Alleviating the Sparsity in Collaborative Filtering using Crowdsourcing

초록

As a novel method for alleviating the sparsity problem in collaborative filtering (CF), we explore crowdsourcing-based CF, namely CrowdCF, which solicits new ratings from the crowd. We study three key questions that need to be addressed to effectively utilize CrowdCF: (1) how to select items to show for crowd workers to elicit extra ratings, (2) how to decide the minimum quantity asked to the crowd, and (3) how to handle the erroneous ratings. We validate the effectiveness of CrowdCF by conducting offline experiments using real-life datasets and online experiments on Amazon Mechanical Turk. The best configuration of CrowdCF improves system-wide MAE by 0.07 and 0.03, and F1-score by 4% and 2% in offline and online experiments, compared to the state-of-the-art CF algorithm.