Language style fused depth-width multi-mode network health rumor detection method and device

Abstract

The invention provides a depth-width multi-mode network health rumor detection method and device fusing language styles, and the method comprises the steps: inputting to-be-recognized social media published content and publisher features into a pre-trained width module for extraction, and obtaining shallow features outputted by the width module; inputting the social media release content to be identified into a pre-trained depth module for extraction, and obtaining deep features which are output by the depth module and represent deep language content; and splicing the shallow-layer features and the deep-layer features, inputting the spliced features into a classifier to obtain the output of the classifier, and detecting the social media release content to be identified as real information, false information or true and false mixed information. According to the method, a social media health rumor detection model is constructed on the basis of a depth-width model, and language style features and deep semantic features of social media published content to be recognized are extracted on the basis of the Aridamod rhetorical theory and used for performing three-classification detection on health rumors.

Description

Technical Field

[0001] The present invention relates to the technical field of rumor detection, and in particular to a deep-wide multimodal network health rumor detection method and device integrating language style. Background Art

[0002] The spread of health rumors on the Internet poses a serious threat to public health and effective detection is urgently needed.

[0003] However, existing rumor detection research mainly focuses on areas such as fake news, fake reviews, and general social media rumors. Unlike fake news and other false information that requires exaggerated fabricated content to attract traffic, health rumors are mostly mixed true and false information.

[0004] Therefore, conventional rumor detection methods targeting fake news are not effective for detecting health rumors. Summary of the invention

[0005] The present invention provides a deep and wide multimodal network health rumor detection method and device integrating language style, so as to make up for the defect that the existing technology has poor detection effect on health rumors, and realize a social media health rumor detection method with higher accuracy and robustness.

[0006] The present invention provides a deep-wide multimodal network health rumor detection method integrating language style, comprising: Inputting the to-be-identified social media content and publisher features into a pre-trained width module for extraction, and obtaining shallow features output by the width module, wherein the shallow features include appeal to logic features, appeal to emotion features, and appeal to personality features; Inputting the social media content to be identified into a pre-trained deep module for extraction, and obtaining deep features representing deep semantics output by the deep module; The shallow features and the deep features are concatenated and input into a classifier to obtain an output of the classifier, and the social media content to be identified is detected as true information, false information, or a mixture of true and false information.

[0007] According to a deep-wide multimodal network health rumor detection method integrating language style provided by the present invention, The emotional appeal features include at least the number of exclamation marks, the number of emoticons, the number of emotional words corresponding to different emotional categories, and the emotional scores of the social media content to be identified; The emotion score corresponding to each emotion category is determined according to the occurrence frequency and emotion intensity of each emotion word of the corresponding emotion category, and the number of degree adverbs and negative words in the context of a preset length of each emotion word.

[0008] According to a deep, wide and multimodal network health rumor detection method integrating language style provided by the present invention, the appeal to logic features include at least the proportion of cognitive process words, the proportion of causal words, the number of functional words, the number of content words in the text of the social media content to be identified, and the picture-text consistency features of the social media content to be identified.

[0009] According to a deep-wide multimodal network health rumor detection method integrating language style provided by the present invention, In the case where the social media content to be identified contains both images and texts, each image and text content in the social media content to be identified is sequentially input into the Clip model to obtain an image-text consistency value of each image and text content output by the Clip model; Calculating the average of the image-text consistency values ​​of all images in the social media published content to be identified, and using the average as the image-text consistency feature of the social media published content to be identified; In the case that the social media content to be identified does not contain an image, an average of the image-text consistency features of any plurality of social media content containing images is used as the image-text consistency feature of the social media content to be identified.

[0010] According to a deep, wide and multimodal network health rumor detection method integrating language style provided by the present invention, the appeal to personality characteristics at least includes the publisher characteristics, the proportion of first-person pronouns and the number of medical terms of the social media content to be identified.

[0011] According to a deep-wide multimodal network health rumor detection method integrating language style provided by the present invention, before the step of inputting the to-be-identified social media publication content and publisher features into a pre-trained wide module for extraction, the method further includes: Collect multiple pieces of health rumor-debunking information from various rumor-debunking platforms, and use the content theme of each piece of health rumor-debunking information as a keyword to search and collect on social media, and obtain the published content data corresponding to the keyword, wherein each piece of the published content data includes the ID, text, and picture of the published content, as well as the ID, gender, number of fans, number of followers, total number of interactions, and VIP status of the publisher of the published content; After the collected published content data is cleaned, a portion of the cleaned published content data is randomly extracted as a data set used in the training process.

[0012] The present invention also provides a deep-wide multimodal network health rumor detection device integrating language style, comprising: A width module is used to input the social media content and publisher features to be identified into the pre-trained width module for extraction, and obtain shallow features output by the width module, wherein the shallow features include appeal to logic features, appeal to emotion features, and appeal to personality features; A deep module, used for inputting the social media publishing content to be identified into a pre-trained deep module for extraction, and obtaining deep features representing deep semantics output by the deep module; The classification module is used to splice the shallow features and the deep features and input them into a classifier to obtain the output of the classifier, and detect the social media content to be identified as true information, false information or a mixture of true and false information.

[0013] The present invention also provides an electronic device, including 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 deep-wide multimodal network health rumor detection method integrating language styles as described in any one of the above-mentioned methods is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the deep-wide multimodal network health rumor detection method integrating language styles as described in any one of the above-mentioned methods is implemented.

[0015] The present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements any of the deep, wide, multimodal network health rumor detection methods that integrate language styles as described above.

