Fake news detection method and system based on joint information propagation within and between images

Abstract

The present invention belongs to the technical field of natural language processing, and provides a fake news detection method and system based on joint information propagation between graphs, including: obtaining the text to be detected, and after preprocessing, respectively constructing a text sequence graph, a text syntax graph and a text semantic graph; based on the text sequence graph, the text syntax graph and the text semantic graph, firstly performing information interaction within the graph, and then performing information interaction between graphs, obtaining a joint information text sequence graph, a joint information text syntax graph and a joint information text semantic graph, and obtaining a vector representation of the text to be detected through a graph attention mechanism and a node attention mechanism, inputting it into a classifier, and obtaining a result of whether the text to be detected is fake news. The feature sparsity problem existing in the vectorization of short texts is solved, the text vector representation is enriched, the problem of missing information interaction between long-distance and non-continuous texts is solved, and the accuracy of text representation is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of natural language processing, and in particular, relates to a fake news detection method and system based on joint information propagation within and between graphs. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the Internet age, every netizen can release news through social media and other channels. The lack of professional ethics in journalism, imperfect restraint mechanisms and rectification measures have led to the emergence of fake news events, which not only mislead readers who do not know the truth, but also cause adverse social impacts.

[0004] In order to detect fake news, many organizations use manual annotation methods to verify the authenticity of news. After a large number of fake news appeared, labor-intensive manual annotation methods can no longer achieve timely processing of fake news. Compared with manual annotation methods, fake news intelligent detection has become a new direction for fake news identification. Existing deep learning methods such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have achieved great success in fake news detection. The biggest advantage of this type of method is that it does not rely on artificial features and can analyze variable-length sequence data and discover complex patterns hidden in text data. However, due to the limitation of the convolution kernel size, CNN cannot capture the dependencies between long-distance and non-continuous words; although the Long Short Term Memory (LSTM) network overcomes the difficulty of RNN long-distance dependency gradient vanishing, it still cannot effectively capture long-distance dependencies.

[0005] The first step in the intelligent detection of fake news is text vectorization, which converts text data into binary digital representations that can be recognized by computers. One-hot encoding corresponds to a vector for each word or character, but the encoded vector cannot reflect the similarity and connection between words, nor does it contain any semantic information. As an improved method, word embedding is a method of representing words as low-dimensional and dense vectors, with real numbers in each dimension. After combining word embedding with fully connected neural networks, a series of deep learning technologies such as Word2vec, Glove and FastText were proposed, providing a variety of numerical representation methods for text. Presenting content data in short text has become an important form of information dissemination. However, the data it contains has a small number of words and weak descriptive information, resulting in sparse text features and difficulty in extracting sample features from the text for text vectorization. Summary of the invention

[0006] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a fake news detection method and system based on joint information propagation within and between graphs, which realizes global interaction of homogeneous and heterogeneous information based on various information existing in the text, solves the feature sparsity problem existing in the vectorization of short texts, enriches the text vector representation, solves the problem of missing interaction of long-distance and non-continuous text information, and improves the accuracy of text representation.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A first aspect of the present invention provides a fake news detection method based on intra-graph and inter-graph joint information propagation, comprising:

[0009] Get the text to be detected and perform preprocessing;

[0010] Based on the preprocessed text, text sequence features, text grammar features and text semantic features are extracted respectively, and then a text sequence graph, a text grammar graph and a text semantic graph are constructed;

[0011] Based on the text sequence graph, text syntax graph and text semantic graph, the information interaction within the graph is first performed, and then the information interaction between graphs is performed to obtain the joint information text sequence graph, the joint information text syntax graph and the joint information text semantic graph;

[0012] Based on the joint information text sequence graph, joint information text syntax graph and joint information text semantic graph, the vector representation of the text to be detected is obtained through the graph attention mechanism and the node attention mechanism;

[0013] The vector representation of the text to be detected is input into the classifier to obtain the result of whether the text to be detected is fake news.

