Improving Automatic Text Document Clustering via Selecting a Small Amount of Labeled Data

摘要

We have investigated an approach which improves automatic text document clustering performance with the help of a small number of labeled documents. An active learning approach is proposed to select informative documents for obtaining user feedbacks on document labels. We make use of the intermediate cluster structure, which is discovered by the clustering process, to guide the active learning. Each cluster is represented by a language model. We make use of the uncertainty of document assignments as a clue for finding informative documents. We have conducted extensive experiments on several real-world corpora. The results demonstrate that our proposed framework is effective.