Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Denoising Time Cycle Modeling for Recommendation

Sicong Xie, Qunwei Li, Weidi Xu, Kaiming Shen, Shaohu Chen, Wenliang Zhong

TL;DR

DiCycle addresses the challenge of noisy temporal signals in recommendations by denoising past user behaviors and explicitly modeling two complementary time-cycle patterns: Absolute Time Cycle ($ATC$) and Relative Time Cycle ($RTC$). It introduces a Time Encoding Unit to produce absolute and relative time representations, a Gated Filter Unit to prune irrelevant histories, and a Time Cycle Module with time-cycle attention, implemented via a User Interest Module and a Time Cycle Module. The framework is validated on three public benchmarks and a real-world IndRec dataset, showing consistent improvements over state-of-the-art baselines and strong ablation support for each component. By differentiating target-related temporal signals and capturing daily/weekly cycles, DiCycle offers practical benefits for CTR prediction in dynamic recommender systems.

Abstract

Recently, modeling temporal patterns of user-item interactions have attracted much attention in recommender systems. We argue that existing methods ignore the variety of temporal patterns of user behaviors. We define the subset of user behaviors that are irrelevant to the target item as noises, which limits the performance of target-related time cycle modeling and affect the recommendation performance. In this paper, we propose Denoising Time Cycle Modeling (DiCycle), a novel approach to denoise user behaviors and select the subset of user behaviors that are highly related to the target item. DiCycle is able to explicitly model diverse time cycle patterns for recommendation. Extensive experiments are conducted on both public benchmarks and a real-world dataset, demonstrating the superior performance of DiCycle over the state-of-the-art recommendation methods.

Denoising Time Cycle Modeling for Recommendation

TL;DR

DiCycle addresses the challenge of noisy temporal signals in recommendations by denoising past user behaviors and explicitly modeling two complementary time-cycle patterns: Absolute Time Cycle () and Relative Time Cycle (). It introduces a Time Encoding Unit to produce absolute and relative time representations, a Gated Filter Unit to prune irrelevant histories, and a Time Cycle Module with time-cycle attention, implemented via a User Interest Module and a Time Cycle Module. The framework is validated on three public benchmarks and a real-world IndRec dataset, showing consistent improvements over state-of-the-art baselines and strong ablation support for each component. By differentiating target-related temporal signals and capturing daily/weekly cycles, DiCycle offers practical benefits for CTR prediction in dynamic recommender systems.

Abstract

Recently, modeling temporal patterns of user-item interactions have attracted much attention in recommender systems. We argue that existing methods ignore the variety of temporal patterns of user behaviors. We define the subset of user behaviors that are irrelevant to the target item as noises, which limits the performance of target-related time cycle modeling and affect the recommendation performance. In this paper, we propose Denoising Time Cycle Modeling (DiCycle), a novel approach to denoise user behaviors and select the subset of user behaviors that are highly related to the target item. DiCycle is able to explicitly model diverse time cycle patterns for recommendation. Extensive experiments are conducted on both public benchmarks and a real-world dataset, demonstrating the superior performance of DiCycle over the state-of-the-art recommendation methods.
Paper Structure (15 sections, 8 equations, 3 figures, 3 tables)

This paper contains 15 sections, 8 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Denoising Time Cycle Modeling
  4. Problem Formulation
  5. Time Encoding Unit
  6. Learning Absolute Time Representation for ATC
  7. Learning Relative Time Representation for RTC
  8. Gated Filter Unit
  9. Time Cycle Modeling
  10. User Interest Module
  11. Time Cycle Module
  12. EXPERIMENTS
  13. Experimental Settings
  14. Results and Discussions
  15. CONCLUSION

Figures (3)

  • Figure 1: Analysis of two typical temporal patterns of user behaviors.
  • Figure 2: The overall architecture of the proposed model.
  • Figure 3: (a), (b): Changing next interaction time for users who had many interactions and a few interactions with target item, respectively.