Research On Early Detection Of Microblog Rumors Based On Latent Semantic Analysis

With the rapid development of social media,emerging media represented by Weibo have become an important platform for information exchange.While bringing convenience to people's lives,the open micro-blog platform has also become an important source of rumors.Traditional rumor detection methods are mostly studied and analyzed by learning the semantic representation of forwarded comment information.At this time,rumors have been widely spread,causing a lot of serious consequences.Early detection of rumors can reduce the spread of rumors to a greater extent,which has more important practical significance.Due to the rumors exposed the characteristics of the incubation period is very limited,the early detection of rumor is very difficult.At the same time,the traditional rumor detection ignores the current huge rumor database and potential semantic features.In view of the above problems,this paper constructs topic features to enhance the effect of early detection of rumors.On the basis of not relying on the comments and forwarded information of the microblogs to be detected,a new rumor detection model is proposed to carry out early detection of rumors,so as to meet the real-time requirements of early detection of rumors in practical work.The main work of this paper is as follows:(1)In order to solve the problems of complex,fuzzy and diversified rumors and the flood of a large number of repetitive rumors,defined title weights and corrected inverse document frequency for similar incident rumors Improved term frequency-inverse document frequency keyword extraction algorithm to obtain seed-sensitive word set,improved point mutual information algorithm to get related word set.On this basis,a sensitive lexicon specifically applied to microblog rumors is constructed,which can improve the semantic understanding and recognition ability of the text for rumor words in the subsequent topic extraction,and assist rumor recognition.(2)In view of the insensitivity of current topic models to low-frequency words in text,we combined the rumor-sensitive thesaurus with the Latent Dirichlet Allocation model,proposed a RUM-LDA topic model,and designed a rumor topic feature extraction method.The comparative experiments show that RUM-LDA model effectively improves the efficiency of sensitive information subject recognition,and subject features can greatly improve the accuracy of follow-up rumor detection,reduce detection time,and enhance the early detection effect of rumors.(3)Aiming at the problems of different text length and randomization of micro-blog,a method of contextual semantic feature extraction based on the integration of attention mechanism and Long Short-Term Memory was designed,and an early detection model of micro-blog rumor based on latent semantic analysis was proposed.This model is named as the RUM-Attlstm model.This model integrates the mapping of traditional social features,rumour thematic features and contextual semantic features to carry out early rumour detection,and solves the gradient disappearance and gradient explosion problems that traditional networks need to deal with.Experiments show that this model can better identify the potential semantic information of text,effectively improve the performance indicators of rumor early detection,and allocate more attention to the key content such as sensitive words.