Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Reinforcement Learning Method to Factual and Counterfactual Explanations for Session-based Recommendation

Han Zhou, Hui Fang, Zhu Sun, Wentao Hu

TL;DR

The paper tackles the opacity of session-based recommender systems by introducing FCESR, a model-agnostic framework that generates both factual and counterfactual explanations for SR predictions using reinforcement learning. It reframes explanation generation as a combinatorial optimization problem solved within an MDP, and then uses the derived explanations as high-quality positive and negative samples in a contrastive learning setup to boost SR accuracy. Extensive experiments across three datasets and five SR backbones show that FCESR improves recommendation performance while delivering meaningful, concise explanations, bridging interpretability and usefulness. This approach advances trustworthy SR by providing true explanations and demonstrating practical gains through explanation-informed training.

Abstract

Session-based Recommendation (SR) systems have recently achieved considerable success, yet their complex, "black box" nature often obscures why certain recommendations are made. Existing explanation methods struggle to pinpoint truly influential factors, as they frequently depend on static user profiles or fail to grasp the intricate dynamics within user sessions. In response, we introduce FCESR (Factual and Counterfactual Explanations for Session-based Recommendation), a novel framework designed to illuminate SR model predictions by emphasizing both the sufficiency (factual) and necessity (counterfactual) of recommended items. By recasting explanation generation as a combinatorial optimization challenge and leveraging reinforcement learning, our method uncovers the minimal yet critical sequence of items influencing recommendations. Moreover, recognizing the intrinsic value of robust explanations, we innovatively utilize these factual and counterfactual insights within a contrastive learning paradigm, employing them as high-quality positive and negative samples to fine-tune and significantly enhance SR accuracy. Extensive qualitative and quantitative evaluations across diverse datasets and multiple SR architectures confirm that our framework not only boosts recommendation accuracy but also markedly elevates the quality and interpretability of explanations, thereby paving the way for more transparent and trustworthy recommendation systems.

A Reinforcement Learning Method to Factual and Counterfactual Explanations for Session-based Recommendation

TL;DR

The paper tackles the opacity of session-based recommender systems by introducing FCESR, a model-agnostic framework that generates both factual and counterfactual explanations for SR predictions using reinforcement learning. It reframes explanation generation as a combinatorial optimization problem solved within an MDP, and then uses the derived explanations as high-quality positive and negative samples in a contrastive learning setup to boost SR accuracy. Extensive experiments across three datasets and five SR backbones show that FCESR improves recommendation performance while delivering meaningful, concise explanations, bridging interpretability and usefulness. This approach advances trustworthy SR by providing true explanations and demonstrating practical gains through explanation-informed training.

Abstract

Session-based Recommendation (SR) systems have recently achieved considerable success, yet their complex, "black box" nature often obscures why certain recommendations are made. Existing explanation methods struggle to pinpoint truly influential factors, as they frequently depend on static user profiles or fail to grasp the intricate dynamics within user sessions. In response, we introduce FCESR (Factual and Counterfactual Explanations for Session-based Recommendation), a novel framework designed to illuminate SR model predictions by emphasizing both the sufficiency (factual) and necessity (counterfactual) of recommended items. By recasting explanation generation as a combinatorial optimization challenge and leveraging reinforcement learning, our method uncovers the minimal yet critical sequence of items influencing recommendations. Moreover, recognizing the intrinsic value of robust explanations, we innovatively utilize these factual and counterfactual insights within a contrastive learning paradigm, employing them as high-quality positive and negative samples to fine-tune and significantly enhance SR accuracy. Extensive qualitative and quantitative evaluations across diverse datasets and multiple SR architectures confirm that our framework not only boosts recommendation accuracy but also markedly elevates the quality and interpretability of explanations, thereby paving the way for more transparent and trustworthy recommendation systems.

Paper Structure

This paper contains 21 sections, 1 theorem, 16 equations, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Explainable Recommendation
  4. Factual and Counterfactual Explanation
  5. Contrastive Learning
  6. Contrastive Learning
  7. Preliminary
  8. Task Formulation
  9. Analysis of the Min-Explanation Problem
  10. Our FCESR Framework
  11. Explanation Generation Module
  12. Recommendation Enhancement Module
  13. Discussion on Optimality and Feasibility
  14. Experiments
  15. Experimental Settings
  16. ...and 6 more sections

Key Result

Theorem 1

The Min-Explanation problem is NP-Hard.

Figures (5)

  • Figure 1: Definitions of factual and counterfactual explanations. Circles represent items in a session, where yellow indicates key counterfactual items and green indicates key factual items: (a) counterfactual reasoning: removing $v_1$ and $v_2$ changes the recommendation from $v_7$ to $v_8$; (b) factual reasoning: the presence of $v_1$ to $v_5$ is sufficient to recommend $v_7$; and (c) counterfactual and factual reasoning: {$v_1$, $v_2$, $v_3$} are both sufficient and necessary conditions for recommending $v_7$.
  • Figure 2: Overview of the FCESR framework.
  • Figure 3: Impact of Positive and Negative Samples.
  • Figure 4: Impact of different lr.
  • Figure 5: Impact of different $\lambda$.

Theorems & Definitions (5)

  • Definition 1
  • Definition 2
  • Definition 3
  • Definition 4
  • Theorem 1