[0014] Furthermore, the preprocessing is to perform data enhancement after normalizing the text to be detected.

[0015] Furthermore, the normalization process includes: removing special characters, word segmentation, deleting stop words, stemming and lowercasing words.

[0016] Furthermore, the data enhancement includes synonym replacement, random insertion, random exchange and random deletion.

[0017] Furthermore, the text sequence feature extraction method is:

[0018] Using a sliding window mechanism to traverse the preprocessed text, calculate the number of times each word pair appears in the same sliding window, and the number of times each word appears in all sliding windows;

[0019] The text order weight is calculated based on the number of times each word pair appears in the same sliding window and the number of times each word appears in all sliding windows.

[0020] Furthermore, the information interaction within the graph adopts a reset gate and an update gate to update the node representations and the weights between the nodes in the text sequence graph, the text syntax graph and the text semantic graph respectively.

[0021] Furthermore, the inter-graph information interaction performs information propagation between the text sequence graph, the text syntax graph and the text semantic graph after the intra-graph information interaction.

[0022] A second aspect of the present invention provides a fake news detection system based on intra-graph and inter-graph joint information propagation, comprising:

[0023] The preprocessing module is configured to: obtain the text to be detected and perform preprocessing;

[0024] A graph construction module is configured to: extract text sequence features, text grammar features and text semantic features based on the preprocessed text, and then construct a text sequence graph, a text grammar graph and a text semantic graph;

[0025] The information propagation module is configured to: based on the text sequence graph, the text syntax graph and the text semantic graph, first perform intra-graph information interaction and then perform inter-graph information interaction to obtain a joint information text sequence graph, a joint information text syntax graph and a joint information text semantic graph;

[0026] A vector representation module is configured to obtain a vector representation of the text to be detected through a graph attention mechanism and a node attention mechanism based on a joint information text sequence graph, a joint information text syntax graph, and a joint information text semantic graph;

[0027] The classification module is configured to: input the vector representation of the text to be detected into the classifier to obtain a result of whether the text to be detected is fake news.

[0028] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the fake news detection method based on joint information propagation within and between graphs as described above.

[0029] A fourth aspect of the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the fake news detection method based on joint information propagation within and between graphs as described above are implemented.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] The present invention provides a fake news detection method based on joint information propagation within and between graphs. The method achieves the purpose of enriching text features by using a third-order text graph tensor to describe text, solves the feature sparsity problem of short texts, and improves the accuracy of fake news detection.

[0032] The present invention provides a fake news detection method based on joint information propagation within and between graphs, which realizes global interaction of homogeneous and heterogeneous information by performing joint information propagation within and between graphs on texts, solves the problem of missing interaction of long-distance and non-continuous texts, and improves the accuracy of text representation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0034] Figure 1 is a flow chart of a fake news detection method based on intra-graph and inter-graph joint information propagation according to Embodiment 1 of the present invention;

[0035] Figure 2 is a text normalization flow chart of the first embodiment of the present invention;

[0036] Figure 3 is a flowchart of grammatical feature extraction according to the first embodiment of the present invention;

[0037] FIG4( a ) is a flowchart of information propagation within a graph according to Embodiment 1 of the present invention;

[0038] FIG4( b ) is a flowchart of information transmission between graphs according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0039] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0040] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0041] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0042] Embodiment 1

[0043] This embodiment provides a fake news detection method based on joint information propagation within and between graphs. First, the text to be detected is obtained and preprocessed (the text to be detected is normalized, and the normalized text is data enhanced to obtain enhanced text); secondly, based on the preprocessed text, the third-order text graph tensor describes three text features to obtain a feature vector; thirdly, based on the feature vector, information propagation between graphs within and between graphs is used to interact with isomorphic and heterogeneous information to obtain a feature graph vector; then, based on the feature graph vector, word weight learning is performed through the graph attention mechanism and the node attention mechanism to obtain a text vector; finally, based on the text vector, the classifier obtains the result of whether the text to be detected is fake news. Specifically, the following steps are included:

[0044] Step (1): Get the text to be detected and normalize it. Clean the text to be detected and remove characters that are irrelevant to the fake news detection task. Figure 2 As shown in the text normalization flowchart, text normalization includes the following steps:

[0045] Step (101): Clean the text: remove special characters such as URL tags and punctuation marks contained in the text to be detected.

