Micro-blog Bursty Event Detection Method Based On Multi-feature Fusion

The Micro-blog's characteristics of fast spreading,timeliness and rich content,make Micro-blog a major social platform,also it provides a data foundation for bursty event detection research.But how to accurately detect bursty event in the massive Micro-blog data is of great significance to social stability and public interest,also it's a research hotspot and difficulty in recent years.Because of the large amount of data,a lot of noise and short text on Micro-blog,the existing bursty detection methods are not rich in feature extraction and the detection result is low in accuracy.In response to the above problems,this paper proposes a Micro-blog Bursty Event Detection Model Based on Multi-feature Fusion of emotional feature and text feature(MF?BDM).The emotional feature model is constructed by Micro-blog emotional symbols,and then use this model to classify Micro-blog text,detect the bursty period through the emotional feature.Secondly,the bursty words extraction in bursty period according to the microblog text feature about word frequency,word frequency growth and hashtag.Finally,the co-word analysis method is used to calculate the similarity between the burst words,and cluster the bursty words to get the Micro-blog bursty events.Experiments show that this model can accurately detect bursty events,and compared with other models,it has improved accuracy and recall rates.In order to solve the problem of extracting bursty words from Micro-blog text features,it is easy to misjudge non-event bursty word sets with high word frequency and high word frequency growth rate as bursty events.This paper incorporates user features to expand and improve the MF?BDM model,and obtain the MF?BDM+.The MF?BDM+model calculates Micro-blog text related to candidate bursty events,and then extracts user features which include the number of reposts,comments and likes from Micro-blog,get the Micro-blog influence and event influence through user features.When the influence of the event and the number of Micro-blog influence related to the event are both greater than the bursty event threshold,it is considered as the final bursty event detection result and analyze the bursty event.The experimental results show that the accuracy of the MF?BDM+ model has been improved,and it also proves the effectiveness of the user characteristics in the application of the model.Finally,based on the above model,this paper designs and implements the Micro-blog bursty events detection system,which displays events through keywords and related Micro-blogs,and the visual analysis of events through text,emotion,and word co-occurrence methods.