A Study of the Effect of Document Representations in Clustering-Based Cross-Document Coreference Resolution

Finding information about people on huge text collections or on-line repositories on the Web is a common activity. We describe experiments aiming at identifying the contribution of semantic information (e.g., named entities) and summarization (e.g., sentence extracts) in a cross-document coreference resolution system. Our system uses a clustering-based algorithm to group documents referring to the same entity. Clustering uses vector representations created by summarization and semantic tagging components. We investigate different clustering configurations and show that selection of the type of summary and the type of term to be used for vector representation is important to achieve good performance.