[0046] Step (102): Word segmentation: Segment the text to be detected into several words after removing special characters.

[0047] Step (103): Delete stop words: Delete some words that frequently appear in the article but are irrelevant to the text, such as prepositions, pronouns, and auxiliary words.

[0048] Step (104): Stem extraction: Use the NLTK tool to perform word form restoration, extract the word stems, and lowercase all the extracted stems.

[0049] like Figure 2 As shown, the obtained text is normalized, and after deleting redundant characters, segmenting words, deleting stop words, extracting stems, and lowercasing words, a neat text is obtained to facilitate the construction of the dictionary.

[0050] Step (2): Perform data enhancement on the normalized text to obtain enhanced text. Specifically: Perform synonym replacement, random insertion, random exchange and random deletion on the text, and combine the text after synonym replacement, random insertion, random exchange and random deletion with the original text to obtain enhanced text, as shown in Table 1, to solve the problems of data imbalance and insufficient label data in the text.

[0051] (1) Synonym replacement: randomly select a word in the sentence and randomly select a synonym of the word to replace it;

[0052] (2) Random insertion: randomly select a word in the sentence, randomly select a synonym of the word and randomly insert it into the sentence;

[0053] (3) Random exchange: Randomly select two words in a sentence and exchange their positions;

[0054] (4) Random deletion: Randomly select a word in the sentence and delete it.

[0055] Table 1. Data enhancement example table

[0056]

[0057] Step (3): Based on the enhanced text, a third-order text graph tensor is constructed to describe three types of text features and obtain feature vectors for learning fake news features to make the classification more accurate. Text order features use the sliding window mechanism to traverse the text and describe the language properties of local co-occurrence between words; text grammatical features obtain the grammatical interdependence between words; text semantic features obtain the topic distribution of the text and the distribution of topic keywords.

[0058] like Figure 1 As shown, the construction of the third-order text graph tensor mainly involves the following processing:

[0059] Step (301): Text sequence feature extraction: With the help of a sliding window mechanism, point-wise mutual information (PMI) is used to describe the sequential context information.

[0060] First, define a sliding window size window_size as the length that can traverse the longest character;

[0061] Secondly, the sliding window mechanism is used to traverse the preprocessed text to be detected, and the number of sliding windows N contained in the text is obtained according to the data length seq_len windows ;

[0062] Finally, according to the number of sliding windows, PMI is used to describe the order weight between any two words, that is, the text order feature:

[0063]

[0064] in, Represents a word pair (w i ,w j ) appears in the same sliding window, N co-occurence (w i ,w j ) represents a word pair (w i ,wj ) appears in the same sliding window; Represents word w i The probability of appearing in all sliding windows, N occurrence (w i ) represents word w i The number of occurrences in all sliding windows.

[0065] Step (302): Extract text grammatical features: Obtain the grammatical dependencies between words. Use Stanford CoreNLP to extract word grammatical dependencies to describe text grammatical features. Figure 3 As shown in the figure, given a text, various word grammatical dependencies can be extracted, where xcomp, dobj, and nbsuj are abbreviations of the names of some dependency relationships; root is the starting symbol of the dependency relationship in a sentence extracted by the CoreNLP tool, which has no practical meaning. Each word grammatical dependency relationship is a one-way triple: (governor, dependent, the name of relation). This triple is a description form of the word grammatical dependency relationship in the text extracted by the Stanford CoreNLP tool, where governor describes the initiating word of the dependency relationship; dependent describes the receiving word of the dependency relationship; the name of relation describes the name of the dependency relationship; given a preprocessed text to be detected, call the CoreNLP function to extract the dependency relationship in the text in the form of a triple. Define the word grammatical weight in the triple, that is, the text grammatical feature, as: E(governor,dependent)=1.

