Discovering hidden concepts in predictive models for texts' polarization

The growth of Internet and the information technology has generated big changes in subjects' communication, which, nowadays, occurs through social media or via thematic forums. This challenges the traditional notion of Customer Relationship Management (CRM) and pushes businesses to prompt and accurate understanding of sentiments expressed, in order to address their marketing actions. In this paper, the authors propose a combined application of a supervised Sentiment Analysis (SA) with a probabilistic kernel discriminant to provide a robust classifier of texts polarization. The partition obtained is also described by means of a statistical characterization of the texts. Such an approach is very promising, not only in terms of classification accuracy, but also in terms of knowledge extraction. A real case study is illustrated in order to test and show the effectiveness of the proposed strategy.