Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RSAttAE: An Information-Aware Attention-based Autoencoder Recommender System

Amirhossein Dadashzadeh Taromi, Sina Heydari, Mohsen Hooshmand, Majid Ramezani

TL;DR

The paper tackles rating prediction in recommender systems by learning feature-informed latent representations through an attention-based autoencoder (RSAttAE) and then applying a supervised XGBoost regressor for rating prediction. Two RSAttAE modules process the rating matrix and side information to produce user and item embeddings, which are concatenated with feature vectors and fed to XGBoost. Using MovieLens 100K, the authors systematically tune the embedding dimension and XGBoost hyperparameters, showing that a dimensionality of 64 and careful boosting parameters yield strong performance. Overall, RSAttAE delivers near-state-of-the-art results on ML 100K, demonstrating robust, feature-aware representation learning with practical impact for improved rating prediction and matrix completion.

Abstract

Recommender systems play a crucial role in modern life, including information retrieval, the pharmaceutical industry, retail, and entertainment. The entertainment sector, in particular, attracts significant attention and generates substantial profits. This work proposes a new method for predicting unknown user-movie ratings to enhance customer satisfaction. To achieve this, we utilize the MovieLens 100K dataset. Our approach introduces an attention-based autoencoder to create meaningful representations and the XGBoost method for rating predictions. The results demonstrate that our proposal outperforms most of the existing state-of-the-art methods. Availability: github.com/ComputationIASBS/RecommSys

RSAttAE: An Information-Aware Attention-based Autoencoder Recommender System

TL;DR

The paper tackles rating prediction in recommender systems by learning feature-informed latent representations through an attention-based autoencoder (RSAttAE) and then applying a supervised XGBoost regressor for rating prediction. Two RSAttAE modules process the rating matrix and side information to produce user and item embeddings, which are concatenated with feature vectors and fed to XGBoost. Using MovieLens 100K, the authors systematically tune the embedding dimension and XGBoost hyperparameters, showing that a dimensionality of 64 and careful boosting parameters yield strong performance. Overall, RSAttAE delivers near-state-of-the-art results on ML 100K, demonstrating robust, feature-aware representation learning with practical impact for improved rating prediction and matrix completion.

Abstract

Recommender systems play a crucial role in modern life, including information retrieval, the pharmaceutical industry, retail, and entertainment. The entertainment sector, in particular, attracts significant attention and generates substantial profits. This work proposes a new method for predicting unknown user-movie ratings to enhance customer satisfaction. To achieve this, we utilize the MovieLens 100K dataset. Our approach introduces an attention-based autoencoder to create meaningful representations and the XGBoost method for rating predictions. The results demonstrate that our proposal outperforms most of the existing state-of-the-art methods. Availability: github.com/ComputationIASBS/RecommSys

Paper Structure

This paper contains 15 sections, 4 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Realted work
  3. Collaborative Filtering methods:
  4. Hybrid Filtering methods:
  5. Dataset
  6. Proposed method
  7. Preprocessing
  8. Framework
  9. Attention-based Autoencoder
  10. Training XGBoost model
  11. Results
  12. Embedding Dimension($d$)
  13. XGBoost regressor hyperparameter tuning
  14. RSAttAE RMSE comparison
  15. Conclusion

Figures (4)

  • Figure 1: The RSAttAE module architecture consists of attention-based encoder and decoder modules.
  • Figure 2: Features distribution of MovieLens 100k
  • Figure 3: The rating prediction procedure using the XGBoost algorithm and the trained RSAttAE encoders.
  • Figure 4: A: Validation loss vs Embedding dimension, which indicates $d=64$ as the best embedding dimension. B: Validation loss vs Epochs for both vanilla Autoencoder and RSAttAE demonstrates the superiority of RSAttAE in convergence.