Harnessing Crowdsourced Recommendation Preference Data from Casual Gameplay

Recommender systems have become a familiar part of our online experiences, suggesting movies to watch, music to listen to, and books to read, among other things. To make relevant suggestions, recommender systems need an accurate picture of our preferences and interests and sometimes even our friends and influencers. This information can be difficult to come by and expensive to source. In this paper we describe a game-with-a-purpose designed to infer useful recommendation data as a side-effect of gameplay. The game is a simple, single-player matching game in which players attempt to match movies with their friends. It has been developed as a Facebook app and harnesses the social graph and likes of players as a source of game data. We describe the basic game mechanics and evaluate the utility of the recommendation knowledge that can be inferred from its gameplay as part of a live-user trial.