the description of the embodiments of this application, "a plurality of" refers to two or more. The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0063] Furthermore, flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, the steps can be processed in reverse order or simultaneously. Additionally, other operations can be added to these processes, or one or more steps can be removed from them.

[0064] Reference Figure 1 This is a flowchart illustrating an optional example of the fake news detection method based on user information and interaction information proposed in this invention. The method can be applied to computer devices, and the fake news detection method proposed in this embodiment may include, but is not limited to, the following steps:

[0065] Step S1: Input user information into a pre-trained multilayer perceptron to obtain the user's pre-score;

[0066] Step S2: Extract features from the text data of the news to be tested to obtain text features;

[0067] Step S3: Construct a user interaction graph and combine it with the user pre-score to generate user features;

[0068] Step S4: Construct a user posting graph and combine it with the user pre-score to generate user posting features;

[0069] Step S5: Based on the text features, user features, and user posting features, perform detection on the news to be tested and obtain the detection results.

[0070] In some feasible embodiments, refer to Figure 2 Step S1 specifically includes:

[0071] S1.1. Train a multilayer perceptron based on a social robot dataset to obtain a pre-trained multilayer perceptron;

[0072] Specifically, the Multilayer Perceptron in the social bot detection module was pre-trained using six social bot datasets (midterm-2018, botowiki, celebrity, cresci-17, Cresci-2015, and Cresci-rtbust-2019). User information (username, personal description, number of followers, number of fans, etc.) from the fake news dataset was transformed into initial user information features and then input into the pre-trained Multilayer Perceptron layers to obtain the user's pre-score.

[0073] S1.2. Convert user information into information features and use them as input to a pre-trained multilayer perceptron to obtain user pre-scores.

[0074] In this embodiment, C is used. u To represent the obtained initial user representation, and using User information is represented by C u To the user's initial score The specific steps for the conversion are as follows:

[0075] The multilayer perceptron (MLP) layer in the pre-trained social robot module is represented as:

[0076]

[0077] C is coming soon u As input to the Multilayer Perceptron (MLP) layer, the final user initial score is obtained as a binary representation. as follows:

[0078]

[0079]

[0080] in b c , b u The parameters have been pre-trained on the social robot dataset (Bot-Detection). During the training for fake news detection, the parameters of this multilayer perceptron (MLP) layer are frozen and remain unchanged.

[0081] In some feasible embodiments, refer to Figure 3 Step S2 specifically includes:

[0082] S2.1 Clean and segment the text data of the news to be tested to obtain preprocessed text data;

[0083] In this process, preprocessed text data is obtained through data special symbol cleaning and word segmentation, and the preprocessed text data is represented as X.

[0084] In addition, the preprocessing may also include operations such as symbol filtering, character cropping, and padding of the text data.

[0085] S2.2. Use the preprocessed text data as input to the representation model to extract features and obtain text features.

[0086] Using X as input to the BERT model, we ultimately obtain unified text features. The specific steps are as follows:

[0087]

[0088] Specifically, a pre-trained bidirectional encoder-representations-from-transformers (BERT) can be used.

[0089] In some feasible embodiments, refer to Figure 4 Step S3 specifically includes:

[0090] S3.1 Construct a user interaction graph based on the interaction information between users;

[0091] Using Figure G u = (V, E) to represent the propagation structure pattern of user interactions, and to obtain the propagation structure feature representation. For each user node V, the given initial node features are: (|V| represents the number of user nodes).

[0092] S3.2. Capture the interaction information between the nodes based on the graph attention mechanism, and adjust the attention of the nodes in combination with the user pre-score to obtain user features.

[0093] use This represents the transformation of user node features from focusing on the node feature updates of each neighbor node to finally obtaining the final user representation, and incorporates the user's initial score during the transformation process.

[0094] Given initial features U (0)The algorithm uses a graph attention network to acquire interaction information between nodes and differentiates the importance of node information aggregation by assigning different levels of attention to different nodes. Furthermore, it utilizes the pre-scores of users obtained above to further reduce the focus on nodes with lower scores (lower scores indicate a higher likelihood of being a bot user), ultimately yielding the final user features.

[0095]

[0096]

[0097] In some feasible embodiments, refer to Figure 5 Step S4 specifically includes:

[0098] S4.1 Construct a user posting graph based on the posts published by users of the news to be tested;

[0099] S4.2. Using the text features of the post as the source node features of the post node, the initial node features in the user posting graph are obtained.

[0100] Use G p = (V,E) represents the posting graph of a user's news post, and represents the relationship between the user's post and the user, represented by p. u The user node that published the source news post is represented by p. u(1) ,p u(2) ,…,p u(|V|-1) This represents a news post published by a user. The initial features of the nodes in the user-published graph are P = {P...} u ,P u(1) ,P u(2) ,…,P u(|V|-1)}

[0101] S4.3. Based on the graph convolutional neural network, the initial features of the nodes in the user posting graph are aggregated and combined with the user pre-score to obtain the user posting features.

[0102] use The transformation from the initial features of nodes in the user-posted graph to user-posted feature representations is as follows: User-posted post features are aggregated using a graph convolutional neural network, and then multiplied with the corresponding pre-scores obtained from the user data to obtain the final user-posted features.

