"Tell me more" : finding related items from user provided feedback
Faculty of Sciences. Mathematics and Computer Science
Faculty of Pharmaceutical, Biomedical and Veterinary Sciences . Biomedical Sciences
Lecture notes in computer science. - Berlin, 1973, currens
, p. 76-90
University of Antwerp
The results returned by a search, datamining or database engine often contains an overload of potentially interesting information. A daunting and challenging problem for a user is to pick out the useful information. In this paper we propose an interactive framework to efficiently explore and (re)rank the objects retrieved by such an engine, according to feedback provided on part of the initially retrieved objects. In particular, given a set of objects, a similarity measure applicable to the objects and an initial set of objects that are of interest to the user, our algorithm computes the k most similar objects. This problem, previously coined as cluster on demand , is solved by transforming the data into a weighted graph. On this weighted graph we compute a relevance score between the initial set of nodes and the remaining nodes based upon random walks with restart in graphs. We apply our algorithm Tell Me More (TMM) on text, numerical and zero/one data. The results show that TMM for almost every experiment significantly outperforms a k-nearest neighbor approach.