A method and system for detecting false information based on comment features

Abstract

The present invention discloses a method and system for detecting false information based on comment features. The method comprises obtaining a model training data set, wherein the model training data set includes the comment content and comment time of each blog post; performing feature extraction on the data set to obtain comment external features, comment language features and comment sentiment features; respectively using the comment external features, the comment language features and the comment sentiment features to select three models with the best prediction performance from a plurality of preset models as three basic models; and using the outputs of the three basic models to select one model with the best prediction performance from a plurality of preset models as a fusion model; and using the multi-feature false information classification model to perform false information detection. The solution of the present invention achieves feature supplementation for existing false information detection by capturing the external features, internal features and fusion features of false information comments, provides a richer detection perspective, and effectively improves the performance of false information detection tasks on social media.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning and natural language processing, and in particular to a method and system for detecting false information based on comment features. Background Art

[0002] The rapid development of the internet has fostered a boom in information sharing on social media. However, as a crucial platform for acquiring information and exchanging opinions, social media has also provided a breeding ground for the dissemination and spread of misinformation. The vast number of platforms and categories, as well as its easily accessible nature, allow misinformation to spread rapidly on social media, impacting our lives in multiple ways. Existing techniques primarily rely on a multi-feature fusion approach to detect misinformation, including rule recognition, calculation of misinformation text features, user reputation features within social media, and images linked to posts. As we all know, social media is not solely about its dissemination properties; interaction is a key application scenario. The amount of information carried by misinformation itself is easily assimilated, making it difficult to identify. From a dissemination perspective, detecting misinformation using forwarded or commented content on social media can capture more features of misinformation. Commenters' opinions can be considered a supplement to the original data. Furthermore, forwarded and commented content often contain the participants' genuine opinions and emotions. Structured comment information provides a more stable feature representation for identifying misinformation.

[0003] Currently, feature mining based on structured comment information primarily relies on simple sentiment polarity analysis and information processing based on text statistics. These methods are insufficient in capturing the subtle differences and deeper features of comments. In particular, this approach overlooks key factors such as the temporal dimension and diverse emotional expressions carried externally in comments. Due to the lack of effective development of these complex features, the accuracy of identifying multi-topic misinformation on social media has been significantly affected. Furthermore, there is a lack of effective grouping and model validation of the hierarchical features of machine learning models. To improve recognition accuracy and efficiency, future research needs to explore and integrate more diverse data features, including time series analysis and multidimensional sentiment analysis, to fully capture and utilize the rich information in social media data. Summary of the Invention

[0004] In order to solve the technical problem that false information detection is insufficient in capturing subtle differences and deep-level features of comments, an embodiment of the present invention provides a false information detection method and system based on comment features.

[0005] The technical solution of the embodiment of the present invention is achieved as follows:

[0006] An embodiment of the present invention provides a method for detecting false information based on comment features, the method comprising:

[0007] Obtain a model training dataset, which includes the comment content and comment time of each blog post; perform feature extraction on the dataset to obtain comment external features, comment language features, and comment sentiment features; the comment external features are basic features of blog post comments, representing the quality or popularity of the blog post; the comment language features are text features of the blog post comments, representing the viewpoint preference of the blog post response; the comment sentiment features represent the sentiment representation carried in the blog post comments;

[0008] The three models with the best prediction performance are selected from the preset multiple models using the external features of the comments, the language features of the comments, and the emotional features of the comments respectively as the three basic models; and the one model with the best prediction performance is selected from the preset multiple models using the outputs of the three basic models as the fusion model; a multi-feature false information classification model is constructed; the multi-feature false information classification model includes three basic models and one fusion model; the three basic models are respectively used to output a first prediction probability for the input external features of the comments, a second prediction probability for the input language features of the comments, and a third prediction probability for the input emotional features of the comments; the fusion model is used to output a final prediction probability based on the first prediction probability, the second prediction probability, and the third prediction probability output by the three basic models respectively;

[0009] Utilize the multi-feature false information classification model to perform false information detection.