[0016] The present invention provides a deep-wide multimodal network health rumor detection method and device integrating language style. A social media health rumor detection model is constructed based on a deep-wide model. Based on Aristotle's rhetoric theory, the appeal to emotion features, appeal to personality features and appeal to logic features of the social media content to be identified are extracted as shallow features, and the deep features of the text information of the social media content to be identified are extracted. The deep features are spliced ​​with the shallow features and then input into a classifier to realize the prediction of whether the social media content to be identified is true, false or a mixture of true and false. The recognition accuracy of health rumors is improved by introducing the language style features of health rumors. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is one of the flow charts of the deep-wide multimodal network health rumor detection method integrating language style provided by the present invention; Figure 2 It is an architecture diagram of a corresponding model of a deep-wide multimodal network health rumor detection method integrating language style provided by the present invention; Figure 3 This is the second flow chart of the deep-wide multimodal network health rumor detection method integrating language style provided by the present invention; Figure 4 It is a structural schematic diagram of a deep-wide multimodal network health rumor detection device integrating language styles provided by the present invention; Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Combine the following Figures 1 to 3 The present invention introduces a deep and wide multimodal network health rumor detection method integrating language style, such as Figure 1 As shown, including: Step 101, inputting the to-be-identified social media content and publisher features into a pre-trained width module for extraction, and obtaining shallow features output by the width module, wherein the shallow features include appeal to logic features, appeal to emotion features, and appeal to personality features; Most current rumor detection models are built based on data and algorithm-driven methods. During the construction process, they mainly focus on the language content characteristics of rumors, but lack attention to language style characteristics. For health rumors, most current detection models are still based on traditional machine learning technology, focusing on language statistical features, network features, user features, interaction features, emotional features, etc.

[0021] However, health rumors are different from general social media rumors in that they are unique in terms of how they are generated, their content characteristics, and their dissemination mechanisms. That is, their writing needs to convince users of the accuracy of their content. Therefore, in the identification of health rumors, features that characterize the language style of text content can be introduced to improve the accuracy of health rumor identification.

[0022] On this basis, the present invention is based on Aristotle's rhetoric theory to extract features that characterize the language style of health rumors.

[0023] Specifically, Aristotle's rhetoric theory divides persuasive styles into three modes: appeal to personality (Ethos), appeal to logic (Logos) and appeal to emotion (Pathos), and points out that persuaders can use the above rhetorical triad to achieve the purpose of persuasion. Among them, appeal to personality is to establish trust by showing the persuader's specific personality traits such as credibility, reputation, moral character and professional knowledge, and is regarded as the most effective means of persuasion. High-level professional or credible communicators can effectively influence the audience and change their attitudes and behaviors towards the issues, products, information or individuals involved in the information. Appeal to logic is a persuasive method based on rationality and logic. It mainly refers to supporting opinions or positions by providing facts, data, statistics and reasonable arguments, so that the audience can evaluate the effectiveness of the arguments, which is regarded as the primary persuasive factor. Appeal to emotion is to influence the audience's opinions by arousing their emotions, aiming to stimulate certain feelings or emotions such as creating fear, arousing sympathy, anger or happiness to enhance the persuasive effect.

[0024] Therefore, based on Aristotle's rhetorical theory, appeal to personality, appeal to logic and appeal to emotion characteristics are extracted from the social media content to be identified and the publisher characteristics, so as to improve the accuracy of health rumor identification based on the above three types of characteristics.

[0025] Optionally, the present invention realizes the extraction of appeal to emotion features, appeal to personality features and appeal to logic features of social media content to be identified in the Chinese context based on the Wide & Deep Model.

[0026] like Figure 2 As shown in the figure, the deep-wide model includes a shallow part, namely the width module, which is mainly responsible for extracting shallow features in the data and memorizing the corresponding outputs in the historical data; while the deep part, namely the depth module, learns the nonlinear relationship and high-order feature combination of features through a neural network.

[0027] Therefore, by pre-training a deep-wide model, the features of the social media posts and publishers to be identified are extracted through its wide module, and the appeal to emotion features, appeal to personality features and appeal to logic features are obtained as shallow features.

[0028] It can be understood that the social media content to be identified is a piece of health-related information published by the publisher on the social media platform.

[0029] Optionally, the social media platform may be any type of social media platform that is primarily text-based, primarily image-based, or primarily video-based.

[0030] Optionally, the social media publishing content to be identified may be in the form of pure text, a mixture of text and image, or a mixture of text and video.

[0031] Optionally, the text in the social media published content to be identified may be text published directly in the form of text, or may be text published in the form of an image or a video.

[0032] Optionally, publisher features are used to characterize the identity portrait of the publisher. The selected publisher features can be adjusted accordingly for different social media platforms, but can generally include the number of followers of the publisher, as well as historical forwarding, commenting and / or likes data. The selection of specific data categories shall be based on the data that can be obtained by the corresponding social media platform.

[0033] Step 102, inputting the to-be-identified social media publication content into a pre-trained deep module for extraction, and obtaining deep features representing deep semantics output by the deep module; The deep module in the deep-wide model is used to extract deep features representing deep semantics from the input social media posts to be identified.

[0034] In one feasible implementation, the deep module uses the BERT (Bidirectional Encoder Representations from Transformers) pre-trained model to extract deep features.

[0035] The BERT model is a deep learning model that has made significant breakthroughs in the field of natural language processing and has powerful text comprehension capabilities. The model is based on the Transformer architecture and uses a two-way training method to fully capture the contextual information in the text, thereby generating richer word vector representations, and has therefore been widely used in rumor detection tasks.

[0036] The corresponding BERT model, in actual use, only inputs the text content in the social media posts to be identified into the deep module for extracting deep features.

[0037] Optionally, the text content may be text content directly published by a publisher.

