Proceedings of the 16th International Conference on statistical analysis of textual data. Vol. 2

This paper aims to present a methodology for monitoring and assessing the public perception on social media. The assessment is based on a procedure using sentiment analysis to build a composite indicator, the Greco-polli (Gp) index. To this aim, we identify a set of hashtags according to the agenda setting to collect data from the Twitter repository and build a corpus for each hashtag. The procedure operates in two steps, and it is applied to the set of corpora, one for each hashtag. first, we automatically assess the sentiment using a non-supervised sentiment analysis procedure based on emotional Text Mining (eTM), and we used the dataset already labeled by eTM to classify the sentiment by Machine learning (Ml). Then, we calculate the Gp index for each sentiment analysis procedure, i.e., we transform the results in a stream of numerical data, and we build a composite index aggregating the information on the sentiment related to topics. We assess the applicability of the automatic labeling system by eTM for the training of Ml algorithms and we compare the results by the Gp index between eTM and Ml. This procedure is applied to a case study on the public perception of security. results highlight that this approach enables the assessment any potential change in the public perception, and that the use of eTM procedure for the automate labeling and the training of Ml algorithms produces good results.