[0010] In one embodiment, obtaining a model training dataset includes:

[0011] Get the comment content and comment time of each blog post based on the link of each blog post;

[0012] The comment content and comment time corresponding to the blog post are taken as a piece of social media blog post data to construct a social media blog post dataset;

[0013] The authenticity of false information in the social media blog data set is marked using labels to obtain a model training data set.

[0014] In one embodiment, the external features of comments include comment strength and comment speed; the comment strength is used to measure the degree of interaction triggered after the blog post is published; it is expressed as ; Indicates the total number of comments;

[0015] The comment speed is used to represent the average time required for a blog post to be commented on; the calculation formula is:

[0016]

[0017] in, Indicates blog post Release time, Comment Release time, and Respectively represent blog posts The latest and earliest comment times among the comments.

[0018] In one embodiment, the comment language features include text length, word frequency, number of sentences, number of emoticons, number of @ symbols used, and topic tags;

[0019] The calculation formula of the text length is:

[0020]

[0021] in, Indicates the total number of comments. Comment The number of bytes;

[0022] The calculation formula of the word frequency is:

[0023]

[0024] in, Comment The number of valid words;

[0025] The calculation formula for the number of sentences is:

[0026]

[0027] in, Comment The number of sentences;

[0028] The calculation formula for the number of emoticons is:

[0029]

[0030] in, Comment The number of emojis;

[0031] The calculation formula for the number of AIT symbols used is:

[0032]

[0033] in, Comment Contains the number of "@";

[0034] The calculation formula for the topic labeling is:

[0035]

[0036] in, Comment The number of topics marked in .

[0037] In one embodiment, the comment sentiment features include positive, negative, and seven main classification sentiment features, and the seven main classification sentiment features include happiness, good, anger, sadness, fear, disgust, and shock.

[0038] In one embodiment, the preset multiple models include: random forest model, extreme gradient boosting model, gradient boosting machine model, adaptive boosting model, K nearest neighbor model, support vector machine model, decision tree model, naive Bayes model and logistic regression model.

[0039] In one embodiment, the three models with the best prediction performance are selected from a plurality of preset models using the external features of the reviews, the language features of the reviews, and the sentiment features of the reviews as the three basic models, including:

[0040] The AUC scoring method was used to evaluate the model performance, and the three models with the best prediction performance were selected from multiple preset models based on the AUC scoring results as the three basic models.

[0041] In one embodiment, the performance of the multi-feature false information classification model is evaluated by accuracy, precision, recall, and F1 score metrics.

[0042] An embodiment of the present invention also provides a false information detection system based on comment features, comprising: a processor and a memory for storing a computer program that can be run on the processor; wherein, when the processor is used to run the computer program, it executes the steps of the above-mentioned method.

[0043] The embodiments of the present invention have the following beneficial effects:

[0044] This embodiment, by comprehensively utilizing the time dimension information and multi-dimensional emotional expressions in comments, can not only identify subtle differences that are easily overlooked in traditional methods, but also explore and integrate the deep-level characteristics of comments, achieve in-depth mining and comprehensive application of comment features, and significantly improve the accuracy and efficiency of false information detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the flow of a method for detecting false information based on comment features according to an embodiment of the present invention;

[0046] Figure 2 This is a diagram of the internal structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

[0048] The embodiment of the present invention provides a method for detecting false information based on comment features, such as Figure 1 As shown, the method includes:

[0049] Step 101: Obtain a model training dataset, which includes the comment content and comment time of each blog post; perform feature extraction on the dataset to obtain comment external features, comment language features, and comment sentiment features; the comment external features are basic features of blog post comments, representing the quality or popularity of the blog post; the comment language features are text features of the blog post comments, representing the opinion preferences of the blog post response; and the comment sentiment features represent the sentiment representation carried in the blog post comments;

[0050] Step 102: Select three models with the best prediction performance from a plurality of preset models using the external features of the comments, the language features of the comments, and the sentiment features of the comments respectively as three basic models; and select one model with the best prediction performance from a plurality of preset models using the outputs of the three basic models as a fusion model; construct a multi-feature false information classification model; the multi-feature false information classification model includes three basic models and one fusion model; the three basic models are respectively used to output a first prediction probability for the input external features of the comments, a second prediction probability for the input language features of the comments, and a third prediction probability for the input sentiment features of the comments; the fusion model is used to output a final prediction probability based on the first prediction probability, the second prediction probability, and the third prediction probability outputted by the three basic models;

[0051] Step 103: Utilize the multi-feature false information classification model to perform false information detection.

