Using Textual Transcripts of Parliamentary Interventions for Profiling Portuguese Politicians

This paper presents an experimental study on the subject of profiling political actors through textual transcriptions of their parliamentary interventions. Supervised learning techniques were used to learn models, which attempt to classify Portuguese politicians according to their gender, their age group, or their political affiliation and orientation. Experiments were made using different types of classification models, using state-of-the-art feature weighting schemes, using stylometric features from state-of-the-art approaches for author profiling, and using features derived from distributional word clustering or from concise semantic analysis. Experiments with the group Lasso regularization technique for logistic regression models were also performed. The experiments showed that language usage is indeed indicative of a person’s characteristics and ideology.