Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Level Sequence Denoising with Cross-Signal Contrastive Learning for Sequential Recommendation

Xiaofei Zhu, Liang Li, Weidong Liu, Xin Luo

TL;DR

This work tackles noisy items in sequential recommendation by introducing MSDCCL, a framework that jointly employs soft and hard denoising guided by cross-signal contrastive learning. It combines a target-aware user-interest extractor to capture both long- and short-term preferences, a dual denoising module with a Gumbel-softmax-based hard signal, and an S-shaped curriculum to better mimic human learning. Empirical results on five public datasets show consistent improvements over strong baselines and existing denoising methods, with ablations confirming the necessity of each component. The approach is compatible with existing recommender backbones and offers practical gains for robust next-item prediction in noisy sequences.

Abstract

Sequential recommender systems (SRSs) aim to suggest next item for a user based on her historical interaction sequences. Recently, many research efforts have been devoted to attenuate the influence of noisy items in sequences by either assigning them with lower attention weights or discarding them directly. The major limitation of these methods is that the former would still prone to overfit noisy items while the latter may overlook informative items. To the end, in this paper, we propose a novel model named Multi-level Sequence Denoising with Cross-signal Contrastive Learning (MSDCCL) for sequential recommendation. To be specific, we first introduce a target-aware user interest extractor to simultaneously capture users' long and short term interest with the guidance of target items. Then, we develop a multi-level sequence denoising module to alleviate the impact of noisy items by employing both soft and hard signal denoising strategies. Additionally, we extend existing curriculum learning by simulating the learning pattern of human beings. It is worth noting that our proposed model can be seamlessly integrated with a majority of existing recommendation models and significantly boost their effectiveness. Experimental studies on five public datasets are conducted and the results demonstrate that the proposed MSDCCL is superior to the state-of-the-art baselines. The source code is publicly available at https://github.com/lalunex/MSDCCL/tree/main.

Multi-Level Sequence Denoising with Cross-Signal Contrastive Learning for Sequential Recommendation

TL;DR

This work tackles noisy items in sequential recommendation by introducing MSDCCL, a framework that jointly employs soft and hard denoising guided by cross-signal contrastive learning. It combines a target-aware user-interest extractor to capture both long- and short-term preferences, a dual denoising module with a Gumbel-softmax-based hard signal, and an S-shaped curriculum to better mimic human learning. Empirical results on five public datasets show consistent improvements over strong baselines and existing denoising methods, with ablations confirming the necessity of each component. The approach is compatible with existing recommender backbones and offers practical gains for robust next-item prediction in noisy sequences.

Abstract

Sequential recommender systems (SRSs) aim to suggest next item for a user based on her historical interaction sequences. Recently, many research efforts have been devoted to attenuate the influence of noisy items in sequences by either assigning them with lower attention weights or discarding them directly. The major limitation of these methods is that the former would still prone to overfit noisy items while the latter may overlook informative items. To the end, in this paper, we propose a novel model named Multi-level Sequence Denoising with Cross-signal Contrastive Learning (MSDCCL) for sequential recommendation. To be specific, we first introduce a target-aware user interest extractor to simultaneously capture users' long and short term interest with the guidance of target items. Then, we develop a multi-level sequence denoising module to alleviate the impact of noisy items by employing both soft and hard signal denoising strategies. Additionally, we extend existing curriculum learning by simulating the learning pattern of human beings. It is worth noting that our proposed model can be seamlessly integrated with a majority of existing recommendation models and significantly boost their effectiveness. Experimental studies on five public datasets are conducted and the results demonstrate that the proposed MSDCCL is superior to the state-of-the-art baselines. The source code is publicly available at https://github.com/lalunex/MSDCCL/tree/main.
Paper Structure (33 sections, 15 equations, 10 figures, 4 tables)

This paper contains 33 sections, 15 equations, 10 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Sequential Recommendation
  4. Contrastive Learning
  5. Denoising Methods
  6. Preliminaries
  7. Methodology
  8. Embedding Layer
  9. Target-Aware User Interest Extractor
  10. Long-term User Interests
  11. Short-term User Interests
  12. Interests Fusion
  13. Soft- and Hard-level Sequence Denoising
  14. User-Item Correlation Generator
  15. Soft-level Denoising
  16. ...and 18 more sections

Figures (10)

  • Figure 1: The architecture of our proposed model MSDCCL.
  • Figure 2: Target-aware convolutional sequence embedding.
  • Figure 3: S-shape curriculum learning vs. Linear shape curriculum learning.
  • Figure 4: Analysis on the effectiveness of different increment patterns in curriculum learning, specifically linear shape versus S-shape.
  • Figure 5: Performance of our model and the two best performing baselines (i.e., AC-BERT4Rec, HSD+BERT4Rec) under different sequence lengths on ML-100k and Beauty.
  • ...and 5 more figures