Topic word set-based text clustering

Clustering is the task of grouping related and similar data without any prior knowledge about the labels. In some real world applications, we face huge amounts of unstructured textual data with no organization. In these situations, clustering is a primitive operation that needs to be done to help future e-commerce tasks. Clustering can be used to enhance different e-commerce applications like recommender systems, customer relationship management systems or personal assistant agents. In this paper we propose a new method for text clustering, by constructing a term correlation graph, and then extracting topic word sets from it and finally, categorizing each document to its related topic with the help of a classification algorithm like SVM. This method provides a natural and understandable description for clusters by their topic word sets, and it also enables us to decide the cluster of documents only when needed and in a parallel fashion, thus significantly reducing the offline processing time. Our clustering method also outperforms the well-known k-means clustering algorithm according to clustering quality measures.