Key points are not available for this paper at this time.
Depression, a prevalent mental health concern, manifests itself on social media, providing opportunities for early detection. This study aims to fill the gap in research for low-resource Indian languages by proposing a comprehensive approach to depression detection across eight languages. By utilizing the Long Short-Term Memory and the Gated Recurrent Unit models, we provide an ensemble model that improves overall performance. Our translation-based dataset solves data restrictions related to language. Drawing insights from various studies, the literature review delves into the complexities of analyzing mental health on social media. We translate these insights into practical applications through model implementation and preprocessing techniques. The LSTM and GRU models, enriched with pretrained FastText embeddings, demonstrate competitive accuracy. The ensemble model, amalgamating their outputs, exhibits robustness in depression detection. Nevertheless, certain limitations persist, including concerns related to dataset representativeness and the nuances of informal social media language. Our study concludes with a roadmap for future research, emphasizing the need to expand datasets, develop language-specific models, and explore advanced architectures. This research contributes to a nuanced understanding of depression detection in linguistically and culturally diverse contexts, building upon insights gained from the literature review and model implementations across various Indian languages.
Rajderkar et al. (Fri,) studied this question.