[0038] Optionally, when the social media platform is a video-based platform, the text content may be the text content in the title and introduction of the video.

[0039] Optionally, when the social media published content to be identified is a mixed content of text and image published in the form of a long image, the text content in the long image is identified as the text content.

[0040] Step 103, concatenating the shallow features and the deep features and inputting them into a classifier to obtain an output of the classifier, and detecting the social media content to be identified as true information, false information, or a mixture of true and false information.

[0041] Alternatively, if Figure 2 As shown, a prediction module is designed to jointly learn and predict labels for deep features and shallow features.

[0042] First, the shallow features and deep features obtained by the width module and the depth module are mapped to the same dimension through two fully connected layers, and then the two vectors of the same dimension are concatenated to obtain a more comprehensive feature representation.

[0043] Furthermore, a self-attention layer is set in the prediction module to receive the spliced ​​features and dynamically adjust the importance of different features through a multi-head self-attention mechanism to capture the long-distance dependencies between features, so as to enhance the model's perception and representation capabilities of key information.

[0044] The core idea of ​​the multi-head self-attention mechanism is to independently calculate the weighted correlation between features in different subspaces to capture richer feature interaction information. By transforming the concatenated feature vectors, we get three vectors: query, key, and value. We then use the attention score to perform a weighted summation on the value vector to get a weighted representation of the feature.

[0045] On this basis, since health rumors often present the characteristics of a mixture of true and false information, different from the binary classification setting in the conventional rumor detection model, in order to solve the possible limitations of the generalization performance of the binary classification task to the real task, the present invention expands the rumor detection task into a multi-classification task, that is, the output detection results are true information, false information or true and false mixed information.

[0046] Therefore, an MLP (Multilayer Perceptron) is trained as a classifier in the prediction module. The MLP is used to learn the nonlinear relationship between features and map the processed features to the final prediction space to generate the final prediction results. The MLP includes three fully connected layers with 256, 64, and 3 neurons respectively, which reduce the feature dimension layer by layer and introduce the ReLU activation function between each layer.

[0047] In general, MLP maps the feature representation of the self-attention output to a 256-dimensional vector space through a linear layer, then applies the ReLU activation function on the vector and maps it to 64 dimensions through a linear layer, and finally outputs the final classification result through a 3-dimensional linear layer, that is, the social media content to be identified as true information, false information, or a mixture of true and false information.

[0048] The present invention constructs a social media health rumor detection model based on a deep-wide model, and based on Aristotle's rhetoric theory, extracts the appeal to emotion features, appeal to personality features and appeal to logic features of the social media content to be identified as shallow features, extracts the deep features of the text information of the social media content to be identified, maps the deep features and the shallow features to the same dimension through two fully connected layers, and then splices the two vectors of the same dimension. The spliced ​​vectors are input into a classifier to realize the classification of the social media content to be identified as true, false or a mixture of true and false, and the recognition accuracy of health rumors is improved by introducing the language style features of health rumors.

[0049] In the deep-wide multimodal network health rumor detection method integrating language style of the present invention, the appeal to emotion features at least include the number of exclamation marks, the number of emoticons, the number of emotion words corresponding to different emotion types and the emotion scores of the social media content to be identified; The emotion score corresponding to each emotion category is determined according to the occurrence frequency and emotion intensity of each emotion word of the corresponding emotion category, and the number of degree adverbs and negative words in the context of a preset length of each emotion word.

[0050] Optionally, appeal to emotion features are extracted from the text of the social media content to be identified. The definition of appeal to emotion features is shown in Table 1: Table 1

[0051] The appeal to emotion feature characterizes the language style feature of the appeal to emotion, taking into account the number and emotion scores of emotion words of different emotion categories contained in the published content to be identified. In a feasible implementation, the emotion words are determined based on an existing emotion dictionary.

[0052] For example, the text of the input social media content to be identified is segmented, and each segmented word is searched and matched with the sentiment dictionary to determine whether it belongs to a sentiment word and the corresponding sentiment category. Finally, the sentiment words corresponding to each sentiment category and the number of corresponding sentiment words are counted.

[0053] Since exclamation marks are usually used to express strong emotions or mood swings, the punctuation marks mainly extract the exclamation mark "!" and count the number of exclamation marks as one of the emotional features.

[0054] Emoticons are widely used as a medium for conveying emotions in online communication, and have become an explicit and fixed way of expressing emotions. Their number can reflect the intensity of the publisher's emotional expression, so the number of emoticons in the text is counted as one of the emotional features.

[0055] In addition, the sentiment score indicates the intensity of the publisher's sentiment level on each sentiment category.

[0056] In one possible implementation, the sentiment score is calculated as follows: For a given text T Specific emotions in e , calculate the text according to formula (1) T The i Words t i Sentiment score :

[0057] in, is the context window size, for The negation value of for The degree adverb value of is shown in formula (2). For words If the word Not in the sentiment dictionary middle, Recorded as 0, otherwise Record as the corresponding sentiment intensity value in the sentiment dictionary, and calculate Negative word value in context and degree adverb value , based on which the final sentiment score of the word is calculated .

[0058] Adverbs of degree are calculated by referring to the adverb entries in WordNet, as shown in formula (3); negative words are calculated by referring to the negative word list in NLTK, as shown in formula (4).

[0059] Then, the text T Each word in Sentiment score Add up and you get the text T Specific emotions e Sentiment score , as shown in formula (5): (5) Symbolic emotional features include the number of exclamation marks and the number of emojis. As important components of text, symbolic contents such as punctuation marks and emojis can complement the text's emotional features. For example, the exclamation mark ("!") is usually used to indicate strong emotional fluctuations, while emojis, as graphically emotional expression symbols widely used in online communication, can convey more subtle emotions, thus achieving the purpose of persuasion.

