DRIFT for Dependency-Based Aspect Sentiment Prediction

DRIFT improves Aspect Sentiment Quadruple Prediction by linearizing dependency parse trees and injecting them directly into LLM prompts. This method enables models to reason over structured aspect-opinion dependencies rather than relying on simple token proximity.