[0066] Step (303): Text semantic feature extraction: extract the subject information and subject-related keywords in the text, and use the subject and keywords to describe the text semantic information. It specifically includes the following steps:

[0067] Step (30301): Use Latent Dirichlet Allocation (LDA) to extract the topic distribution T = (t1, t2, ..., t k )(k is the number of topics);

[0068] Step (30302): For each topic t i , which can be represented by the word probability distribution

[0069] Step (30303): To further accurately represent the topic, it is determined that each topic can be represented by the M words that are most relevant to it:

[0070] Step (30304): Given any text x = (w1, w2, ... w h ,…,w n )(n is the number of words, containing n words), the text corresponds to a topic t j , t j is described by a set of words If there is a word in the text x Then we can define w h and any other word w in text x k The word pairs (w i ,w k ), that is, the semantic weight of the text, is E(w i ,w k )=1(i=h,k≠h).

[0071] Based on the text sequence features, text grammatical features and text semantic features, the text sequence graph G is constructed respectively. seq , text grammar graph G syn and text semantic graph G sem Each text graph can be represented as G = (V, E), where V (|V| = n) represents a node set, and E describes the weights between nodes. The sequence graph corresponds to d in step 301. sequential (w i ,w j ), the syntax diagram corresponds to E(governor,dependent) in step 302, and the semantic diagram corresponds to E(w i ,w k ).

[0072] Step (4): Joint information propagation between graphs within and between graphs, for the text sequence graph G seq , text grammar graph G syn and text semantic graph G sem First, information interaction is performed within the graph, and then information interaction is performed between graphs to obtain a joint information feature graph. Based on the feature vector, information propagation is performed within the graph to achieve global interaction of homogeneous information and obtain homogeneous graph vectors; based on homogeneous graph vectors, information propagation is performed between graphs to achieve global interaction of heterogeneous information and obtain feature graph vectors. The specific implementation is:

[0073] Step (401): Information interaction within the graph, using reset gates and update gates to update the node representations and weights between nodes in the text sequence graph, text syntax graph, and text semantic graph, respectively. Specifically, in the same graph, each word node exchanges isomorphic information with other nodes in the graph and updates the node representation, as shown in Figure 4(a). Given the representation of the current node x t , and the state h passed down from the previous node t-1 , reset gate r t =σ(W r ·[h t-1 ,x t ]) decides how to combine the new input information with the previous information and updates the gate z t =σ(W z ·[h t-1 ,x t ]) Define the amount of previous information saved to the current time, and get the state h passed to the next node t =(1-z t )*h t-1 +z t *h′, where h′=tanh(W h′ ·[r t *h t-1 ,x t ]) and the updated representation of the current node y t =σ(W y ·h t ), W r , W z , W h′ and W y All represent trainable parameters. Text sequence graph G seq , text grammar graph G syn and text semantic graph G sem Update the nodes and the weights between nodes respectively to obtain the updated text sequence graph G seq , text grammar graph G syn and text semantic graph G sem .

[0074] Step (402): Inter-graph information interaction: refers to the information propagation between the text sequence graph, text syntax graph and text semantic graph after the intra-graph information interaction, so as to realize the information interaction between heterogeneous features. After the intra-graph update mechanism, each node has a separate representation for different text features, but the features are still separated from each other. The purpose of the inter-graph information interaction mechanism is to integrate different information and realize the interaction of heterogeneous information, as shown in Figure 4(b). The node representations between different graphs are added to the corresponding node representations, i.e., G′ seq =G seq +Gsem ,G′ sem =G sem +G syn , G′ syn =G syn +G seq After calculation, we get the joint information text sequence graph, joint information text syntax graph and joint information text semantic graph, G′ seq , G′ syn and G′ sem , realize the propagation of information between graphs and update the graph representation. seq , G syn and G sem Updated text sequence graph, text syntax graph, and text semantic graph for intra-graph information propagation.

