Unsupervised topic discovery in micro-blogging networks

Unsupervised automatic topic discovery in micro-blogging social networks is a very challenging task, as it involves the analysis of very short, noisy, ungrammatical and uncontextual messages. Most of the current approaches to this problem are basically syntactic, as they focus either on the use of statistical techniques or on the analysis of the co-occurrences between the terms. This paper presents a novel topic discovery methodology, based on the mapping of hashtags to WordNet terms and their posterior clustering, in which semantics plays a centre role. The paper also presents a detailed case study in the field of Oncology, in which the discovered topics are thoroughly compared to a golden standard, showing promising results. (C) 2015 Elsevier Ltd. All rights reserved.