Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Criteria-Aware Graph Filtering: Extremely Fast Yet Accurate Multi-Criteria Recommendation

Jin-Duk Park, Jaemin Yoo, Won-Yong Shin

TL;DR

This work tackles the inefficiency of training-based multi-criteria recommender systems by introducing CA-GF, a training-free, matrix-decomposition-free graph-filtering framework that leverages a criteria-aware MC graph. It constructs an MC user-expansion graph, applies criterion-specific second-order polynomial low-pass filters on an adjusted item–item graph, and aggregates signals using a criteria-preference mechanism to produce final recommendations. CA-GF achieves extremely fast runtimes (below $0.2$ seconds on large datasets) and up to $24\%$ gains in $NDCG@5$ compared with strong MC baselines, while also offering interpretability through per-criterion attribution maps. The approach demonstrates scalable, accurate MC recommendations with practical impact for real-time decision support and business insights, and it is reproducible via publicly available code.

Abstract

Multi-criteria (MC) recommender systems, which utilize MC rating information for recommendation, are increasingly widespread in various e-commerce domains. However, the MC recommendation using training-based collaborative filtering, requiring consideration of multiple ratings compared to single-criterion counterparts, often poses practical challenges in achieving state-of-the-art performance along with scalable model training. To solve this problem, we propose CA-GF, a training-free MC recommendation method, which is built upon criteria-aware graph filtering for efficient yet accurate MC recommendations. Specifically, first, we construct an item-item similarity graph using an MC user-expansion graph. Next, we design CA-GF composed of the following key components, including 1) criterion-specific graph filtering where the optimal filter for each criterion is found using various types of polynomial low-pass filters and 2) criteria preference-infused aggregation where the smoothed signals from each criterion are aggregated. We demonstrate that CA-GF is (a) efficient: providing the computational efficiency, offering the extremely fast runtime of less than 0.2 seconds even on the largest benchmark dataset, (b) accurate: outperforming benchmark MC recommendation methods, achieving substantial accuracy gains up to 24% compared to the best competitor, and (c) interpretable: providing interpretations for the contribution of each criterion to the model prediction based on visualizations.

Criteria-Aware Graph Filtering: Extremely Fast Yet Accurate Multi-Criteria Recommendation

TL;DR

This work tackles the inefficiency of training-based multi-criteria recommender systems by introducing CA-GF, a training-free, matrix-decomposition-free graph-filtering framework that leverages a criteria-aware MC graph. It constructs an MC user-expansion graph, applies criterion-specific second-order polynomial low-pass filters on an adjusted item–item graph, and aggregates signals using a criteria-preference mechanism to produce final recommendations. CA-GF achieves extremely fast runtimes (below seconds on large datasets) and up to gains in compared with strong MC baselines, while also offering interpretability through per-criterion attribution maps. The approach demonstrates scalable, accurate MC recommendations with practical impact for real-time decision support and business insights, and it is reproducible via publicly available code.

Abstract

Multi-criteria (MC) recommender systems, which utilize MC rating information for recommendation, are increasingly widespread in various e-commerce domains. However, the MC recommendation using training-based collaborative filtering, requiring consideration of multiple ratings compared to single-criterion counterparts, often poses practical challenges in achieving state-of-the-art performance along with scalable model training. To solve this problem, we propose CA-GF, a training-free MC recommendation method, which is built upon criteria-aware graph filtering for efficient yet accurate MC recommendations. Specifically, first, we construct an item-item similarity graph using an MC user-expansion graph. Next, we design CA-GF composed of the following key components, including 1) criterion-specific graph filtering where the optimal filter for each criterion is found using various types of polynomial low-pass filters and 2) criteria preference-infused aggregation where the smoothed signals from each criterion are aggregated. We demonstrate that CA-GF is (a) efficient: providing the computational efficiency, offering the extremely fast runtime of less than 0.2 seconds even on the largest benchmark dataset, (b) accurate: outperforming benchmark MC recommendation methods, achieving substantial accuracy gains up to 24% compared to the best competitor, and (c) interpretable: providing interpretations for the contribution of each criterion to the model prediction based on visualizations.

Paper Structure

This paper contains 33 sections, 4 theorems, 35 equations, 8 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Problem Definition
  4. Graph Signal Processing
  5. Graph Filtering for Single-Criterion Recommendation
  6. Proposed Method: CA-GF
  7. Overview
  8. Graph Construction
  9. MC user-expansion graph construction
  10. Item--item similarity graph with adjustment
  11. Graph Filtering Harnessing Criteria Awareness
  12. Criterion-Specific Graph Filtering
  13. Polynomial Graph Filtering
  14. Criteria Preference-Infused Aggregation
  15. Experimental Evaluation
  16. ...and 18 more sections

Key Result

lemma 1

park2024turboshen2021powerful: The matrix polynomial $\sum_{k=1}^K{a_{c,k}}\bar{P}_f^k$ is a graph filter for graph $\bar{P}_f$, with the frequency response function of $h(\lambda) = \sum_{k=1}^K{a_{c,k}}(1 - \lambda)^k$.

Figures (8)

  • Figure 1: An illustration showing (a) four criteria ratings in a restaurant domain and (b) a comparison of the training time for 100 epochs between using single-criterion ratings ( i.e., overall ratings) and MC ratings across three benchmark MC recommendation methods on the TripAdvisor dataset. Additionally, the processing time is measured for CA-GF that does not need any training process.
  • Figure 2: The schematic overview of CA-GF.
  • Figure 3: Log-scaled runtime comparison of CA-GF (with and without GPU) and $\text{GF-CF}_\text{MC}$ (CPU) using various synthetic datasets, where OOM (GPU) indicates out-of-memory issues on GPU.
  • Figure 4: The effect of three polynomial LPFs $f(\bar{P}_f, c)$ (L, I, and O) for each criterion on the Recall@10.
  • Figure 5: The effect of adjustment parameter $s_{f}$ for three polynomial LPFs (L, I, and O) on the Recall@10.
  • ...and 3 more figures

Theorems & Definitions (8)

  • definition 1
  • definition 2
  • definition 3
  • lemma 1
  • Corollary 1
  • Corollary 2
  • Corollary 3
  • definition 4