Document Clustering with User feedback

In this paper, we focus on the problem of incorporating user input into an automated document clustering process to improve clustering performance. Before the start of the clustering process, the user can provide a small set of labeled documents that form the initial descriptions of the clusters. If the user provides no initial information, the clustering process has to form the initial descriptions of the clusters by itself. At the end of the clustering process, the user can navigate the clusters, assess the clustering quality and if necessary, provide feedback to the clustering process. With this user feedback, the clustering process will retrain itself to obtain new clusters that best describe the nature of the data and the desire of user. The results show that our methods for initializing the clustering process are valuable and that user feedback improves the quality of the clusters.