KAN and RinSCut: Lazy Linear Classifier and Rank-in-Score Threshold in Similarity-Based Text Categorization

Two important research areas in statistical approaches for automated text categorization are similarity-based learning algorithms and associated thresholding strategies. The combination of these techniques significantly influences the overall performance of text categorization systems. After researching common techniques in both areas, we describe a lazy linear classifier known as the keyword association network (KAN) and a rank-in-score (RinSCut) thresholding strategy to improve the categorization performance over existing techniques. Extensive experiments have been conducted on the Reuters-21578 data set. The experimental results show that KAN outperforms two linear classifiers, Rocchio and Widrow-Hoff, and gives results competitive with k-NN. All implemented classifiers except Widrow-Hoff show performance improvements with RinSCut.