Transformer-Based Abstractive Summarization for Depression Detection Literature for Enhanced Medical Insights
Akshi Kumar, Aditi Sharma, Saurabh Raj Sangwan · 2025
The overwhelming surge in depression detection research presents significant challenges for mental health professionals and researchers in keeping pace with new advancements. This issue is particularly critical as timely access to insights from recent studies is essential for effective diagnosis and intervention strategies. Manually summarizing the growing body of literature is labour-intensive and prone to inconsistencies, creating an urgent need for automated summarization tools. This study introduces DepressiLex, a specialized corpus comprising 40 research papers from 2023-2024 focused on depression detection. Using transformer-based models like including Pre-training with Extracted Gap-Sentences for Abstractive Summarization (PEGASUS), Bidirectional and Auto-Regressive Transformers (BART), the Text-to-Text Transfer Transformer (T5-Base), the Longformer-Encoder-Decoder (LED), and ProphetNet, to evaluate their effectiveness in generating abstractive summaries. We assessed their performance using metrics such as the Bilingual Evaluation Understudy (BLEU) and the Recall-Oriented Understudy for Gisting Evaluation (ROGUE), with the Longformer-Encoder-Decoder consistently outperforming the others. This engineering advancement fulfils a critical need in healthcare, providing mental health professionals with streamlined, artificial intelligence (AI)-enabled access to key insights, thereby significantly reducing the time and cognitive load involved in reviewing complex research. Additionally, word cloud visualizations highlight the dominant themes and terms across the summaries, underscoring the potential of transformer models to transform access to mental health knowledge.