[0052] The method of this embodiment captures the external features, internal features, and fusion features of false information comments to supplement the features of existing false information detection, provide a richer detection perspective, and effectively improve the performance of false information detection tasks on social media.

[0053] Specifically, when obtaining the model training data set, this embodiment can crawl the corresponding comment content and comment time according to the link of each blog post, and mark it according to the post ID to ensure that it matches the original text; then use labels to mark the true attributes of the false information in each blog post, thereby generating a model training data set.

[0054] This embodiment obtains three types of features based on the model training data set: review external features, review language features, and review sentiment features.

[0055] Specifically, the comment external features represent the basic features of blog comments, reflecting the quality or popularity of the blog. It includes two sub-indicators: comment intensity and comment speed. Among them, the comment intensity is used to measure the degree of interaction triggered after the blog is published, and is measured by the number of comments. The comment speed describes the average time required from the publication of the blog to the time when the blog is commented, and is calculated by using the comment interval time in the blog comments and the time difference between the latest and earliest comment times.

[0056] The comment language features represent the text features of blog comments, reflecting the opinion preferences of blog responses. The comment language features include six sub-indicators: text length feature, word frequency feature, sentence number feature, number of emojis, number of "@" used, and topic annotation. Among them, the text length feature is measured by the total number of bytes of the non-segmented text in all the comments of the blog. The word frequency feature is measured by the number of effective words after processing with the Chinese jieba segmentation package and stop words in all the comments of the blog. The sentence number feature is measured by counting the number of sentences after splitting the comment text with full stops, question marks, exclamation marks, etc. (.,!,?,!?…), and then removing the blank sentences. The number of emojis is measured by using a self-built regular expression to match the emojis appearing in the comment text and counting the number of emojis. The number of "@" used is counted by matching the "@" symbol in the text. The topic annotation is measured by using a regular expression to match and then using the quantity statistical measure.

[0057] The comment sentiment feature describes the emotional representation carried in Chinese expressions. In this embodiment, by introducing the Chinese emotion word ontology library, the positive, negative, and seven main classification emotion features of "happy", "good", "angry", "sad", "fearful", "hateful", "surprised" existing in social media are calculated and classified.

[0058] That is, the comment sentiment feature includes the following nine indicators:

[0059] "Happy" represents the perceived intensity of a pleasant mood in blog comments in the Chinese context, with example words such as "joy" and "relieved", etc.;

[0060] "Good" represents the perceived intensity of approval and agreement in blog comments in the Chinese context, with example words such as "respectful" and "admiring", etc.; ;

[0061] "Angry" represents the perceived intensity of emotional excitement in blog comments in the Chinese context, with example words such as "angry" and "annoyed", etc.;

[0062] "Sad" represents the perceived intensity of a low mood in blog comments in the Chinese context, with example words such as "sad" and "guilty", etc.;

[0063] "Fear" represents the perceived intensity of danger or threat in blog comments in the Chinese context, with example words such as "flustered" and "timid";

[0064] "Hatred" represents the perceived intensity of unpleasant things in blog comments in the Chinese context, with example words such as "irritated" and "反感";

[0065] "Surprise" represents the perceived intensity of the reaction to unexpected events in blog comments in the Chinese context, with example words such as "strange" and "miracle";

[0066] Positive sentiment polarity is the integration of the lexicons of "joy", "good", and "surprise";

[0067] Negative sentiment polarity is the integration of the lexicons of "anger", "sorrow", "fear", and "hatred".

