A Simple yet Effective Negative Sampling Plugin for Constructing Positive Sample Pairs in Implicit Collaborative Filtering

TL;DR

This paper tackles implicit collaborative filtering by addressing a blind spot in negative sampling: the quality of positive supervision. It introduces PSP-NS, a plugin that constructs high-confidence positive sample pairs via a weighted user-item graph that blends global interaction patterns from randomized SVD with local signals, and reinforces learning through replication-based weighting and activity-aware user weights. Theoretical analyses show margin improvements for true positives and amplified gains for inactive users, while experiments on four real-world datasets demonstrate consistent, substantial gains in ranking metrics and robustness across seeds. PSP-NS is designed as a plug-and-play enhancement that can boost a wide range of implicit CF models and negative sampling strategies, offering a practical approach to more accurate user preference learning.

Abstract

Most implicit collaborative filtering (CF) models are trained with negative sampling, where existing work designs sophisticated strategies for high-quality negatives while largely overlooking the exploration of positive samples. Although some denoising recommendation methods can be applied to implicit CF for denoising positive samples, they often sparsify positive supervision. Moreover, these approaches generally overlook user activity bias during training, leading to insufficient learning for inactive users. To address these issues, we propose a simple yet effective negative sampling plugin, PSP-NS, from the perspective of enhancing positive supervision signals. It builds a user-item bipartite graph with edge weights indicating interaction confidence inferred from global and local patterns, generates positive sample pairs via replication-based reweighting to strengthen positive signals, and adopts an activity-aware weighting scheme to effectively learn inactive users' preferences. We provide theoretical insights from a margin-improvement perspective, explaining why PSP-NS tends to improve ranking quality (e.g., Precision@k/Recall@k), and conduct extensive experiments on four real-world datasets to demonstrate its superiority. For instance, PSP-NS boosts Recall@30 and Precision@30 by 32.11% and 22.90% on Yelp over the strongest baselines. PSP-NS can be integrated with various implicit CF recommenders or negative sampling methods to enhance their performance.

Figures (3)

• Figure 1: An overview of PSP-NS.
• Figure 2: (a)-(b) Recall@20 vs. wall-clock time (s). (c)-(d) Performance of integrating PSP-NS with different negative sampling methods.
• Figure 3: The impact of $q$, $a$, and $s$ on Recall@20 and Precision@20.