Social media platforms contain interesting information that can be used to directly measure people' feelings and, thanks to the use of communication technologies, also to geographically locate these feelings. Unfortunately, the understanding is not as easy as one may think. Indeed, the large volume of data makes the manual approach impractical and the diversity of language combined with the brevity of the texts makes the automatic approach quite complicated. In this paper, we consider the gamification approach to sentimentally classify tweets and we propose TSentiment, a game with a purpose that uses human beings to classify the polarity of tweets (e.g., positive, negative, neutral) and their sentiment (e.g., joy, surprise, sadness, etc.). We created a dataset of more than 65,000 tweets, we developed a Web-based game and we asked students to play the game. Obtained results showed that the game approach was well accepted and thus it can be useful in scenarios where the identification of people' feelings may bring benefits to decision making processes.

Twitter is the most popular real-time micro-blogging service and it is a platform where users provide and obtain information at rapid pace. In this scenario, one of the biggest challenge is to find a way to automatically identify the most influential users of a given topic. Currently, there are several approaches that try to address this challenge using different Twitter signals (e.g., number of followers, lists, metadata), but results are not clear and sometimes conflicting. In this paper, we propose TRank, a novel method designed to address the problem of identifying the most influential Twitter users on specific topics identified with hashtags. The novelty of our approach is that it combines different Twitter signals (that represent both the user and the user's tweets) to provide three different indicators that are intended to capture different aspects of being influent. The computation of these indicators is not based on the magnitude of the Twitter signals alone, but they are computed taking into consideration also human factors, as for example the fact that a user with many active followings might have a very noisy time lime and, thus, miss to read many tweets. The experimental assessment confirms that our approach provides results that are more reasonable than the one obtained by mechanisms based on the sole magnitude of data.