Support-vector-based iteratively adjusted centroid classifier for text categorization

To address the lackness of centroid-based classifier (CC) that is prone to generate inductive bias or model misfit, a support-vector-based iteratively-adjusted centroid classifier (IACC_SV) was proposed, which employs support vectors found by some routines, e.g., linear support vector machines (SVMs) to construct centroid vectors for CC, and then iteratively adjusts the initial centroid vectors according to the misclassified training samples. Compared with traditional classification algorithms, IACC_SV achieves better performance in terms of macro-F1 and micro-F1, and the extensive experiments on 8 real-world text corpora demonstrate the effectiveness of the proposed algorithm, especially on text corpora with highly imbalanced classes.