Inferring Users’ Preferences from Crowdsourced Pairwise Comparisons: A Matrix Completion Approach

摘要: Inferring user preferences over a set of items is an important problem that has found numerous applications. This work focuses on the scenario where explicit feature representation unavailable, setup similar to collaborative filtering. In order learn user's from his/her response only small number pairwise comparisons, we propose leverage comparisons made by many crowd users, refer as crowdranking. The proposed crowdranking framework based theory matrix completion, and present efficient algorithms for solving related optimization problem. Our theoretical analysis shows that, average, O ( r log m ) queries are needed accurately recover ranking list target user, rank unknown rating matrix, << . empirical study with two real-world benchmark datasets filtering one dataset collected via Amazon Mechanical Turk promising performance algorithm compared state-of-the-art approaches.