Human-Level Interpretable Learning for Aspect-Based Sentiment Analysis

This paper proposes human-interpretable learning of aspect-based sentiment analysis (ABSA), employing the recently introduced Tsetlin Machines (TMs). We attain interpretability by converting intricate position-dependent textual semantics into binary form, mapping all features bag-of-words (BOWs). The form BOWs are encoded so that information on aspect and context words nearly lossless for classification. further adapt as input to TM, enabling patterns in propositional logic. To evaluate accuracy, we conducted experiments two widely used ABSA datasets SemEval 2014: Restaurant 14 Laptop 14. show how each relevant feature takes part conjunctive clauses contain corresponding word, demonstrating human-level interpretability. At same time, obtained accuracy is competitive with existing neural network models, reaching 78.02% 73.51%