[Paper Review] TFPS: A Temporal Filtration-enhanced Positive Sample Set Construction Method for Implicit Collaborative Filtering

Jiayi Wu, Zhengyu Wu+3 more|arXiv (Cornell University)|Feb 26, 2026

Recommender Systems and Techniques0 citations

TL;DR

TFPS introduces temporal filtration to build a weighted, layered positive sample set from implicit feedback, improving Recall@k and NDCG@k when used with negative-sampling-based implicit CF. It weights recent interactions, creates layered subgraphs via graph filtration, and applies layer-enhancement to emphasize current-user preferences.

ABSTRACT

The negative sampling strategy can effectively train collaborative filtering (CF) recommendation models based on implicit feedback by constructing positive and negative samples. However, existing methods primarily optimize the negative sampling process while neglecting the exploration of positive samples. Some denoising recommendation methods can be applied to denoise positive samples within negative sampling strategies, but they ignore temporal information. Existing work integrates sequential information during model aggregation but neglects time interval information, hindering accurate capture of users' current preferences. To address this problem, from a data perspective, we propose a novel temporal filtration-enhanced approach to construct a high-quality positive sample set. First, we design a time decay model based on interaction time intervals, transforming the original graph into a weighted user-item bipartite graph. Then, based on predefined filtering operations, the weighted user-item bipartite graph is layered. Finally, we design a layer-enhancement strategy to construct a high-quality positive sample set for the layered subgraphs. We provide theoretical insights into why TFPS can improve Recall@k and NDCG@k, and extensive experiments on three real-world datasets demonstrate the effectiveness of the proposed method. Additionally, TFPS can be integrated with various implicit CF recommenders or negative sampling methods to enhance its performance.

Motivation & Objective

• Motivate improved learning from implicit feedback by focusing on high-quality positive samples that reflect current user preferences.
• Introduce a data-level temporal filtration approach to construct a layered, enhanced positive sample set.
• Provide theoretical justification for margin amplification and empirical evidence across real-world datasets.
• Demonstrate TFPS’s compatibility with various implicit CF models and negative sampling strategies.

Proposed method

• Weight each user-item interaction with a time-decay based on the most recent user interaction time.
• Apply edge-weight filtration to partition the weighted bipartite graph into multiple layered subgraphs by weight intervals.
• Use a layer-enhancement scheme that duplicates high-weight edges according to their layer index to form the positive sample set (PSS).
• Train with BPR loss on PSS, which implicitly reweights samples by frequency of occurrence in PSS.
• Provide theoretical analysis showing margin amplification from data-level reweighting and its impact on Recall@k and NDCG@k.

Experimental results

Research questions

• RQ1 Can TFPS improve implicit CF performance by incorporating temporal information into positive samples?
• RQ2 How does the layer-enhancement intensity affect performance and generalization?
• RQ3 Is TFPS compatible with different negative sampling strategies and CF models?
• RQ4 How does TFPS compare with sequential models under timestamp-partitioned evaluation?

Key findings

• TFPS outperforms state-of-the-art baselines on three real-world datasets (AmazonCDs, LastFM, Ta-Feng) across Recall@20/30 and NDCG@20/30.
• Layer-enhancement emphasizes recent interactions, improving model focus on current preferences without sacrificing long-term signal.
• TFPS can be integrated with STAM and other negative sampling methods to boost performance further (TFPS-STAM shown).
• Theoretical results show that duplicating high-weight positives amplifies margin gains, translating to better ranking metrics under timestamp-partitioned evaluation.
• TFPS maintains linear time complexity O(|E|) and requires a single pre-processing pass before training.