Implicit feature detection for sentiment analysis

Kim Schouten, Flavius Frasincar

Abstract: Implicit feature detection is a promising research direction that has not seen much research yet. Based on previous work, where co-occurrences between notional words and explicit features are used to find implicit features, this research critically reviews its underlying assumptions and proposes a revised algorithm, that directly uses the co-occurrences between implicit features and notional words. The revision is shown to perform better than the original method, but both methods are shown to fail in a more realistic scenario.