An Enhanced LDA Topic Model Approach for Event Extraction from Twitter

Topic models are the strongest technique to identify latent patterns from the huge amount of text content. They are applied in a wide range of areas including recent work for event extraction from Twitter. There are a heated discussion on social media when major events are occurred around the world and now a day's finding out important information from those contents generated by human becoming a challenging task. In this paper we compare different topic model techniques like PLSA, LDA, CTM and NMF apply directly on tweets and the experimental result shows that LDA perform well. Though applying traditional LDA topic model directly on tweets posses’ two challenges, 1st Data scarceness problem due to the nature of short text length of the tweets and 2nd Generated summaries contain words that are somewhat general and independent to the topic that is failed to understand the denotation of twitter data. To this end, this paper proposed a novel approach named W-LDA (weighted LDA) that not only solve the short text problem of tweet but also solve the problem of uncertainty about assignment of proper topic to documents (tweets).Experimental results shows that our proposed method manifest clearly an improved accuracy than the existing event detection methods.