Comparison-Based Interactive Collaborative Filtering

We study the interactive model of comparison-based collaborative filtering. Each player prefers one object from each pair of objects. However, revealing what is a player preference between two objects can be done only by asking the player specifically about that pair, an action called probing. The goal is to (approximately) reconstruct the players’ preferences with the smallest possible number of probes per player. The per-player number of probes can be reduced if there are many players who share a similar taste, but a priori, players do not know who to collaborate with. In this paper, we present the model of comparison-based interactive collaborative filtering, analyze a few possible taste models and present distributed algorithms whose output is close to the best possible approximation to the players’ taste.