Bursty Event Detection in Social Networks by Integrating Word Features and Word2Vec Model

By constructing bursty event detection model in social networks, this paper helps public opinion supervision departments monitor and guide public opinion for various emergencies, reduces negative impacts and has great significance to maintain social security and stability. This paper proposes a novel bursty event detection method in social networks by integrating word features and Word2Vec model to make up for the lack of extracting deep semantic features of text information. Firstly, we crawl the Weibo text data which is sliced into multiple time windows, and extract bursty words by computing the features including word frequency feature and word frequency growth rate feature for each word in each time window. Secondly, we compute word vectors of bursty words and obtain deep semantic features of bursty words by using the Word2Vec model. Thirdly, we construct similarity matrix for bursty words by considering both of the co-occurrences and semantic features of bursty words, and detect bursty events in different time windows by adopting the cohesive hierarchical clustering. The experimental evaluation results show that the word features and Word2Vec based method has excellent performance in the bursty event detection task and the proposed method has a high improvement compared to the baseline methods.