An approach to lexicon filtering for author profiling

This paper studies the influence of a general Spanish lexicon and a domain-specific lexicon on a text classification problem. Specifically, we address the impact of the choice of lexicons for user modelling. To do so, we identify gender and profession as demographic traits, and political ideology as a psychographic trait from a set of tweets. We experimented with machine learning and supervised learning methods to create a prediction model with which we evaluated our specific lexicon. Our results show that the choice and/or construction of lexicons to support the resolution of this task can follow a given strategy, characterised by the domain of the lexicon and the type of words it contains.