Adversarial Binary Collaborative Filtering for Implicit Feedback

• DOI: 10.1609/aaai.v33i01.33015248
• Publication Date: 2019-08-30T07:41:49Z
• Authors: Haoyu Wang; Nan Shao; Defu Lian
• Keywords: Discriminative model; Generative model; Binary classification

Fast item recommendation based on implicit feedback is vital in practical scenarios due to data-abundance, but challenging because of the lack negative samples and large number recommended items. Recent adversarial methods unifying generative discriminative models are promising, since model, as a sampler, gradually improves iteration continues. However, binary-valued model still unexplored within min-max framework, important for accelerating recommendation. Optimizing difficult non-smooth nondifferentiable. To this end, we propose two novel relax binarization error function Gumbel trick so that can be optimized by many popular solvers, such SGD ADMM. The then evaluated framework four real-world datasets shown its superiority competing hashing-based algorithms. In addition, our proposed approximate discrete variables precisely applied solve other optimization problems.