[0060] In the method for detecting health rumors in a deep and wide multi-modal network integrating language styles according to the present invention, the appeal-to-logic features at least include the proportion of cognitive process words, the proportion of causal words, the proportion of function words and the quantity in the text of the social media post to be recognized, and the text-image consistency feature of the social media post to be recognized.

[0061] Optionally, the appeal-to-logic features are extracted from the text and images of the social media post to be recognized, and the definition of the appeal-to-logic features is shown in Table 2: Table 2 ; The appeal-to-logic features represent the feature dimensions of the persuasive language style appealing to logic. Optionally, the proportion of causal words, the quantity of function words, the quantity of content words, the proportion of cognitive process words and the text-image consistency in the text are used as the appeal-to-logic features.

[0062] Specifically, causal words such as "because" and "so" can reflect how the publisher constructs arguments, connects viewpoints and shows causal relationships, which helps to understand the relationships between sentences and paragraphs in the text and shows the logic and coherence of the text. Therefore, the occurrence frequency or proportion of different types of causal words in the text is usually used to measure the logical structure of the text.

[0063] Function words refer to words that do not carry specific semantics in a sentence, such as "is" and "of". Although function words themselves do not have strong semantics, their use will affect the grammatical structure, causal relationship, coherence and fluency of the text, thus affecting the overall logic and rigor of the text.

[0064] Optionally, LIWC-2015 (Linguistic Inquiry and Word Count, a text analysis software) is used to calculate the proportion and quantity of 8 common types of function words in the Chinese scenario.

[0065] Cognitive process words (such as causal words and difference words) can indicate that the author is expressing logically strong ideas such as reasoning, analysis and causal relationships. Therefore, their proportion is related to the logic of the published content.

[0066] Content words (such as nouns, verbs, etc.) often carry more semantic information and can reflect the content density of the published content. More content words may mean that the published content is rich in information, more compact in structure, and conducive to logical expression.

[0067] The number and proportion of words in the above categories are obtained based on text statistics of the social media content to be identified.

[0068] In addition, since the consistency between images and texts represents the consistency of the information content expressed in the image and the text, it is generally believed that the higher the consistency between images and texts, the clearer the logic of the information, and the more conducive it is to the effective dissemination of information and the audience's understanding of the information.

[0069] Therefore, the consistency between image and text is taken as one of the logical features to calculate the consistency between the image and text of the social media posts to be identified.

[0070] Optionally, when there are multiple images in the social media publication to be identified, the image-text consistency between each image and the entire text is calculated, and then the image-text consistency average of the entire text is used as the image-text consistency of the social media publication to be identified.

[0071] Optionally, when there are multiple images in the social media content to be identified, and each image has clear text content corresponding to it, the image-text consistency of each image and the corresponding text content is calculated, and then the image-text consistency average of all texts is used as the image-text consistency of the social media content to be identified.

[0072] Optionally, when the social media content to be identified includes video and text, several key frames of the video are captured as images of the social media content to be identified, which are used to calculate the image-text consistency of the social media content to be identified.

[0073] On this basis, in a feasible implementation, the number and proportion of each functional word, the proportion of causal words, the proportion of cognitive process words, the number of content words and the consistency of pictures and texts are taken as the acquired appeal to logic features, that is, the proportion of causal words is counted separately. In other feasible implementations, similarly, the proportion of difference words in cognitive process words can also be counted separately, and / or the total proportion of all functional words can be counted as one of the output appeal to logic features, which can be adjusted according to the training results.

[0074] It should be noted that, unlike conventional rumor-debunking models that directly extract images from published content to obtain features of the image modality, the present invention takes into account that the semantic features of images alone are difficult to achieve accurate rumor detection. Therefore, it does not use pre-trained models such as ResNet to extract its features. Instead, it chooses to calculate the image-text consistency features of the image and text content, and uses this as the feature in the width module to assist in the rumor detection task.

[0075] In the deep-wide multimodal network health rumor detection method integrating language style of the present invention, when the social media published content to be identified contains both images and texts, each image and text content in the social media published content to be identified is sequentially input into the Clip model to obtain the image-text consistency value of each image and text content output by the Clip model; Calculating the average of the image-text consistency values ​​of all images in the social media published content to be identified, and using the average as the image-text consistency feature of the social media published content to be identified; Specifically, the Clip model is used to calculate the consistency between images and text and incorporate it into the model. Clip (Contrastive Language-Image Pre-training) is a multimodal model based on contrastive learning, which uses 400 million text-image pairs for pre-training. The Clip model has a shared embedding space that can embed images and text into the same space to calculate consistency.

[0076] When the social media posts to be identified contain both images and texts, the Clip model is used to calculate the image-text consistency value of each image and the complete text respectively, and then the image-text consistency values ​​of all images are averaged, and the obtained mean is used as the image-text consistency feature of the social media posts to be identified.

[0077] The above method can reduce the deviation caused by individual images and obtain a more stable and reliable similarity measurement.

[0078] In the case that the social media content to be identified does not contain an image, an average of the image-text consistency features of any plurality of social media content containing images is used as the image-text consistency feature of the social media content to be identified.

[0079] When the social media post to be identified does not contain an image, the average of the image-text consistency features of multiple other social media posts containing images is used for filling in order to ensure the integrity of the data and improve the robustness of the model so that it can be applied to detect the authenticity of social media posts with / without images at the same time.

[0080] In a specific implementation, during the training phase of the model, social media posts containing images are first used for training to obtain the mean image-text consistency of all social media posts containing images used for training. This mean is then used to fill in the image-text consistency features of social media posts that do not contain images during the detection process.

