이용 전 선호도를 이용한 추천 시스템 내 누락된 평점 추론

Abstract

Data sparsity is the biggest problem faced by collaborative filtering used in recommender systems. Data imputation alleviates the data sparsity problem by inferring missing ratings and imputing them to the original rating matrix. Existing imputation approaches, however, have failed to take the unique characteristics of missing ratings into account. In this paper, we identify the limitations of existing data imputation approaches and make three claims that any data imputation approaches should follow. Furthermore, we hypothesize that in most cases pre-use preference leads to post-use preference. Based on our hypothesis, we propose to impute NPP(Normalized Post-use preferences originate from Pre-use preference)-values, which satisfy all three claims. We also define the notion of confidence for each NPP-value, and propose Confidence-Awareness SVD (CA-SVD) that incorporates such confidences into SVD. Through extensive experiments, we demonstrate that our solution outperforms existing state-of-the-arts significantly.