[0103] P (1) =GCN(P)

[0104]

[0105] In some feasible embodiments, refer to Figure 6 Step S5 specifically includes:

[0106] S5.1. Concatenate the text features, user features, and user posting features to obtain the final representation;

[0107] S5.2. The final expression is passed through a fully connected layer and a classification layer to complete the detection and classification of the news to be tested, and the detection result is obtained.

[0108] The above-obtained text features of the news posts are represented as follows User Feature Representation and user-published features The final news post was obtained by piecing together the information. Then, a fully connected layer and softmax are used to classify the news posts under test as real or fake news, as follows:

[0109]

[0110]

[0111] Where Concatenate is the tensor concatenation operation, W is the weight matrix of the fully connected layer, b is the bias value, and Softmax is the softmax function. To predict the probability, based on the predicted probability and combined with the set probability experience value, it is determined whether the news to be tested is fake news.

[0112] like Figure 7 As shown, a fake news detection system based on user information and interaction information includes:

[0113] The social robot module is used to input user information into a pre-trained multilayer perceptron to obtain a pre-score for the user;

[0114] The text content extraction module is used to extract features from the text data of the news to be tested, and obtain text features;

[0115] The user interaction module is used to construct a user interaction graph and, in conjunction with the user pre-score, generate user features.

[0116] The user post information aggregation module is used to construct a user posting graph and, in conjunction with the user pre-score, generate user posting features;

[0117] The fake news detection module is used to detect the news to be tested based on the text features, the user features, and the user posting features, and obtain the detection results.

[0118] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0119] A fake news detection device based on user information and interaction information:

[0120] At least one processor;

[0121] At least one memory for storing at least one program;

[0122] When the at least one program is executed by the at least one processor, the at least one processor implements the fake news detection method based on user information and interaction information as described above.

[0123] The content of the above method embodiments is applicable to the device embodiments. The specific functions implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0124] A storage medium storing processor-executable instructions, which, when executed by a processor, are used to implement a fake news detection method based on user information and interaction information as described above.

[0125] The content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0126] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for detecting fake news based on user information and interaction information, characterized in that, Includes the following steps: User information is input into a pre-trained multilayer perceptron to obtain a user pre-score; Feature extraction is performed on the text data of the news to be tested to obtain text features; Construct a user interaction graph and combine it with the user pre-score to generate user features; Construct a user posting graph and combine it with the user pre-score to generate user posting features; Based on the text features, user features, and user posting features, the news to be tested is detected, and the detection results are obtained. The step of constructing a user interaction graph and combining it with the user pre-score to generate user features specifically includes: Construct a user interaction graph based on the interaction information between users; The interaction information between nodes is captured based on the graph attention mechanism, and the attention of nodes is adjusted in combination with the user pre-score to obtain user characteristics; The process of generating the user features is as follows: in, The initial characteristics of nodes in the user interaction graph. This indicates the user's pre-scoring score. Indicates user characteristics, Represents a graph attention mechanism network; The step of constructing a user posting graph and combining it with the user pre-score to generate user posting features specifically includes: Construct a user posting graph based on the posts published by users of the news to be tested; Using the text features of the post as the source node features of the post node, the initial node features in the user posting graph are obtained; Based on graph convolutional neural networks, the initial features of nodes in the user posting graph are aggregated and combined with the user pre-score to obtain user posting features; The process of generating the user-published features is as follows: in, This represents the initial characteristics of the nodes in the user-published graph. This represents a graph convolutional neural network. This indicates the user's pre-score corresponding to the image posted by that user. This indicates the user-published characteristics.

2. The method for detecting fake news based on user information and interaction information according to claim 1, characterized in that, The step of inputting user information into a pre-trained multilayer perceptron to obtain a user pre-score specifically includes: A pre-trained multilayer perceptron is obtained by training a multilayer perceptron based on a social robot dataset. User information is converted into information features and used as input to a pre-trained multilayer perceptron to obtain user pre-scores.

3. The method for detecting fake news based on user information and interaction information according to claim 1, characterized in that, The step of extracting features from the text data of the news to be tested to obtain text features specifically includes: The text data of the news to be tested is cleaned and segmented to obtain preprocessed text data; The preprocessed text data is used as input to the representation transformation model for feature extraction to obtain text features.

4. The method for detecting fake news based on user information and interaction information according to claim 1, characterized in that, The step of detecting the news to be tested based on the text features, the user features, and the user posting features, and obtaining the detection result, specifically includes: The text features, user features, and user posting features are concatenated to obtain the final representation; The final expression is passed through a fully connected layer and a classification layer to complete the detection and classification of the news to be tested, and the detection results are obtained.

5. A fake news detection system based on user information and interaction information, characterized in that, A method for performing fake news detection based on user information and interaction information as described in claim 1 includes: The social robot module is used to input user information into a pre-trained multilayer perceptron to obtain a pre-score for the user; The text content extraction module is used to extract features from the text data of the news to be tested, and obtain text features; The user interaction module is used to construct a user interaction graph and, in conjunction with the user pre-score, generate user features. The user post information aggregation module is used to construct a user posting graph and, in conjunction with the user pre-score, generate user posting features; The fake news detection module is used to detect the news to be tested based on the final feature representation obtained by concatenating the text features, the user features, and the user posting features, and obtain the detection result.

6. A fake news detection device based on user information and interaction information, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the fake news detection method based on user information and interaction information as described in any one of claims 1-4.

Images