[0068] This embodiment uses the following sentiment analysis steps to obtain the above nine indicators: perform Chinese word segmentation using jieba.lcut. Create and maintain a set of unique words and their frequencies. Calculate the occurrence frequency of each word by comparing the words with predefined sentiment lists (such as positive and negative). Count the occurrence times of each sentiment for sentiment tendency analysis.

[0069] The above three types of features: comment external features, comment language features, and comment sentiment features will be respectively input into 3 basic models for training. The set of predicted probabilities output by the 3 trained basic layer models will be used as input features to train the fusion model.

[0070] That is, this embodiment uses a two-layer model structure for fused feature learning. One layer is 3 basic models. The second layer is the fusion model.

[0071] This embodiment will respectively input the above three types of features into the following 9 excellent machine learning models and select the model with the best performance: Random Forest, eXtreme Gradient Boosting, Gradient Boosting Machine, Adaptive Boosting, K-Nearest Neighbors, Support Vector Machine, Decision Tree, Naïve Bayes, Logistic Regression.

[0072] The first-layer basic model trains machine learning models based on the review external feature set, review language feature set, and review sentiment feature set. The second-layer fusion model uses the predicted probabilities output by the first-layer basic model as input to train the final machine learning model.

[0073] Specifically, this embodiment uses the same training data (75% of the data) and test data (25% of the data) to build the model.

[0074] For the first-tier model performance, K-fold cross-validation (K=5) was used for each model, and the AUC score was used to evaluate model performance. The overall model performance was evaluated using accuracy, precision, recall, and F1 score metrics to compare and evaluate model performance.

[0075] Accuracy evaluates the accuracy of the model's predictions, that is, the ratio of correct predictions to the total number of observations. Precision measures the proportion of samples predicted to be positive that are actually positive, and is calculated as the number of true positives (TP) divided by the number of all predicted positives (TP + FP). Recall measures the proportion of actual positive samples identified by the model, that is, the number of true positives (TP) divided by the total number of actual positives (TP + FN). The F1 score is the harmonic mean of precision and recall, and is used to comprehensively consider precision and recall.

[0076] In this example, the accuracy metric is used to primarily evaluate the prediction accuracy of each machine learning model. In addition, three other metrics will be used to comprehensively evaluate the overall performance of these models. Finally, the advantages of a multi-feature misinformation classification scheme based on a stacking structure for misinformation detection will be explained. This will be applied as a misinformation detection model based on review features.

[0077] The present embodiment will be described below with a specific example.

[0078] In one application scenario, this embodiment includes a social media data collection module, a feature selection and calculation module based on false information comments, a multi-feature false information classification model design module based on a Stacking structure, and a false information detection optimal model evaluation module.

[0079] Social media data collection module, used to collect and crawl social media blog data.

[0080] The social media data collection module requires crawling the corresponding comments content and comment time based on the link of each blog post, and marking them according to the post ID to ensure that they match the original text;

[0081] The social media data collection module uses labels to mark the true and false attributes of each blog post and generate classified data groups;

[0082] The social media data collection module ultimately generates a dataset that can be used for model learning and testing.

[0083] In addition, the complete process of this embodiment is:

[0084] S1: Obtain social media blog post data, and crawl the corresponding comment content and comment time based on the link of each blog post, and mark them according to the post ID to ensure that they match the original text.

[0085] S11: Use the authenticity of false information to label each blog post as false information / non-false information to construct a false information detection dataset.

[0086] S12: This embodiment uses an existing dataset of false information from Weibo, and statistics on samples in the dataset are shown in Table 1:

[0087] Table 1

[0088]

[0089] S2: Feature selection and calculation based on false information comments. In this step, three types of features from blog comments are selected: external features of comments, language features of comments, and sentiment features of comments. The specific features and descriptions are shown in Table 2:

[0090] Table 2

[0091]

[0092] S21: Comment external feature categories represent the basic features of blog comments and reflect the quality or popularity of blog posts.

