Abstract

Conventional keyword-based indexing and retrieval techniques for textual documents lack of precision when a long query string is employed in order to discover documents containing a specific “event”, such as “Einstein discovered relativity”. This paper proposes a framework to resolve such a problem. In our proposal, we apply semantic role labeling and coreference techniques in order to parse each sentence within textual documents into three elements: subject, object and predicates. These elements can subsequently be used for indexing and retrieval. Our primitive evaluation experiments have shown that this promising methodology raises the retrieval precision if we compared it to conventional literal termmatching techniques.

Citations