Research On The Feature Extraction Of Microblog Rumors And The Method Of Rumor Recognition Based On Multi-model Fusion Strategy

With the rapid development of the Internet,social media,represented by Sina Weibo,has gradually become the main platform for people to spread information.While social media has brought convenience to people in sharing information,it has also provided a way for online rumors to spread.The widespread spread of rumors on social media can damage the online environment and affect social life,therefore,the implementation of automated detection of online rumors is of great significance to maintain social stability and protect people's interests.Traditional rumor recognition methods ignore the importance of word frequency information of rumor keywords and the extraction of deep semantic features from Weibo body text when extracting features from Weibo.Therefore,this paper designs a rumor recognition model by extracting a feature named Relevance from a rumor database based on word frequency information and using deep learning methods to extract deep semantic features from the text of Weibo.The main work includes:Firstly,a rumor recognition model based on the feature of Relevance and statistical machine learning was designed.Considering the importance of word frequency information for rumor identification,web crawlers were used to obtain the latest rumor and non-rumor data from the Sina Weibo platform.The Weibo rumor database and experimental dataset were constructed by combining the existing datasets.The APRIORI algorithm was used to extract the frequent itemsets of the Weibo rumor database,and the frequent itemsets were then extended by extracting near-sense words through the Word2 Vec model.Eventually,the feature of Relevance was extracted from the frequent itemset.The feature of Relevance was combined with other extracted statistical features into a statistical machine learning model for training,and experiments showed that the feature of Relevance was effective in improving the model's rumor recognition.The GBDT-R model was also designed for rumor detection,and the effectiveness of the model was verified.Secondly,A rumor recognition model based on a multi-model fusion strategy was designed.In view of the limitations of a single model and the improvement of the classification effect by multi-model fusion strategy,this paper designed an ALBERT-Text CNN-SVM model combining deep learning and machine learning.A novel pre-training model ALBERT and combined with Text CNN was designed to introduce ALBERT-Text CNN for deep feature extraction of Weibo text.Simultaneously,the deep semantic features of Weibo text extracted by ALBERT-Text CNN combined with some Weibo statistical features are input into the SVM model for training,and named ALBERT-Text CNN-SVM,considering the improvement of other statistical features of Weibo for rumor recognition.In comparison with ALBERT-Text CNN which uses Weibo text alone,ALBERT-Text CNN-SVM has improved 0.68%,2.18% and 0.78% in accuracy,recall and F1 value respectively.Meanwhile,the model reached the optimal values in the comparison experiments with other literature in terms of accuracy,precision and F1 value,which were 95.98%,97.57% and 95.78% respectively.At the same time,the recall rate reached 94.06%.The experiments demonstrated that the ALBERT-Text CNN-SVM model can effectively identify Weibo rumors.