[0093] Specifically, comment intensity is used to measure the degree of interaction triggered by a blog post, expressed as , Indicates the total number of comments.

[0094] Comment velocity describes the average time it takes for a blog post to be commented on and is calculated as follows:

[0095]

[0096] in, Indicates blog post Release time, Comment Release time, and Indicates blog post The latest and earliest comment times among the comments.

[0097] S22: The comment language feature category represents the textual features of the blog comments and reflects the opinion preferences of the blog responses.

[0098] Specifically, the text length is the sum of the number of bytes in all the comments in the blog post without word segmentation, which is recorded as , The number of bytes per comment.

[0099] The word frequency is the number of valid words in all the comments of the blog post after being processed by the Chinese Jieba word segmentation package and stop words, recorded as , Indicates the number of valid words in each comment.

[0100] Sentence count: Segment the review text by period, question mark, exclamation mark, etc. (.!?!?...), and then remove blank sentences. The statistical result of the number of sentences is expressed as: , Indicates the number of sentences in each review.

[0101] Number of emojis, using a self-built regular expression Match the emojis that appear in the comment text. The cumulative number of emojis is expressed as: , Indicates the number of emojis per comment.

[0102] The number of "@" is counted by matching the "@" symbol in the text, expressed as: , Indicates the number of "@" characters in each comment.

[0103] Topic annotation, that is, the topic tags introduced in the comment text. There can be multiple topic annotations, using regular expressions Match. Expressed as: , Indicates the number of topics tagged in each comment.

[0104] S23: The comment sentiment feature category describes the sentiment representation carried in Chinese expressions. This embodiment introduces a Chinese sentiment vocabulary ontology library to calculate and classify positive, negative and seven main classification sentiment features existing in social media.

[0105] This lexicon divides emotional expressions in detail, classifying emotions into seven main categories: "happiness", "good", "anger", "sorrow", "fear", "disgust", "surprise", and further subdividing them into 21 subcategories. The emotional intensity of emotional words has five levels: 1, 3, 5, 7, 9. In addition, emotional polarity is divided into neutral, positive, and negative, corresponding to the values 0, 1, 2 respectively. This dictionary includes 27,466 entries, and each entry includes the word, the type of word part, the number of word meanings, the meaning serial number, the main emotional category, the emotional intensity, the emotional polarity, and the auxiliary emotional category.

[0106] In the emotional dictionary, emotional words have multiple emotional expressions. The main emotional category describes the main emotional tendency of the word, while the auxiliary emotional category includes other possible emotional aspects. There are seven types of word parts in the emotional vocabulary ontology: noun (noun), verb (verb), adjective (adj), adverb (adv), network word (nw), idiom (idiom), prepositional phrase (prep). Each word under each emotional category has a definite polarity. In the polarity marking, 3 represents that the word has both positive and negative genders. This system also specifically marks negative words and degree adverbs. Negative words can reverse the emotional intensity, and degree adverbs indicate different levels of emotional tendency.

[0107] The technical steps of emotional analysis include:

[0108] Use jieba.lcut for Chinese word segmentation.

[0109] Create and maintain a set of unique words and their frequencies.

[0110] Calculate the occurrence frequency of each word by comparing the word with a predefined emotional list (such as positive and negative).

[0111] Count the occurrence times of each emotion for analyzing the emotional tendency.

[0112] This method provides a systematic framework to analyze the emotional color in the text in a scientific way, which helps to accurately understand and process the language expressions on social media. The calculation method for each emotion is as follows.

