Toward Reliable Depression Detection in Negation-Heavy Twitter Corpora

Sawant Dessai, Vishal Vishvanath (2025) Toward Reliable Depression Detection in Negation-Heavy Twitter Corpora. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (775kB) | Preview
Preview	PDF (Configuration Manual) Download (657kB) | Preview

Abstract

Negation and informal language are frequent in social media posts indicating depressive moods, but these are hard to recognize with an automated classifier when labels are synthesized using rules of lexicons instead of human labeling. The paper assesses the claim that the new DeBERTa-v3 transformer exhibits any quantifiable benefits as compared to BERT and RoBERTa in these settings of realistic weak labeling. Based on the 57k-tweet Multi-Labeled Depression Corpus, we have maintained negation, emojis, and casual speech and used stratified 80/10/10 partitions and the same fine-tuning parameters to all architectures. Evaluation of performance is done with the use of Macro-F1 on the entire test set as well as a subset which includes a larger number of negation in the test set. With three randomly chosen seeds, DeBERTa-v3 also has the highest macro-F1, both with 3-class (0.9187) and with binary classification (0.9510), and all pair-wise differences are all statistically significant (p < 0.0001). It was also found that the model will be more robust on negative heavy tweets. These outcomes suppose that the architectural enhancements of DeBERTa-v3, especially disentangled attention, are real advantages towards depression detection in noisy, language-noisy, weakly labelled social-media settings.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Horn, Christian UNSPECIFIED
Uncontrolled Keywords:	Depression detection; Weakly labeled data; Transformer models (BERT, RoBERTa, DeBERTa-v3); negation handling; Twitter (X); Social media
Subjects:	P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing R Medicine > RA Public aspects of medicine > RA790 Mental Health Z Bibliography. Library Science. Information Resources > ZA Information resources > ZA4150 Computer Network Resources > The Internet > World Wide Web > Websites > Online social networks T Technology > TK Electrical engineering. Electronics. Nuclear engineering > Telecommunications > The Internet > World Wide Web > Websites > Online social networks
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	09 Sep 2026 08:39
Last Modified:	09 Sep 2026 08:39
URI:	https://norma.ncirl.ie/id/eprint/9906

Actions (login required)

View Item