Micro-blog Rumor Combination Recognition Model Based On Time Series

In the information age,the emergence of social application platform has greatly reduced the cost of information dissemination.Micro-blog,as one of the most widely used social applications,promotes the spread of rumors with its features of convenience,grassroots,originality and interactivity.It is particularly important to accurately identify rumour microblog and to stop it in time at the early stage.This paper takes micro-blog as the object and extracts 22 attribute features from four aspects,including user characteristics,text content,propagation characteristics and time effect.Logistic Regression,Support Vector Machine,Random Forest and improved BP neural network models as classical classification models are established based on micro-blog user information,text content and propagation characteristics.According to the comment information with time effect,a two-layer Bi-GRU deep learning model under the attention mechanism is established.In order to improve the recognition accuracy,the optimal classical classification model is combined with the deep learning model based on time effect,and finally the micro-blog rumor combination recognition model based on time series is established.Logistic Regression,SVM,Random Forest and improved BP neural network model are compared and analyzed.It is found that the comprehensive classification ability of Random Forest is the best,and its accuracy,precision,recall and F1 value reach 0.9632,0.9773,0.9457 and 0.9613 respectively.Meanwhile,the accuracy,precision,recall and F1 values of the two-layer Bi-GRU model under the attention mechanism based on time series are respectively 0.9666,0.9855,0.9446 and 0.9646.By combining and weighting the two classification models,a micro-blog rumor combination under simple weighted method recognition model based on time series is established.The accuracy,precision,recall and F1 values of the model are respectively 0.9894,0.9892,0.9903 and 0.9897 which is far better than a single sub-model.Compared with the deep learning model established by Liao X W and et al.(2018)and Sun W C and et al.(2020),the recognition accuracy of the combined model in this paper is increased by 2.14%and 0.84%.Through the analysis of Logistic and Random Forest,it is found that the number of blogs,the number of followers,the registration duration,whether he is an ordinary user,whether the micro-blog contains#,the absolute deviation between the number of comments and the sum of the number of being liked and being forwarded are remarkable and contribute a lot to the model.When the duration of account registration is shorter,the number of blogs,followers and the absolute deviation are smaller,and the micro-blog does not contain the#symbol,it is more likely that the micro blog released is a rumor micro blog.