Utilization of global ranking information in GraphBased biomedical literature clustering

Paper presented at the 9th International Conference on Data Warehousing and Knowledge Discovery, DaWaK 2007, Regensburg, Germany.In this paper, we explore how global ranking method in conjunction with local density method help identify meaningful term clusters from ontology enriched graph representation of biomedical literature corpus. One big problem with document clustering is how to discount the effects of class-unspecific general terms and strengthen the effects of class-specific core terms. We claim that running global ranking method on a well constructed term graph can identify class-specific core terms. In detail, PageRank and HITS are applied on a direct abstract-title graph to target class specific core terms. Then k dense terms clusters (graph) are identified from these terms. Finally, a document is assigned to the closest term graph. A series of experiments are conducted on a document corpus collected from PubMed. Experimental results show that our approach is very effective to identify class-specific core terms and thus help document clustering