[0113] "Happiness" represents the perceived intensity of the pleasant mood in the blog post comments in the Chinese scenario. Example words such as "joy" and "relieved", are recorded as:

[0114]

[0115] "Good" represents the perceived intensity of approval and agreement in the blog post comments in the Chinese scenario. Example words such as "respectful" and "admiring", are recorded as:

[0116]

[0117] "Anger" represents the intensity of perceived emotional agitation in blog post comments in the Chinese context. Example words include "indignant" and "irritated", etc., denoted as:

[0118]

[0119] "Sadness" represents the intensity of perceived low mood in blog post comments in the Chinese context. Example words include "sorrowful" and "guilty", etc., denoted as:

[0120]

[0121] "Fear" represents the intensity of perceived danger or threat in blog post comments in the Chinese context. Example words include "flustered" and "timid", etc., denoted as:

[0122]

[0123] "Disgust" represents the intensity of perceived unpleasant things in blog post comments in the Chinese context. Example words include "fidgety" and "反感" (it seems there is a Chinese word here, if it's a typo, please correct), etc., denoted as:

[0124]

[0125] "Surprise" represents the intensity of perceived reaction to unexpected events in blog post comments in the Chinese context. Example words include "strange" and "miracle", etc., denoted as:

[0126]

[0127] Positive emotional polarity is the integration of the lexicons of "Joy", "Good", and "Surprise", denoted as:

[0128]

[0129] Among them, Positive includes words classified as Happy, Good, and Surprise.

[0130] Negative emotional polarity is the integration of the lexicons of "Anger", "Sadness", "Fear", and "Disgust", denoted as:

[0131]

[0132] Among them, Negative includes words classified as Anger, Sad, Fear, and Disgust.

[0133] S3: Design of a multi-feature misinformation classification model based on a stacking architecture. This step uses a two-layer model structure. The first layer serves as the base model, training multiple machine learning models based on feature groups and selecting the model with the best predictive performance for each feature group. The second layer serves as the meta model, using the prediction results of the first layer model as input to train the final classifier.

[0134] S31: First-level Base Model training step. Machine learning models are trained based on the feature group classifications—external review feature group, language feature group, and sentiment feature group. The model with the best prediction performance in each feature group is selected.

[0135] According to the features calculated in step S2, the input feature group consists of three types of features: review external features, review language features, and review sentiment features:

[0136]

[0137] in, , , . As the target variable, it identifies whether the information is false.

[0138] As a binary classification task, false information detection uses nine high-performing machine learning models as the basic model training: Random Forest, eXtreme Gradient Boosting, Gradient Boosting Machine, Adaptive Boosting, K-Nearest Neighbors, Support Vector Machine, Decision Tree, Naïve Bayes, and Logistic Regression.

[0139] These nine methods have been widely used in data prediction and each has its own advantages in different classification tasks. In this step, the same training data (75% of the data) and test data (25% of the data) are used to build the model.

[0140] For each model , using K-fold cross validation (K=5), calculate the AUC score:

[0141]

[0142] Select the model with the highest AUC As the final model of this type of feature. The optimal model of the three feature groups finally determined is expressed as , , .

[0143] The first layer prediction probability obtained by training: , , .

[0144] The output of the first layer model is:

[0145] S32: Second-layer Meta Model training, the prediction probability of the first-layer model As input features, on this basis, the final Meta Model is trained:

[0146]

[0147] in, Select in all features The model with the best prediction AUC to classify.

[0148] S4: Model performance evaluation. This is done using the following four metrics: accuracy, precision, recall, and F1 score. The definitions of each metric are as follows:

[0149] Accuracy: This metric evaluates the accuracy of the model's predictions, i.e., the ratio of the number of correct predictions to the total number of observations. It is expressed as:

[0150]

[0151] Among them, TP is the number of true positives, TN is the number of true negatives, FP is the number of false positives, and FN is the number of false negatives.

[0152] Precision: This metric measures the proportion of samples predicted to be positive that are actually positive. It is calculated as the number of true positives (TP) divided by the number of all predicted positives (TP + FP). It is expressed as:

[0153]

[0154] Recall: This metric measures the proportion of actual positive samples identified by the model, that is, the number of true positives (TP) divided by the total number of actual positives (TP + FN). It is expressed as:

[0155]

[0156] F1 score: This metric is the harmonic mean of precision and recall, and is used to comprehensively consider precision and recall. It is expressed as:

[0157]

[0158] This embodiment mainly evaluates the prediction accuracy of each model through the accuracy index. At the same time, the other three indicators will also be used to comprehensively evaluate the comprehensive performance of each machine learning model. Finally, by comparing , , , , , to explain the advantages of the multi-feature false information classification scheme based on the Stacking structure in false information detection. And apply it as a false information detection model based on comment features.

