Abstract

Collaborative Filtering systems commonly use neighbor-based approaches in order to find relevant items. In this paper we study the effects of several furthest neighbor-based models using two common similarity metrics, comparing the recommendations obtained when recommending items unliked by those least similar to oneself. Our results show that the proposed furthest neighbor method provides more diverse recommendations with a tolerable loss in precision compared to traditional nearest neighbor methods. The recommendations obtained by k furthest neighbor-based approaches are almost completely orthogonal to those obtained by their k nearest neighbors-based counterparts.

@inproceedings{said2012d,
author = {Said, Alan and Jain, Brijnesh J. and Kille, Benjamin and Albayrak, Sahin},
title = {Increasing Diversity Through Furthest Neighbor-Based Recommendation},
booktitle = {Proceedings of the WSDM'12 Workshop on Diversity in Document Retrieval (DDR'12) },
year = {2012},
location = {Seattle, WA, USA},
}