An Improvement of Author Feature Extraction Using ̂β - Reduction

Abstract

This study aims to improve the method for deriving concise rules that represent an author’s stylistic features by adopting ̂β-reduction, a technique based on variable precision rough sets that allows reversals of decision attributes. Compared to conventional Lβ-reduction, this approach enables the extraction of shorter and more interpretable rules. In conclusion, the application of ̂β-reduction led to a reduction of five attributes, with a 0.01 improvement in coverage compared to Lβ-reduction. Although the confidence value decreased by 0.05, this is attributed to more aggressive attribute reduction under a relaxed threshold. These results indicate that ̂β-reduction is effective in simplifying stylistic rules while broadening their applicability.