UniOR NLP at eRisk 2021: Assessing the Severity of Depression with Part of Speech and Syntactic Features

This paper describes the participation of the UniOR NLP Research Group team in task 3 (T3) within the CLEF eRisk 2021 lab. We report the approaches used to address eRisk 2021 T3, which aims to measure the severity of the signs of depression in social media users. This year’s eRisk T3 consists of exploring methods for automatically filling out a 21-question depression questionnaire, namely Beck’s Depression Inventory (BDI). We explored and tried different combinations of text pre-processing and feature extraction steps in order to grasp self-referential pieces of text and two main methods for representing the text features as input data for traditional machine learning classifiers.