Dictionary-based Sentiment Analysis at Subword Level

While deep learning can offer a promising approach to sentiment analysis, it often presents challenges in complexity and explainability. Alternatively, sentiment analysis based on dictionaries has been explored. In this paper, subword-level dictionaries in English are considered to address performance degradation resulting from domain mismatches. To work at subword level, a framework based on naïve bayes machine learning algorithm is exploited. Furthermore, stopwords at the subword level have been proposed to remove additional interference intrinsic to subword tokenization. Numerical experiments demonstrate that the proposed method achieves higher accuracy and F1 scores compared to the conventional dictionary-based method when there is a mismatch between the dictionary and the corpus of documents while performing marginally worse than state-of-the-art deep learning methods when applied to datasets from the same domain.