[0081] It can be understood that the image-text consistency feature represented by the above mean is not biased, that is, the image-text consistency feature of the content published on social media without images does not have a biased influence on the judgment of its authenticity.

[0082] In the deep-wide multimodal network health rumor detection method integrating language style of the present invention, the appeal to personality characteristics at least includes the publisher characteristics of the social media content to be identified, the proportion of first-person pronouns and the number of medical terms.

[0083] Optionally, publisher characteristics, first-person pronoun ratios, and the number of medical terms are extracted from the input publisher characteristics and the text information of the social media content to be identified as appeal to personality characteristics. The definition of appeal to personality characteristics is shown in Table 3: Table 3 ; Appeal to personality traits characterizes the characteristic dimensions of persuasive language style that appeals to personality, specifically taking into account the publisher’s characteristics, the proportion of the first person in the text of the social media posts to be identified, and the number of medical terms.

[0084] Optionally, publisher characteristics may include the number of fans, number of followers, gender, total number of interactions, authentication information and VIP status of the publisher on the corresponding social media platform, which is specifically determined based on the information that can be obtained by the corresponding social media platform.

[0085] These features can reflect the publisher’s status, social influence, and credibility on the platform. Among them, the number of followers of the publisher represents the publisher’s influence to a certain extent. Users with greater influence tend to maintain their user reputation and image, so their posts are more likely to be authentic.

[0086] Similarly, people who follow fewer people tend to reduce the false interactions that come with following each other, and are therefore generally considered to have greater autonomy and independence, and the information they post is more reliable.

[0087] Publishers of different genders often have different tendencies and preferences when publishing and processing information. Moreover, the gender of the publisher of information will directly affect the receiver's perception of the credibility and authority of the information, especially in the field of health.

[0088] In addition, if the content published by a publisher is often forwarded, commented on and liked by a large number of users, it means that the publisher has a certain degree of authority and influence in the field or community. Such publishers usually pay more attention to their own reputation and image, and the authenticity of their posts may be higher.

[0089] Publishers with certified information are generally considered to have higher authority and credibility in their professional fields, and their posts are less likely to be rumors.

[0090] VIP status refers to the user's subscription status, which to a certain extent reflects the publisher's cost investment in the platform. VIP users may have a higher marginal cost for spreading rumors and a lower probability of spreading rumors.

[0091] In addition, the proportion of first-person pronouns and the number of medical terms are calculated from the text information of the social media posts to be identified as personality traits.

[0092] The proportion of first-person pronouns can reflect the credibility and authority that the publisher is trying to demonstrate. Therefore, a higher proportion of first-person pronouns usually indicates that the publisher pays more attention to expressing personal positions in information transmission and tries to enhance the audience's trust through self-disclosure.

[0093] The number of medical terms describes the amount of health information contained in the published content, and to a certain extent can affect the audience's perception of the professionalism and authority of the information publisher.

[0094] The ability of the recipient will affect the trust level of the trustgiver. The more frequently medical terms are used, the more credible the publisher's ability tends to appear in the eyes of the audience, thereby enhancing the persuasiveness of health rumors.

[0095] On this basis, the number of fans, number of followers, gender, total number of interactions and authentication information of the publisher of the social media content to be identified, as well as the proportion of first-person pronouns and the number of medical terms are used as the appeal to personality characteristics.

[0096] In the deep-wide multimodal network health rumor detection method integrating language style of the present invention, before the step of inputting the to-be-identified social media publishing content and publisher features into the pre-trained wide module for extraction, the method further includes: Collect multiple pieces of health rumor-debunking information from various rumor-debunking platforms, and use the content theme of each piece of health rumor-debunking information as a keyword to search and collect on social media, and obtain the published content data corresponding to the keyword, wherein each piece of the published content data includes the ID, text, and picture of the published content, and one or more of the ID, number of fans, number of followers, gender, total number of interactions, authentication information, and VIP status of the publisher of the published content; Since the existing data sets are hardly suitable for training the model corresponding to the method of the present invention, it is necessary to construct a corresponding data set.

[0097] Specifically, first, multiple health and wellness rumor-debunking information are manually collected from various Internet joint rumor-debunking platforms and / or each social platform’s own official rumor-debunking platform. In this implementation, a total of 300 pieces of information are collected, and the collected rumor-debunking information is used as the basis for manual labeling.

[0098] Furthermore, the content theme of each rumor-refuting message was extracted and used as search keywords to construct a keyword list. Each keyword in the keyword list was used as a search keyword to search on the social media platform to obtain all the published content data related to the search keyword, and a total of 79,409 data were obtained to form the original data set.

[0099] Among them, each piece of published content data obtained includes the ID, text, pictures of the published content, and one or more of the ID, number of fans, number of followers, gender, total number of interactions, authentication information and VIP status of the publisher of the published content, which shall be based on what can be obtained by the corresponding social media platform.

[0100] At the same time, the image data is saved and numbered, and an index is built to associate and map each post text with the image to ensure the correlation between data of different modalities.

[0101] After the collected published content data is cleaned, a portion of the cleaned published content data is randomly extracted as a data set used in the training process.

[0102] To improve data quality, the data was preprocessed by deleting duplicate values, removing noise, and cleaning text. Then, 10% of the original data set was randomly sampled to generate 7,620 data to form the initial experimental data set, which was used as the data set for training deep and wide models and classifiers.

[0103] Furthermore, the initial experimental data set was annotated. In order to improve the accuracy of annotation and reduce the annotation deviation caused by subjective factors, the official rumor-refuting information was used as the annotation basis, and the cross-validation method was used for annotation.

