Adaptive Centroid-Based Clustering Algorithm for Text Document Data

Document clustering is a significantly popular research, which aims to partition a corpus into many subgroups of homogeneous documents. Traditional clustering approaches catholically lack of considerations of word weights with clusters. To address this problem, we propose an Adaptive Centroid-based Clustering (ACC) algorithm. As a successful supervised centroid-based classifier, Class-Feature-Centroid (CFC) algorithm takes relationships among words into account. ACC attempts to employ this discriminative CFC vector to drive the clustering procedure. Since clustering is unsupervised, ACC begins with hundreds of small clusters for acceptable CFC vectors, and then iteratively regroups clusters of documents until convergence. As ACC is self-organized, it can determine the number of clusters adaptively. The experimental results validate that ACC achieves competitive performance with the state-of-art clustering approaches.