An interpretable framework for fine-grained sentiment analysis

This paper presents GENIS, a novel and fully realized methodology for grading sentiment in e-commerce reviews, with a particular emphasis on transparency and interpretability. unlike opaque black-box models, GENIS follows a simple, replicable, and operational pipeline that systematically identifies sentiment-bearing substantives—nouns and noun phrases that carry evaluative meaning—within textual reviews. these linguistic cues form the basis for assigning fine-grained sentiment scores on a 1–10 scale, enabling more nuanced evaluation than traditional binary or categorical sentiment classification. experimental results demonstrate that GENIS delivers performance comparable to state-of-the-art zero-shot large language models (LLMs) when measured against human-assigned scores, while significantly enhancing interpretability. by explicitly surfacing the linguistic rationale behind each decision, GENIS not only facilitates trustworthy AI but also supports practical applications in review analysis, customer feedback management, And model auditing.