ayaanzhaque/SDCNL

Implements a clustering-based label correction pipeline that uses dimensionality reduction and confidence thresholding to automatically denoise noisy Reddit labels without requiring prior noise distribution assumptions. Leverages pre-trained transformer embeddings fed into a supervised classifier trained on the corrected labels, addressing the clinical challenge of distinguishing depression from suicidal ideation in web-scraped text data.

No commits in the last 6 months.