[0075] Step (5): Based on the joint information text sequence graph, the joint information text syntax graph, and the joint information text semantic graph, word weight learning is performed through the graph attention mechanism and the node attention mechanism to obtain the text vector. Specifically, the following processing is included:

[0076] Step (501): Construct a graph attention mechanism to learn the impact of different text graphs on the final classification results and obtain the final text graph representation G f , obtained from formula (1):

[0077]

[0078] Among them, β t Represents the t-th text graph (G1 is G′ seq All nodes of G2 are G′ syn All nodes of G3 are G′ sem The importance of all nodes is expressed by (2) and (3):

[0079] u t =tanh(W t G t +b t ) (2)

[0080]

[0081] Among them, u t G t The hidden representation obtained by sending it to the fully connected layer, W t , b t are parameters that can be trained.

[0082] Step (502): Since G fThe node information in has been fully represented. In order to learn the influence of different words on the final classification results, the word attention mechanism is used to calculate and obtain the final vector representation w of the text to be detected. f , calculated by formula (4) and formula (5):

[0083] w i =σ(f1(w i ))⊙tanh(f2(w i )) (4)

[0084]

[0085] Among them, w i It's G f The representation of the i-th word in , where both f1 and f2 are perceptrons.

[0086] Step (6): Classifier: The vector representation w of the text to be detected f The input is sent to the classifier for classification prediction to obtain the result of whether the text to be tested is fake news, and the classification result is presented to the user in a visual way.

[0087] The present invention realizes the global interaction of isomorphic and heterogeneous information based on the various information existing in the text. First, based on the text content, various information between texts is extracted; then, based on the various text information, the interaction of isomorphic and heterogeneous information is realized by joint information propagation within and between graphs to obtain a feature graph vector; based on the feature graph vector, a text vector is obtained through a graph attention mechanism and a node attention mechanism; finally, based on the text vector, a classifier obtains the result of whether the text to be detected is fake news. The invention solves the problem of feature sparsity when short text is vectorized, enriches the text vector representation; solves the problem of missing long-distance and non-continuous text information interaction, and improves the accuracy of text representation.

[0088] The present invention extracts diverse information from texts and conducts joint information propagation within and between graphs, effectively solving the problem of sparse features in short texts and the problem of missing interactions in long-distance, non-continuous texts, thereby improving the accuracy of detecting fake news in network environments.

[0089] Embodiment 2

[0090] This embodiment provides a fake news detection system based on intra-graph and inter-graph joint information propagation, which specifically includes the following modules:

[0091] The preprocessing module is configured to: obtain the text to be detected and perform preprocessing;

[0092] A graph construction module is configured to: extract text sequence features, text grammar features and text semantic features based on the preprocessed text, and then construct a text sequence graph, a text grammar graph and a text semantic graph;

[0093] The information propagation module is configured to: based on the text sequence graph, the text syntax graph and the text semantic graph, first perform intra-graph information interaction and then perform inter-graph information interaction to obtain a joint information text sequence graph, a joint information text syntax graph and a joint information text semantic graph;

[0094] A vector representation module is configured to obtain a vector representation of the text to be detected through a graph attention mechanism and a node attention mechanism based on a joint information text sequence graph, a joint information text syntax graph, and a joint information text semantic graph;

[0095] The classification module is configured to: input the vector representation of the text to be detected into the classifier to obtain a result of whether the text to be detected is fake news.

[0096] It should be noted here that each module in this embodiment corresponds to each step in Example 1 one by one, and the specific implementation process is the same, which will not be repeated here.

[0097] Embodiment 3

[0098] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps in the fake news detection method based on joint information propagation within and between graphs as described in the first embodiment above are implemented.

