结合文本与用户信息的微博谣言检测方法研究

Research on Micro-blog Rumor Detection Method Based on Text and User Information

微博已成为人们获取信息的重要来源,但微博中也充斥着大量网络谣言,会对个人和社会造成严重危害。由于谣言的模糊性和隐蔽性,导致微博谣言的自动识别非常困难,这给微博监管带来极大挑战。针对微博谣言的文本信息,构建了句子级细粒度的情感分类算法,对微博评论进行情感倾向性分析,计算负面情感评论比例;同时结合微博的用户信息,分析微博转发和评论列表,判断是否有潜在辟谣用户参与,并利用发文用户的等级、信用等信息,定义并计算用户信誉值;最后训练分类器并完成对微博未知样本的预测。实验结果表明,所提出的新特征能有效提高微博谣言检测的准确率。

While micro-blog has become an important source for people to obtain information,there are also a large number of internet rumors,which can cause serious harm to individuals and society.Because of the fuzziness and concealment of rumors,the automatic identification of micro-blog rumors is very difficult,which brings great challenges to micro-blog′s supervision.For the text information of rumors,this paper constructed a new classification method of sentence level fine-grained emotion,and used this method to analyze the emotional orientation of comments,and then calculated the proportion of negative emotional comments.Combined with user information at the same time,it analyzed the retweet lists and comment lists,and judged whether there were potential authoritative users that participated in the propagation process of rumors.Moreover,it defined and calculated user reputation value by using the information of the user′s grade and credit.Finally,the classifier was trained and the prediction of unknown micro-blog samples was completed.The experimental results show that the new features proposed can effectively improve the accuracy of micro-blog rumors detection.