Among the various social media platforms that dominate the internet today, Twitter has established itself as a major player and has become a preferred choice for expressing one’s opinion on almost everything. Ordinary citizens aside, it is used by the governments world over to connect directly with their citizens, by the media houses, companies, research organizations, and so on. Given the omnipresent role it plays, it has thus become imperative to aggressively pursue topic detection from Twitter so that its benefits can be implemented in a range of applications such as natural disaster warnings, fake news detection, and user opinion assessment among other uses. This paper outlines different types of topic detection techniques that are employed frequently such as exemplar based topic detection, clustering based topic detection, frequent pattern mining, two level message clustering, combination of k-means clustering methods and singular value decomposition. While exemplar based topic detection technique uses the most significant tweets to detect topics, the clustering based technique uses different clustering methods such as sequential k-means, spherical k-means, DBSCAN, and bngram to detect the topics. In frequent pattern mining, the FP-growth algorithm along with its variation can be used to detect the topics. In two level message clustering, topics are clustered using different phases. A blend of both algorithms is employed in the combination of k-means clustering methods and singular value decomposition. A detailed discussion of these topic detection techniques has been done in this paper.

Keywords : Text Mining; Topic Detection; Clustering; Twitter