A network approach to dimensionality reduction in Text Mining

There is an ever-increasing interest in developing statistical tools for extracting information from documental repositories. In a Text Mining frame, a knowledge discovery process usually implies a dimensionality reduction of the vocabulary, via a feature selection and/or a feature extraction. Here we propose a strategy designed for reducing dimensionality through a network-based approach. Network tools allow performing the reduction by considering the most important relations among the terms. The effectiveness of the strategy will be shown on a set of tweets about the 2018 Italian General Election.