WSD-AF for Robust Sentiment Analysis
Adaptive fusion and word-sense disambiguation for noisy sentiment classification.
Overview
WSD-AF is a COMP550 Natural Language Processing project at McGill University. It studies sentiment classification under noisy, adversarial, and semantically conflicting inputs, where standard fine-tuned language models can miss implicit negation or mixed sentiment.
The project combines word-sense disambiguation with an adaptive fusion module that balances global sentence semantics and context-focused semantics before final classification.
My Contribution
- Led the project direction, manuscript writing, and code development.
- Designed the adaptive fusion framework and word-sense disambiguation pipeline.
- Implemented disambiguation-related code, dependency parsing, and fusion-model experiments.
- Created the project figures and organized the public GitHub repository.
Notes
The repository contains the course report, datasets, main experiment scripts, and project figures. I treat this as a completed course project and implementation prototype; the README explains the evaluation modes, but it does not currently include a polished final result table.