Umigon-lexicon: rule-based model for interpretable sentiment analysis and factuality categorization

We present umigon-lexicon, a set of lexicons in English and associated conditions to evaluate sentiment in a text. These lexicons are curated to capture the subjective sentiment conveyed by the author specifically, as opposed to the identification of the overall sentiment. We provide a comparison with existing lexicons before evaluating the performance on sentiment and factuality classification tasks. These contributions highlight the long-lasting relevance of lexicon-based solutions for algorithmic, inherently interpretable models in sentiment analysis and factuality categorization.