Contrastive Self-supervised Sequential Recommendation with Robust Augmentation

TL;DR

CoSeRec tackles data-sparsity and noisy interactions in sequential recommendation by integrating contrastive self-supervised learning with robust, informative augmentations that exploit item correlations and sequence length. It introduces substitute and insert augmentations, along with a hybrid correlation framework, and trains the SR model jointly with SSL in a multi-task setup. Extensive experiments on three real-world datasets demonstrate consistent performance gains and improved robustness, with detailed ablations validating the superiority of informative augmentations and the joint training approach. The work advances practical SR by addressing cold-start and noise through targeted data augmentation and self-supervised signals, and provides open-source code for reproducibility.

Abstract

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

Figures (8)

• Figure 1: Overall framework. (a) illustrates the structure of Transformer. (b) presents the sequence encoding and prediction of Transformer; $v_i$ denotes an item at position $i$ of the sequence $s_u$. (c) demonstrates the contrastive SSL of a sequence. It first augments a sequence as two views with two augmentation methods. Then, it encodes the sequence by concatenating embedding outputs from Transformer. Finally, it maximizes the agreement between the two views.
• Figure 2: Five augmentations adopted in this paper. The original sequence is $[v_1,v_2,v_3,v_4]$. Informative augmentations incorporate correlations among items.
• Figure 3: Performance comparison (in NDCG@5) w.r.t. different augmentation sets on Sports and Beauty datasets. ‘w/o' indicates one type is removed from the full augmentation set $\{\mathsf{M}, \mathsf{C}, \mathsf{R}, \mathsf{S}, \mathsf{I}\}$. CoSeRec adopts the full set.
• Figure 4: Performance comparison (in $\mathrm{NDCG}@5$) w.r.t. different augmentation pairs on Sports and Beauty datasets.
• Figure 5: Performance comparison w.r.t. different augmentation sets for short sequences on Sports and Beauty datasets.