| Resumen |
Pre-trained language models are able to capture a broad range of knowledge and language patterns in text and can be fine-tuned for specific tasks. In this paper, we focus on evaluating the effectiveness of various traditional machine learning and pre-trained language models in identifying depression through the analysis of text from social media. We examined different feature representations with the traditional machine learning models and explored the impact of pre-training on the transformer models and compared their performance. Using BoW, Word2Vec, and GloVe representations, the machine learning models with which we experimented achieved impressive accuracies in the task of detecting depression. However, pre-trained language models exhibited outstanding performance, consistently achieving high accuracy, precision, recall, and F1 scores of approximately 0.98 or higher. © 2024 Instituto Politecnico Nacional. All rights reserved. |
