Social information discovery enhanced by sentiment analysis techniques

In recent years, the massive diffusion of social networks has made available a large amount of user-generated content, for the most part in the form of textual data that contain people's thoughts and emotions about a great variety of topics. In order to exploit these publicly available information, in this work we introduce a social information discovery system which elaborates simultaneously over more-than-one social network in an integrated scenario. The system is designed to ensure flexibility and scalability, thus enabling for (near-)real-time analysis even in case of high rates of content's creation and large amounts of heterogeneous data. Furthermore, a noise detection technique ensures a high relevance of analyzed posts/tweets to the domain of interest. We also propose a lexicon-based sentiment analysis algorithm to extract and measure users’ opinion, in order to support collaboration and open innovation. Polysemous words and negations are typically challenging for lexicon-based approaches: for this reason, we introduce both a word sense disambiguation algorithm and a negation handling technique. Experiments on several datasets have proven that the combined use of both techniques improves the classification accuracy on 3-class sentiment analysis.