[0104] First, five research assistants were fully trained to ensure that they were familiar with the labeling rules and all rumor-debunking information. Then, 50 posts were randomly selected for a pre-test. Five annotators were asked to read each post independently and judge the category of the post. During the labeling process, the annotators could not refer to the judgment of others. This was done to assess the annotators' understanding and consistency, and to fully discuss and resolve potential problems.

[0105] Based on the pre-test, five annotators conducted formal annotation. After the independent annotation, for samples with consistent annotations, the annotations of the annotators were directly used as the final labels of the samples; for samples with different annotation results, the five annotators fully discussed the medical big model and the opinions of field experts and finally reached a consistent label.

[0106] Finally, the experimental data set contained 2546 two-sided health information, 1830 health rumor information and 3244 real health information. The processed data set was randomly divided in a ratio of 8:2 to form the training set and test set required for model training.

[0107] On this basis, the overall process of the present invention is as follows Figure 3 shown.

[0108] In order to evaluate and test the performance of the model corresponding to the method of the present invention, a comparative experiment was set up during the experiment. The baseline model selected in the comparative experiment is: Traditional machine learning models: including two benchmark models, Random Forest (RF) and XGBoost, which mainly use shallow language style features for prediction.

[0109] Deep learning models: including two benchmark models, BERT and bi-LSTM, which mainly use deep language content for prediction.

[0110] LSTM-Attention: A rumor detection model that integrates content features and user features. The model uses LSTM based on the attention mechanism to extract deep semantic information of the text and perform task prediction.

[0111] HMCAN: A hierarchical multimodal contextual attention network rumor detection model. The model uses BERT and ResNet models to extract deep semantic features of text and image features respectively, then fuses features through a multimodal contextual attention mechanism and concatenates the fused features, and finally completes the classification task through a fully connected layer.

[0112] The above baseline model and model are tested on the dataset, and the results are shown in Table 4: Table 4 ; In the table, the MWDHRD-ART framework is the model corresponding to the method of the present invention. It can be found that the MWDHRD-ART framework proposed in the present invention is superior to the five benchmark models in all performance indicators. First, compared with the traditional machine learning methods RF and XGBoost, MWDHRD-ART improves the F1 value by 5.5% and 4.26%, respectively. Secondly, compared with the deep learning methods BERT and bi-LSTM, MWDHRD-ART improves the F1 value by 6.16% and 5.78%, respectively. Finally, compared with the two baseline models LSTM-Attention and HMCAN in existing studies, MWDHRD-ART improves the F1 value by 7.86% and 1.54%, respectively. Compared with LSTM-Attention and HMCAN, MWDHRD-ART uses both deep language content features and shallow language style features for prediction, especially considering the amount of medical information in the text, and is therefore more suitable for rumor identification tasks in the health field. The above results show that the MWDHRD-ART framework proposed in the present invention is effective for carrying out social media health rumor detection tasks.

[0113] It is worth mentioning that in the baseline model of the present invention, the F1 values ​​of the two machine learning models that only use shallow language style features are slightly higher than the F1 value of the deep learning model, which to a certain extent shows that the language style features extracted in this paper are of high quality. Therefore, in the face of the real-life scenario of a large number of tasks to be predicted, only by reasonably extracting lower-dimensional language style features can a health rumor detection effect similar to that based on the deep language content feature model be achieved. It can be seen that language style features have higher efficiency and practicality in health rumor detection tasks, especially when computing resources are limited, and their application prospects are even broader.

[0114] The deep, wide, multimodal network health rumor detection device with integrated language style provided by the present invention is described below. The deep, wide, multimodal network health rumor detection device with integrated language style described below and the deep, wide, multimodal network health rumor detection method with integrated language style described above can be referenced to each other.

[0115] like Figure 4 As shown, the deep-wide multimodal network health rumor detection device integrating language style includes a width module 401, a depth module 402 and a classification module 403: The width module 401 is used to input the social media content and publisher features to be identified into the pre-trained width module for extraction, and obtain the shallow features output by the width module, wherein the shallow features include appeal to logic features, appeal to emotion features and appeal to personality features; Most current rumor detection models are built based on data and algorithm-driven methods. During the construction process, they mainly focus on the language content characteristics of rumors, but lack attention to language style characteristics. For health rumors, most current detection models are still based on traditional machine learning technology, focusing on language statistical features, network features, user features, interaction features, emotional features, etc.

[0116] However, health rumors are different from general social media rumors in that they are unique in terms of how they are generated, their content characteristics, and their dissemination mechanisms. That is, their writing needs to convince users of the accuracy of their content. Therefore, in the identification of health rumors, features that characterize the language style of text content can be introduced to improve the accuracy of health rumor identification.

[0117] On this basis, the present invention is based on Aristotle's rhetoric theory to extract features that characterize the language style of health rumors.

[0118] Specifically, Aristotle's rhetoric theory divides persuasive styles into three modes: appeal to personality (Ethos), appeal to logic (Logos) and appeal to emotion (Pathos), and points out that persuaders can use the above rhetorical triad to achieve the purpose of persuasion. Among them, appeal to personality is to establish trust by showing the persuader's specific personality traits such as credibility, reputation, moral character and professional knowledge, and is regarded as the most effective means of persuasion. High-level professional or credible communicators can effectively influence the audience and change their attitudes and behaviors towards the issues, products, information or individuals involved in the information. Appeal to logic is a persuasive method based on rationality and logic. It mainly refers to supporting opinions or positions by providing facts, data, statistics and reasonable arguments, so that the audience can evaluate the effectiveness of the arguments, which is regarded as the primary persuasive factor. Appeal to emotion is to influence the audience's opinions by arousing their emotions, aiming to stimulate certain feelings or emotions such as creating fear, arousing sympathy, anger or happiness to enhance the persuasive effect.

