Dual Adversarial Variational Embedding for Robust Recommendation

Robust recommendation aims at capturing true preference of users from noisy data, for which there are two lines methods have been proposed. One is based on noise injection, and the other to adopt generative model Variational Auto-encoder (VAE). However, existing works still face challenges. First, injection often draw a fixed distribution given in advance, while real world, distributions different items may differ each due personal behaviors item usage patterns. Second, VAE models not expressive enough capture since yields an embedding space single modal, user-item interactions usually exhibit multi-modality user distribution. In this paper, we propose novel called Dual Adversarial Embedding (DAVE) robust recommendation, can provide personalized reduction items, space, by combining advantages adversarial training between introduced auxiliary discriminators variational inference networks. The extensive experiments conducted datasets verify effectiveness DAVE recommendation.