Efficient Cluster Representation in Similar Document Search.
Shankaran Sitarama, Uma Mahadevan, Mani Abrol · 2003
Similar document search is the problem of retrieving documents that resemble a given document. In this paper, we describe a cluster-based retrieval scheme that approximates the classic nearest neighbor search scheme, by identifying the clusters that are closest to the input document and restricting attention to these clusters only. Cluster signatures play an important role in the effectiveness of this approximation, since the inclusion of a cluster in the restricted search depends entirely on whether its signature matches the given document. We study three different representations of cluster signatures and their role in performing a similar document search, while examining only a fraction of the documents from the target corpus.