Detalhes do Documento

CeDRI at eRisk 2021: a naive approach to early detection of psychological disorders in social media

Autor(es): Lopes, Rui Pedro

Data: 2021

Identificador Persistente: http://hdl.handle.net/10198/24037

Origem: Biblioteca Digital da UPB

Assunto(s): Early risk detection; Tf-Idf; Word2Vec; Recursive neural networks; Dataset heuristics; DL4J


Descrição

This paper describes the participation of the CeDRI team in eRisk 2021 tasks, particularly, the Task 1: Early Detection of Signs of Pathological Gambling and Task 2: Early Detection of Signs of Self-Harm. The main difference between these two is that the first is a “test only” challenge, where no training data is supplied. The second task has labeled data available, which can be used for training. Both tasks were addressed using the same algorithms, using a custom training set for Task 1 and the provided data in the second. The algorithms were TfIdf vectorizer with a Logistic Regression layer, Word2Vec vectorizer with LSTM and Word2Vec vectorizer with CNN. All vectorizers and Neural Networks were trained solely with the training data. As expected, the algorithms did not state-of-the-art, but the experience allowed to reflect in several aspects related to the importance of proper dataset preparation and processing. © 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

Tipo de Documento Comunicação em conferência
Idioma Inglês
Contribuidor(es) Biblioteca Digital da UPB
Licença CC
facebook logo  linkedin logo  twitter logo 
mendeley logo

Documentos Relacionados