Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Interpretable Triplet Importance for Personalized Ranking

Bowei He, Chen Ma

TL;DR

This work introduces ITIPR, a framework that assigns interpretable Shapley-based importance values to each (user, positive item, negative item) triplet in Bayesian personalized ranking. It constructs a Monte Carlo-based, variance-controlled estimator for Triplet Shapley values and uses a TIP model to resample high-importance triplets, training a weighted BPR objective to boost performance. Empirical results across six public datasets and multiple backbone models (MF and GNNs) show consistent gains over state-of-the-art baselines, with improved stability and interpretability. The approach offers a principled, scalable pathway to more transparent and effective personalized ranking from implicit feedback.

Abstract

Personalized item ranking has been a crucial component contributing to the performance of recommender systems. As a representative approach, pairwise ranking directly optimizes the ranking with user implicit feedback by constructing (\textit{user}, \textit{positive item}, \textit{negative item}) triplets. Several recent works have noticed that treating all triplets equally may hardly achieve the best effects. They assign different importance scores to negative items, user-item pairs, or triplets, respectively. However, almost all the generated importance scores are groundless and hard to interpret, thus far from trustworthy and transparent. To tackle these, we propose the \textit{Triplet Shapley} -- a Shapely value-based method to measure the triplet importance in an interpretable manner. Due to the huge number of triplets, we transform the original Shapley value calculation to the Monte Carlo (MC) approximation, where the guarantee for the approximation unbiasedness is also provided. To stabilize the MC approximation, we adopt a control covariates-based method. Finally, we utilize the triplet Shapley value to guide the resampling of important triplets for benefiting the model learning. Extensive experiments are conducted on six public datasets involving classical matrix factorization- and graph neural network-based recommendation models. Empirical results and subsequent analysis show that our model consistently outperforms the state-of-the-art methods.

Interpretable Triplet Importance for Personalized Ranking

TL;DR

This work introduces ITIPR, a framework that assigns interpretable Shapley-based importance values to each (user, positive item, negative item) triplet in Bayesian personalized ranking. It constructs a Monte Carlo-based, variance-controlled estimator for Triplet Shapley values and uses a TIP model to resample high-importance triplets, training a weighted BPR objective to boost performance. Empirical results across six public datasets and multiple backbone models (MF and GNNs) show consistent gains over state-of-the-art baselines, with improved stability and interpretability. The approach offers a principled, scalable pathway to more transparent and effective personalized ranking from implicit feedback.

Abstract

Personalized item ranking has been a crucial component contributing to the performance of recommender systems. As a representative approach, pairwise ranking directly optimizes the ranking with user implicit feedback by constructing (\textit{user}, \textit{positive item}, \textit{negative item}) triplets. Several recent works have noticed that treating all triplets equally may hardly achieve the best effects. They assign different importance scores to negative items, user-item pairs, or triplets, respectively. However, almost all the generated importance scores are groundless and hard to interpret, thus far from trustworthy and transparent. To tackle these, we propose the \textit{Triplet Shapley} -- a Shapely value-based method to measure the triplet importance in an interpretable manner. Due to the huge number of triplets, we transform the original Shapley value calculation to the Monte Carlo (MC) approximation, where the guarantee for the approximation unbiasedness is also provided. To stabilize the MC approximation, we adopt a control covariates-based method. Finally, we utilize the triplet Shapley value to guide the resampling of important triplets for benefiting the model learning. Extensive experiments are conducted on six public datasets involving classical matrix factorization- and graph neural network-based recommendation models. Empirical results and subsequent analysis show that our model consistently outperforms the state-of-the-art methods.
Paper Structure (39 sections, 4 theorems, 19 equations, 3 figures, 5 tables, 2 algorithms)

This paper contains 39 sections, 4 theorems, 19 equations, 3 figures, 5 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Problem Formulation
  5. Bayesian Personalized Ranking
  6. Methodology
  7. Triplet Importance for Pairwise Learning
  8. Triplet Shapley as Interpretable Importance
  9. Approximating Triplet Shapley
  10. Identical Transformation to Triplet Shapley
  11. Monte Carlo Approximation
  12. Efficient Triplet Shapley Approximation
  13. Stabilizing Triplet Shapley Approximation
  14. Importance-aware Triplet Resampling
  15. Overall Learning Procedure
  16. ...and 24 more sections

Key Result

Theorem 1

Triplet Shapley from Joining Process Perspective. Denote by $\Pi^{T}$ the set of all possible permutations of the triplets in $D^{T}$, each of which representing a distinct joining order. Moreover, let $P^{\pi}_{(u,i,j)}$ denote the set of triplets that precede $(u,i,j)$ in the permutation $\pi \in

Figures (3)

  • Figure 1: The illustration of vanilla BPR and corresponding four categories of improvement methods.
  • Figure 2: Distribution (in violin plot) of the estimates for triplet Shapley (w/o control variate) and triplet Shapley, based on 6 randomly selected triplets across 5 independent runs on Amazon-Books and Amazon-CDs datasets.
  • Figure 3: Training time comparison of various methods with different base models on Amazon-Books and ML-20M.

Theorems & Definitions (5)

  • Definition 1
  • Theorem 1
  • Theorem 2
  • Lemma 1
  • Theorem 3