Analysis of Twitter Data Using a Multiple-level Clustering Strategy

Twitter, currently the leading microblogging social network, has attracted a great body of research works. This paper proposes a data analysis framework to discover groups of similar twitter messages posted on a given event. By analyzing these groups, user emotions or thoughts that seem to be associated with specific events can be extracted, as well as aspects characterizing events according to user perception. To deal with the inherent sparseness of micro-messages, the proposed approach relies on a multiple-level strategy that allows clustering text data with a variable distribution. Clusters are then characterized through the most representative words appearing in their messages, and association rules are used to highlight correlations among these words. To measure the relevance of specific words for a given event, text data has been represented in the Vector Space Model using the TF-IDF weighting score. As a case study, two real Twitter datasets have been analysed.