Contrastive Self-supervised Sequential Recommendation with Robust Augmentation

• DOI: 10.48550/arxiv.2108.06479
• Publication Date: 2021-01-01
• Authors: Zhiwei Liu; Yongjun Chen; Jia Li; Philip S. Yu; Julian McAuley; Caiming Xiong
• Keywords: FOS: Computer and information sciences; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; 0202 electrical engineering, electronic engineering, information engineering; 02 engineering and technology; Information Retrieval (cs.IR); Computer Science - Information Retrieval

Sequential Recommendationdescribes a set of techniques to model dynamic user behavior in order predict future interactions sequential data. At their core, such approaches transition probabilities between items sequence, whether through Markov chains, recurrent networks, or more recently, Transformers. However both old and new issues remain, including data-sparsity noisy data; can impair the performance, especially complex, parameter-hungry models. In this paper, we investigate application contrastive Self-Supervised Learning (SSL) recommendation, as way alleviate some these issues. Contrastive SSL constructs augmentations from unlabelled instances, where agreements among positive pairs are maximized. It is challenging devise framework for due its discrete nature, correlations items, skewness length distributions. To end, propose novel framework, Self-supervised Recommendation (CoSeRec). We introduce two informative augmentation operators leveraging item create high-quality views learning. Experimental results on three real-world datasets demonstrate effectiveness proposed method improving performance robustness against sparse Our implementation available online at \url{https://github.com/YChen1993/CoSeRec}