[0119] Therefore, based on Aristotle's rhetorical theory, appeal to personality, appeal to logic and appeal to emotion characteristics are extracted from the social media content to be identified and the characteristics of the publisher, so as to improve the accuracy of health rumor identification based on the above three types of characteristics.

[0120] Optionally, the present invention realizes the extraction of appeal to emotion features, appeal to personality features and appeal to logic features of social media content to be identified in the Chinese context based on the Wide & Deep Model.

[0121] like Figure 2As shown in the figure, the deep-wide model includes a shallow part, namely the width module, which is mainly responsible for extracting shallow features in the data and memorizing the corresponding outputs in the historical data; while the deep part, namely the depth module, learns the nonlinear relationship and high-order feature combination of features through a neural network.

[0122] Therefore, by pre-training a deep-wide model, the features of the social media posts and publishers to be identified are extracted through its wide module, and the appeal to emotion features, appeal to personality features and appeal to logic features are obtained as shallow features.

[0123] It can be understood that the social media content to be identified is a piece of health-related information published by the publisher on the social media platform.

[0124] Optionally, the social media platform may be any type of social media platform that is primarily text-based, primarily image-based, or primarily video-based.

[0125] Optionally, the social media publishing content to be identified may be in the form of pure text, a mixture of text and image, or a mixture of text and video.

[0126] Optionally, the text in the social media published content to be identified may be text published directly in the form of text, or may be text published in the form of an image or a video.

[0127] Optionally, publisher features are used to characterize the identity portrait of the publisher. The selected publisher features can be adjusted accordingly for different social media platforms, but can generally include the number of followers of the publisher, as well as historical forwarding, commenting and / or likes data. The selection of specific data categories shall be based on the data that can be obtained by the corresponding social media platform.

[0128] A depth module 402 is used to input the social media content to be identified into a pre-trained depth module for extraction, and obtain deep features representing deep semantics output by the depth module; The deep module in the deep-wide model is used to extract deep features representing deep semantics from the input social media posts to be identified.

[0129] In one feasible implementation, the deep module uses the BERT (Bidirectional Encoder Representations from Transformers) pre-trained model to extract deep features.

[0130] The BERT model is a deep learning model that has made significant breakthroughs in the field of natural language processing and has powerful text comprehension capabilities. The model is based on the Transformer architecture and uses a two-way training method to fully capture the contextual information in the text, thereby generating richer word vector representations, and has therefore been widely used in rumor detection tasks.

[0131] The corresponding BERT model, in actual use, only inputs the text content in the social media posts to be identified into the deep module for extracting deep features.

[0132] Optionally, the text content may be text content directly published by a publisher.

[0133] Optionally, when the social media platform is a video-based platform, the text content may be the text content in the title and introduction of the video.

[0134] Optionally, when the social media published content to be identified is a mixed content of text and image published in the form of a long image, the text content in the long image is identified as the text content.

[0135] The classification module 403 is used to splice the shallow features and the deep features and input them into a classifier to obtain the output of the classifier, and detect the social media content to be identified as true information, false information or a mixture of true and false information.

[0136] Alternatively, if Figure 2 As shown, a prediction module is designed to jointly learn and predict labels for deep features and shallow features.

[0137] First, the shallow features and deep features obtained by the width module and the depth module are mapped to the same dimension through two fully connected layers, and then the two vectors of the same dimension are concatenated to obtain a more comprehensive feature representation.

[0138] Furthermore, a self-attention layer is set in the prediction module to receive the spliced ​​features and dynamically adjust the importance of different features through a multi-head self-attention mechanism to capture the long-distance dependencies between features, so as to enhance the model's perception and representation capabilities of key information.

[0139] The core idea of ​​the multi-head self-attention mechanism is to independently calculate the weighted correlation between features in different subspaces to capture richer feature interaction information. By transforming the concatenated feature vectors, we get three vectors: query, key, and value. We then use the attention score to perform a weighted summation on the value vector to get a weighted representation of the feature.

[0140] On this basis, since health rumors often exhibit the characteristics of a mixture of true and false information, different from the binary classification setting in conventional rumor detection models, in order to solve the possible limitations of the generalization performance of binary classification tasks for real tasks, the present invention expands the rumor detection task into a multi-classification task, that is, the output detection results are true information, false information, or true and false mixed information.

[0141] Therefore, an MLP (Multilayer Perceptron) is trained as a classifier in the prediction module. The MLP is used to learn the nonlinear relationship between features and map the processed features to the final prediction space to generate the final prediction results. The MLP includes three fully connected layers with 256, 64, and 3 neurons respectively, which reduce the feature dimension layer by layer and introduce the ReLU activation function between each layer.

[0142] In general, MLP maps the feature representation of the self-attention output to a 256-dimensional vector space through a linear layer, then applies the ReLU activation function on the vector and maps it to 64 dimensions through a linear layer, and finally outputs the final classification result through a 3-dimensional linear layer, that is, the social media content to be identified as true information, false information, or a mixture of true and false information.

[0143] The present invention constructs a social media health rumor detection model based on a deep-wide model, and based on Aristotle's rhetoric theory, extracts the appeal to emotion features, appeal to personality features and appeal to logic features of the social media content to be identified as shallow features, extracts the deep features of the text information of the social media content to be identified, maps the deep features and the shallow features to the same dimension through two fully connected layers, and then splices the two vectors of the same dimension obtained. The spliced ​​vectors are input into a classifier to realize the detection of true, false or mixed true and false content of the social media content to be identified, and improves the recognition accuracy of health rumors by introducing the language style features of health rumors.

