Combining passage-level and document-level evidence in information retrieval

Abstract

Large collections of full-text documents are now commonly used in automated information retrieval. When the stored documents are long, the retrieval of complete documents may not be in the user's best interest. In this case, efficient and effective retrieval results may be obtained by retrieving text excerpts of varying size in response to statements of the user's interest. This study explores a particular approach of segmenting text into passages for retrieval.Documents are segmented into passages by extracting a fixed number of contagious words from the whole document.The process of extracting passages can be viewed as using a window of certain size that slides through the whole document.