Author:
Patricia Martin-Rodilla
Affiliation:
Information Retrieval Lab (IRLab), Facultade de Informática, University of A Coruña, Spain
Keyword(s):
Ontology Learning, Time, Ontology Evolution, Text Mining, Social Media, Depression, Early Risk Prediction.
Abstract:
Approaches to early detection of depression based on individual’s language are receiving increasing attention, with detection software systems based on lexical, grammatical or discursive components applied to medical corpus or social media texts. However, these first detection systems are defragmented, each attending to a specific feature or linguistic level, and not addressing a more conceptual level. Existing ontology learning (OL) methods extract the ontology referred in the text. In addition, existing systems perform language analysis for the detection of depression as a snapshot of each individual, regardless of their temporal dimension. Is it possible that suitable linguistic features to detect early signs of depression vary over time? And the underlying ontology? This paper presents a model that adds the temporal component to current ontology learning models to perform evolutionary analysis of both linguistic and ontological features to texts from social networks. The model ha
s been applied to an external corpus of depression from social media texts, with a two-fold goal: 1) validating the model by contrasting it with OL models without temporal component 2) producing a corpus of evolutionary OL results applied to the depression detection from social media texts.
(More)