Clustering Improvement via Integrating with Sparse Topical Coding

Topic modeling can improve document clustering by projecting documents into a topic space. By document, we mean a general concept. Document can be an image, a video, a textual document or each data which can be described in bag-of-words model based on the histogram of its features. In this paper, we introduce a clustering method based on Sparse Topical Coding (STC). In the proposed method, document clustering and topic modeling are integrated into a unified framework and jointly performed to achieve the best clustering performance. Our method clusters the documents based on STC topic modeling used for mining the topics and K-means clustering used for discovering latent groups in document collection. Experimental results show the effectiveness of our proposed clustering approach.