You are here

A Game with a Purpose for Recommender Systems


Barry Smyth, Rachael Rafter, Sam Banks

Publication Type: 
Refereed Conference Meeting Proceeding
Recommender systems help us choose what to read, watch and buy, and are now a common feature of online services. They do this by learning about our preferences and relationships in order to generate targeted suggestsions that we are likely to find relevant (Adomavicius and Tuzhilin 2005; Desrosiers and Karypis 2011). The data on which they rely can be difficult to collect at scale and in this paper we consider a novel approach to collecting this data by using a so-called game-with-a-purpose (GWAP); see (von Ahn and Dabbish 2004; Law and von Ahn 2009; Salvador et al. 2013; Gligorov et al. 2011; Cooper et al. 2010). In the pastGWAPs have been used for challenging tasks such as object recognition and protein folding; might they also be used to help build better recommender systems? To explore this we describe a single-player matching game in which players attempt to match movies with their friends. These matches can be used to infer the strength of relationships between users and the predicted level of interest a user might have for a particular movie. Both of these types of data can be useful in a recommendation context.We evaluate the utility of this data as part of a live-user trial.
Conference Name: 
Third AAAI Conference on Human Computation and Crowdsourcing
Digital Object Identifer (DOI):
Publication Date: 
Research Group: 
National University of Ireland, Dublin (UCD)
Open access repository: