(User selects MCQ options for each of the 21 BDI-II parameters)
(Options mapped to 0%, 33%, 66%, 100% based on severity)
(LLM to predict per-parameter risk scores)
(Combine manual and LLM scores using Outlier-Adjusted Weighted Mean)
(Use LLM to create a profile summary)
(Collect 100+ peer-reviewed papers and clinical resources, chunk them, create vector embeddings)
(Store all vectors for fast semantic search)
(Match user’s mental health summary with relevant literature and strategies)
(Use LLM with search results + Final Suicide Risk Percentage to create customized strategies)
1) Final Suicide Risk Percentage2) Suicide prevention strategies
(Select posts that reflect emotional, cognitive, or behavioral expressions)
(Tokenization, noise removal, normalization)
(Assign sentiment scores and Class: positive, negative, neutral)
(Apply the trained model to user's filtered posts)
(Use labeled Reddit data (20,000+ posts) to fine-tune Transformer model)
(Use LLM to predict per-parameter BDI-II scores for filtered posts)
1. Manual BDI-II scores (from user MCQ)2. Social media BDI-II scores (via LLM)3. Model-predicted suicide risk scores (from trained classifier)
by Raghu
