Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Long-term Embedding with Denoising and Augmentation for Recommendation

Zahra Akhlaghi, Mostafa Haghir Chehreghani

TL;DR

ALDA4Rec addresses noise and dynamic user preferences in graph-based sequential recommendations by combining a similarity-based item-item graph with community-detection denoising and edge augmentation. It learns short-term representations via a GCN on time-sliced graphs, and long-term preferences through interval-level GRU/attention, instant-level attention, and mean-level averaging, with an MLP-driven adaptive weight to balance components. Empirical results on four real-world datasets show consistent improvements over state-of-the-art baselines in both accuracy and robustness, validating the preprocessing-denoising and adaptive long-term strategies. The approach offers practical benefits for robust, temporal recommendation in real systems, with scalable preprocessing and flexible long-term modeling to capture evolving user interests.

Abstract

The rapid growth of the internet has made personalized recommendation systems indispensable. Graph-based sequential recommendation systems, powered by Graph Neural Networks (GNNs), effectively capture complex user-item interactions but often face challenges such as noise and static representations. In this paper, we introduce the Adaptive Long-term Embedding with Denoising and Augmentation for Recommendation (ALDA4Rec) method, a novel model that constructs an item-item graph, filters noise through community detection, and enriches user-item interactions. Graph Convolutional Networks (GCNs) are then employed to learn short-term representations, while averaging, GRUs, and attention mechanisms are utilized to model long-term embeddings. An MLP-based adaptive weighting strategy is further incorporated to dynamically optimize long-term user preferences. Experiments conducted on four real-world datasets demonstrate that ALDA4Rec outperforms state-of-the-art baselines, delivering notable improvements in both accuracy and robustness. The source code is available at https://github.com/zahraakhlaghi/ALDA4Rec.

Adaptive Long-term Embedding with Denoising and Augmentation for Recommendation

TL;DR

ALDA4Rec addresses noise and dynamic user preferences in graph-based sequential recommendations by combining a similarity-based item-item graph with community-detection denoising and edge augmentation. It learns short-term representations via a GCN on time-sliced graphs, and long-term preferences through interval-level GRU/attention, instant-level attention, and mean-level averaging, with an MLP-driven adaptive weight to balance components. Empirical results on four real-world datasets show consistent improvements over state-of-the-art baselines in both accuracy and robustness, validating the preprocessing-denoising and adaptive long-term strategies. The approach offers practical benefits for robust, temporal recommendation in real systems, with scalable preprocessing and flexible long-term modeling to capture evolving user interests.

Abstract

The rapid growth of the internet has made personalized recommendation systems indispensable. Graph-based sequential recommendation systems, powered by Graph Neural Networks (GNNs), effectively capture complex user-item interactions but often face challenges such as noise and static representations. In this paper, we introduce the Adaptive Long-term Embedding with Denoising and Augmentation for Recommendation (ALDA4Rec) method, a novel model that constructs an item-item graph, filters noise through community detection, and enriches user-item interactions. Graph Convolutional Networks (GCNs) are then employed to learn short-term representations, while averaging, GRUs, and attention mechanisms are utilized to model long-term embeddings. An MLP-based adaptive weighting strategy is further incorporated to dynamically optimize long-term user preferences. Experiments conducted on four real-world datasets demonstrate that ALDA4Rec outperforms state-of-the-art baselines, delivering notable improvements in both accuracy and robustness. The source code is available at https://github.com/zahraakhlaghi/ALDA4Rec.

Paper Structure

This paper contains 33 sections, 21 equations, 5 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. Graph-based recommendation
  4. Sequential recommender systems
  5. Denoising in recommender systems
  6. Preliminaries
  7. Methodology
  8. Preprocessing
  9. Similarity-based graph construction
  10. Noise detection
  11. Edge augmentation
  12. Embedding learning and prediction
  13. Short-term embedding learning
  14. Long-term embedding learning
  15. Interval-level long-term embedding
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Overall framework of the proposed ALDA4Rec model
  • Figure 2: Extracting user interactions from $I_t^{sim}$
  • Figure 3: Performance of our model for various values of $\lambda_1$ in HR@10.
  • Figure 4: Analysis of the Performance of Our Model for Various Values of $\beta$ in HR@10.
  • Figure 5: Impact of $min_{sim}$ variation on model performance in HR@10.