Research On Rumor Detection Model And Method Via Internet Public Opinion

With the development of Web3.0 technology,Internet becomes a more and more popular spreading carrier in public opinion topic,and produces various data,including true and false data.Through a hot public opinion topic,some media and people make rumors with obvious trends on purpose,and these rumors put a serious threat to normal Internet order.Therefore,automaticly and early detecting rumors from big Internet public opinion data can reasonably guide the direction of public opinion,and promote social harmony and stability.A beneficial feature set is the key to rumor detection in Internet public opinion.Nowadays,time feature,space feature,user feature and sentiment feature have been applied to rumor detection.Internet public opinion with rumors has a obvious sentiment,so obtaining sentiment feature from Internet public opinion makes great sense in rumor detection.In this paper,between-class relative document frequency distribution with a certain word is introduced into recent improved information gain algorithm to do more accurate feature selection and sentiment analysis.Sentiment feature by sentiment analysis together with other time feature,space feature and user feature are used to establish public opinion rumor detection index system.Based on a beneficial index system,good classification model plays a significant role in Internet public opinion rumor detection.Rumor detection is a classification problem,and selecting a beneficial classification model is help to improve rumor detection prediction accuracy.In this paper,considering limitations of C4.5 decision tree,random forest,naive bayes,neural network and support vector machine these single algorithms,a fusion model combining decision tree and logistic regression,logistic model tree model,are used to detect rumors,and the best classifier are selected to confirm the most important features to rumor detection.The experimental results show that in Internet public opinion rumor detection,firstly,improved information gain algorithm via relative document frequency distribution can improve classification performance of sentiment analysis,and then logistic model tree's classification performance is better than other common classification algorithms,more importantly,sentiment feature is one of the most significant features to detect rumors.In a conclusion,establishing a beneficial features set and classification model can effectively detect rumor in Internet public opinion.