[0159] This embodiment has the following beneficial effects:

[0160] This embodiment significantly improves the accuracy and efficiency of false information detection through in-depth mining and comprehensive application of comment features. Traditional methods often rely only on basic sentiment polarity analysis and text statistics when processing structured comment data, and have difficulty capturing subtle differences and complex features in comments, such as time series information and diverse emotional expressions, which are key factors affecting the accuracy of false information identification. By comprehensively utilizing the time dimension information and multi-dimensional emotional expressions in comments, this embodiment can not only identify subtle differences that are easily overlooked in traditional methods, but also explore and integrate the deep-level features of comments. The multi-feature false information classification model for comments based on the Stacking structure integrates more fused features. This comprehensive analysis method of the fused feature model based on comment features enables this embodiment to more accurately capture complex patterns related to false information, thereby achieving higher accuracy in the identification of multi-topic false information.

[0161] Furthermore, the feature selection mechanism of this embodiment is specifically optimized for the unique components of social media comments, including but not limited to user behavior patterns, posting time series, and diverse sentiment. This makes disinformation detection more than just simple text analysis, but a comprehensive feature analysis that enables a more comprehensive understanding and prediction of disinformation behavior on social media. Therefore, through its highly innovative and forward-looking design, this embodiment ensures more accurate and reliable disinformation identification in complex information environments.

[0162] Finally, this embodiment adopts a multi-model horizontal training strategy. By analyzing the performance of the constructed base model and the multi-feature misinformation classification model based on the stacking structure on key performance indicators such as accuracy, precision, recall, and F1-score, it is possible to derive the model's pros and cons, and select the optimal model for misinformation detection. This approach not only optimizes the overall performance of the model but also enhances the system's adaptability and robustness to complex social media environments. Therefore, the solution provided by this embodiment effectively addresses the shortcomings of existing technologies and provides a comprehensive and precise technical solution to address increasingly complex information security challenges.

[0163] In order to implement the method of an embodiment of the present invention, an embodiment of the present invention also provides a false information detection system based on comment features, including: a processor and a memory for storing a computer program that can be run on the processor; wherein, when the processor is used to run the computer program, it executes the steps of the above-mentioned method.

[0164] The above-mentioned system provided in this embodiment and the above-mentioned method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0165] To implement the method of an embodiment of the present invention, an embodiment of the present invention further provides a computer program product. The computer program product includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of the above method.

[0166] Based on the hardware implementation of the above program modules and in order to implement the method of the embodiment of the present invention, the embodiment of the present invention further provides an electronic device (computer device). Specifically, in one embodiment, the computer device can be a terminal, and its internal structure diagram can be as follows: Figure 2As shown. The computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05, and a memory (not shown in the figure) connected via a system bus. The processor A01 of the computer device is used to provide computing and control capabilities. The memory of the computer device includes an internal memory A03 and a non-volatile storage medium A06. The non-volatile storage medium A06 stores an operating system B01 and a computer program B02. The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 in the non-volatile storage medium A06. The network interface A02 of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor A01, the method of any of the above embodiments is implemented. The display screen A04 of the computer device can be a liquid crystal display or an electronic ink display. The input device A05 of the computer device can be a touch layer covering the display screen, or it can be a key, trackball, or touchpad provided on the computer device housing, or it can be an external keyboard, touchpad, or mouse.

[0167] Those skilled in the art will understand that Figure 2 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0168] The device provided by an embodiment of the present invention includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the method of any one of the above embodiments is implemented.

[0169] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0170] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. 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.

[0171] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work 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 The function specified in one or more boxes.

[0172] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device 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.

