Separating-Plane Factorization Models: Scalable Recommendation from One-class Implicit Feedback

Abstract

We study the large-scale video recommendation problem based on user viewing logs instead of explicit ratings. As viewing records are implicitly positive samples, existing matrix factorization methods fail to generate discriminative recommendations based on such one-class data. We propose a scalable approach called separating-plane matrix factorization (SPMF) to make effective recommendations based on one-class implicit feedback, with a learning complexity only comparable to matrix factorization. With extensive offline evaluation in Tencent Data Warehouse (TDW) based on big data, we show that our approach outperforms a wide range of state-of-the-art methods. We also deployed our system online to test with real users in Tencent QQ Browser mobile app. Results show that our approach can increase the video click through rate by 23% over implicit-feedback collaborative filtering (IFCF), a scheme implemented in Spark’s MLlib.