[0144] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the social media health rumor detection method based on Aristotle's rhetoric theory, the method comprising: inputting the social media publishing content to be identified and the publisher's features into a width module for extraction, and obtaining the shallow features output by the width module, wherein the shallow features include appeal to logic features, appeal to emotion features and appeal to personality features; inputting the social media publishing content to be identified into a pre-trained deep module for extraction, and obtaining the deep features representing the deep semantics output by the deep module; splicing the shallow features and the deep features and inputting them into a classifier to obtain the output of the classifier, and detecting the social media publishing content to be identified as true information, false information or true and false mixed information.

[0145] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0146] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the deep-wide multimodal network health rumor detection method that integrates language styles provided by the above-mentioned methods, and the method includes: inputting the social media content to be identified and the publisher characteristics into a pre-trained wide module for extraction, and obtaining the shallow features output by the wide module, wherein the shallow features include appeal to logic features, appeal to emotion features, and appeal to personality features; inputting the social media content to be identified into a pre-trained deep module for extraction, and obtaining the deep features representing deep semantics output by the deep module; splicing the shallow features and the deep features and inputting them into a classifier to obtain the output of the classifier, and detecting the social media content to be identified as true information, false information, or a mixture of true and false information.

[0147] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the deep-wide multimodal network health rumor detection method that integrates language styles provided by the above-mentioned methods, the method comprising: inputting the social media content to be identified and the publisher's features into a pre-trained wide module for extraction, and obtaining shallow features output by the wide module, wherein the shallow features include appeal to logic features, appeal to emotion features, and appeal to personality features; inputting the social media content to be identified into a pre-trained deep module for extraction, and obtaining deep features representing deep semantics output by the deep module; splicing the shallow features and the deep features and inputting them into a classifier to obtain the output of the classifier, and detecting the social media content to be identified as true information, false information, or a mixture of true and false information.

[0148] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, i.e., they may be located in one place or distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement the scheme without creative effort.

[0149] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep-wide multimodal network health rumor detection method integrating language style, characterized in that: include: Inputting the to-be-identified social media content and publisher features into a pre-trained width module for extraction, and obtaining shallow features output by the width module, wherein the shallow features include appeal to logic features, appeal to emotion features, and appeal to personality features; Inputting the social media content to be identified into a pre-trained deep module for extraction, and obtaining deep features representing deep semantics output by the deep module; The shallow features and the deep features are concatenated and input into a classifier to obtain an output of the classifier, and the social media content to be identified is detected as true information, false information, or a mixture of true and false information.

2. The deep-wide multimodal network health rumor detection method integrating language style according to claim 1 is characterized in that: The emotional appeal features include at least the number of exclamation marks, the number of emoticons, the number of emotional words corresponding to different emotional categories, and the emotional scores of the social media content to be identified; The emotion score corresponding to each emotion category is determined according to the occurrence frequency and emotion intensity of each emotion word of the corresponding emotion category, and the number of degree adverbs and negative words in the context of a preset length of each emotion word.

3. The deep-wide multimodal network health rumor detection method integrating language style according to claim 1 is characterized in that: The appeal to logic features include at least the proportion of cognitive process words, the proportion of causal words, the number of functional words, the number of content words in the text of the social media content to be identified, and the picture-text consistency feature of the social media content to be identified.

4. The deep-wide multimodal network health rumor detection method integrating language style according to claim 3 is characterized in that: In the case where the social media content to be identified contains both images and texts, each image and text content in the social media content to be identified is sequentially input into the Clip model to obtain an image-text consistency value of each image and text content output by the Clip model; the average of the image-text consistency values ​​of all images in the social media content to be identified is calculated, and the average is used as the image-text consistency feature of the social media content to be identified; In the case that the social media content to be identified does not contain an image, an average of the image-text consistency features of any plurality of social media content containing images is used as the image-text consistency feature of the social media content to be identified.

5. The deep-wide multimodal network health rumor detection method integrating language style according to claim 1 is characterized in that: The appeal to personality characteristics at least includes publisher characteristics, the proportion of first-person pronouns, and the number of medical terms of the social media content to be identified.

6. The deep-wide multimodal network health rumor detection method integrating language style according to any one of claims 1 to 5 is characterized in that: Before the step of inputting the to-be-identified social media content and publisher features into the pre-trained width module for extraction, the method further includes: Collect multiple pieces of health rumor-debunking information from various rumor-debunking platforms, and use the content theme of each piece of health rumor-debunking information as a keyword to search and collect on social media, and obtain the published content data corresponding to the keyword, wherein each piece of the published content data includes the ID, text, and picture of the published content, as well as the ID, gender, number of fans, number of followers, total number of interactions, and VIP status of the publisher of the published content; After the collected published content data is cleaned, a portion of the cleaned published content data is randomly extracted as a data set used in the training process.

7. A deep and wide multimodal network health rumor detection device integrating language style, characterized in that: include: A width module is used to input the social media content and publisher features to be identified into the pre-trained width module for extraction, and obtain shallow features output by the width module, wherein the shallow features include appeal to logic features, appeal to emotion features, and appeal to personality features; A deep module, used for inputting the social media publishing content to be identified into a pre-trained deep module for extraction, and obtaining deep features representing deep semantics output by the deep module; The classification module is used to splice the shallow features and the deep features and input them into a classifier to obtain the output of the classifier, and detect the social media content to be identified as true information, false information or a mixture of true and false information.

8. An electronic 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 deep, wide, multimodal network health rumor detection method integrating language styles as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the deep, wide and multimodal network health rumor detection method integrating language styles as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the deep, wide and multimodal network health rumor detection method integrating language styles as described in any one of claims 1 to 6 is implemented.

Images