[0173] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0174] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0175] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0176] It is understood that the memory of the embodiments of the present invention may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. Among them, non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disk, or compact disc read-only memory (CD-ROM); magnetic surface memory may be magnetic disk or tape memory. Volatile memory may be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus random access memory (DRRAM).The memories described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memories.

[0177] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0178] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for detecting false information based on comment features, characterized in that: The method comprises: Obtain a model training dataset, which includes the comment content and comment time of each blog post; perform feature extraction on the dataset to obtain comment external features, comment language features, and comment sentiment features; the comment external features are basic features of blog post comments, representing the quality or popularity of the blog post; the comment language features are text features of the blog post comments, representing the viewpoint preference of the blog post response; the comment sentiment features represent the sentiment representation carried in the blog post comments; The three models with the best prediction performance are selected from the preset multiple models using the external features of the comments, the language features of the comments, and the emotional features of the comments respectively as the three basic models; and the one model with the best prediction performance is selected from the preset multiple models using the outputs of the three basic models as the fusion model; a multi-feature false information classification model is constructed; the multi-feature false information classification model includes three basic models and one fusion model; the three basic models are respectively used to output a first prediction probability for the input external features of the comments, a second prediction probability for the input language features of the comments, and a third prediction probability for the input emotional features of the comments; the fusion model is used to output a final prediction probability based on the first prediction probability, the second prediction probability, and the third prediction probability output by the three basic models respectively; Detecting false information using the multi-feature false information classification model; The comment language features include text length, word frequency, number of sentences, number of emoticons, number of @ symbols used, and topic tags; The calculation formula of the text length is: in, Indicates the total number of comments. Comment The number of bytes; The calculation formula of the word frequency is: in, Comment The number of valid words; The calculation formula for the number of sentences is: in, Comment The number of sentences; The calculation formula for the number of emoticons is: in, Comment The number of emojis; The calculation formula for the number of AIT symbols used is: in, Comment Contains the number of "@"; The calculation formula for the topic labeling is: in, Comment The number of topics marked in .

2. The method for detecting false information based on comment features according to claim 1, characterized in that: Obtain the model training dataset, including: Get the comment content and comment time of each blog post based on the link of each blog post; The comment content and comment time corresponding to the blog post are taken as a piece of social media blog post data to construct a social media blog post dataset; The authenticity of false information in the social media blog data set is marked using labels to obtain a model training data set.

3. The false information detection method based on comment features according to claim 1 is characterized in that: The external features of comments include comment strength and comment speed; the comment strength is used to measure the degree of interaction triggered after the blog post is published; it is expressed as ; Indicates the total number of comments; The comment speed is used to represent the average time required for a blog post to be published and commented on; The calculation formula is: in, Indicates blog post Release time, Comment Release time, and Respectively represent blog posts The latest and earliest comment times among the comments.

4. The method for detecting false information based on comment features according to claim 1, characterized in that: The comment sentiment features include positive, negative and seven main classification sentiment features, and the seven main classification sentiment features include happiness, good, anger, sadness, fear, hate and surprise.

5. The method for detecting false information based on comment features according to claim 1, characterized in that: The preset multiple models include: random forest model, extreme gradient boosting model, gradient boosting machine model, adaptive boosting model, K nearest neighbor model, support vector machine model, decision tree model, naive Bayes model and logistic regression model.

6. The method for detecting false information based on comment features according to claim 1, characterized in that: The three models with the best prediction performance are selected from a plurality of preset models using the external features of the reviews, the language features of the reviews, and the sentiment features of the reviews as the three basic models, including: The AUC scoring method was used to evaluate the model performance, and the three models with the best prediction performance were selected from multiple preset models based on the AUC scoring results as the three basic models.

7. The method for detecting false information based on comment features according to claim 1, characterized in that: The performance of the multi-feature misinformation classification model was evaluated using accuracy, precision, recall, and F1 score metrics.

8. A false information detection system based on comment features, characterized in that: include: A processor and a memory for storing a computer program that can be run on the processor; wherein, when the processor is used to run the computer program, the steps of the method according to any one of claims 1 to 7 are performed.

Images