Experiments in Term Weighting and Keyword Extraction in Document Clustering

We study methods to initialize or bias different clustering methods using prior information about the "importance" of a keyword w.r.t. the whole document collection or a specific cluster. These studies give us hints on how to initialize clustering methods in order to improve performance if prior knowledge is available. This can be especially useful if a user-specific clustering of a document collection or web search result set is desired. Furthermore, we discuss whether one should draw on information measures to extract keywords that can be used as describing features for document clusters.

• Type: Conference
• Year: 2004
• Where: LWA
• Authors: Christian Borgelt, Andreas Nürnberger
• Keywords: Clustering Methods, Document Collection, LWA 2004, LWA 2007, User-specific Clustering