The Best of Both Worlds: Combining Ranked List and Clustering,
Anton Leuski, James F. Allen · 1999
Abstract : An information retrieval system arranges retrieved documents into a list and ranks them in the order they are expected to be relevant. The ranked list is a well-known and widely accepted technique for information presentation for the task of finding relevant documents as quick as possible. Clustering is also a well-known and successful method for grouping similar documents together - in particular, for grouping relevant documents. In this paper we present a system that combines the ranked list with a clustering approach, accepts a modest amount of user feedback, and yields an approach that exceeds the retrieval effectiveness of a traditional ranked list, with and without relevance feedback.