[0099] Embodiment 4

[0100] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the fake news detection method based on joint information propagation within and between graphs as described in the first embodiment are implemented.

[0101] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) containing computer-usable program code.

[0102] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0103] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0105] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A fake news detection method based on joint information propagation within and between graphs, characterized in that: include: Get the text to be detected and perform preprocessing; Based on the preprocessed text, text sequence features, text grammar features and text semantic features are extracted respectively, and then a text sequence graph, a text grammar graph and a text semantic graph are constructed; Based on the text sequence graph, text syntax graph and text semantic graph, the information interaction within the graph is first performed, and then the information interaction between graphs is performed to obtain the joint information text sequence graph, the joint information text syntax graph and the joint information text semantic graph; Based on the joint information text sequence graph, joint information text syntax graph and joint information text semantic graph, the vector representation of the text to be detected is obtained through the graph attention mechanism and the node attention mechanism; Input the vector representation of the text to be tested into the classifier to obtain the result of whether the text to be tested is fake news; The method for extracting the text sequence feature is as follows: using a sliding window mechanism to traverse the preprocessed text, and calculating the number of times each word pair appears in the same sliding window, and the number of times each word appears in all sliding windows; Based on the number of times each word pair appears in the same sliding window and the number of times each word appears in all sliding windows, obtaining a text sequence feature; The information interaction within the graph adopts a reset gate and an update gate to respectively update the node representations and the weights between the nodes in the text sequence graph, the text syntax graph and the text semantic graph; The inter-graph information interaction transmits information between the text sequence graph, the text syntax graph and the text semantic graph after the intra-graph information interaction.

2. The fake news detection method based on intra-graph and inter-graph joint information propagation as claimed in claim 1, characterized in that: The preprocessing is to perform data enhancement after normalizing the text to be detected.

3. The fake news detection method based on intra-graph and inter-graph joint information propagation as claimed in claim 2, characterized in that: The normalization process includes: removing special characters, word segmentation, deleting stop words, stemming and lowercasing words.

4. The fake news detection method based on intra-graph and inter-graph joint information propagation as claimed in claim 2, characterized in that: The data enhancement includes synonym replacement, random insertion, random swap and random deletion.

5. A fake news detection system based on joint information propagation within and between graphs, characterized by: include: The preprocessing module is configured to: obtain the text to be detected and perform preprocessing; A graph construction module is configured to: extract text sequence features, text grammar features and text semantic features based on the preprocessed text, and then construct a text sequence graph, a text grammar graph and a text semantic graph; The information propagation module is configured to: based on the text sequence graph, the text syntax graph and the text semantic graph, first perform intra-graph information interaction and then perform inter-graph information interaction to obtain a joint information text sequence graph, a joint information text syntax graph and a joint information text semantic graph; A vector representation module is configured to obtain a vector representation of the text to be detected through a graph attention mechanism and a node attention mechanism based on a joint information text sequence graph, a joint information text syntax graph, and a joint information text semantic graph; A classification module is configured to: input the vector representation of the text to be detected into a classifier to obtain a result of whether the text to be detected is fake news; The method for extracting the text sequence feature is as follows: using a sliding window mechanism to traverse the preprocessed text, and calculating the number of times each word pair appears in the same sliding window, and the number of times each word appears in all sliding windows; Based on the number of times each word pair appears in the same sliding window and the number of times each word appears in all sliding windows, obtaining a text sequence feature; The information interaction within the graph adopts a reset gate and an update gate to respectively update the node representations and the weights between the nodes in the text sequence graph, the text syntax graph and the text semantic graph; The inter-graph information interaction transmits information between the text sequence graph, the text syntax graph and the text semantic graph after the intra-graph information interaction.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the fake news detection method based on joint information propagation within and between graphs as described in any one of claims 1 to 4 are implemented.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the fake news detection method based on joint information propagation within and between graphs as described in any one of claims 1 to 4 are implemented.

Images