Fidelity-Aware Recommendation Explanations via Stochastic Path Integration

TL;DR

This paper tackles the problem of explanation fidelity in recommender systems, where explanations must reflect the true reasoning of the model rather than merely appear plausible. It introduces SPINRec, a model-agnostic method that adapts path-integration to sparse, implicit data by using stochastic baseline sampling from the empirical user-history distribution and selecting the most faithful attribution path. The approach yields superior fidelity across MF, VAE, and NCF on ML1M, Yahoo! Music, and Pinterest, validated with both original AUC-based and refined fixed-length counterfactual metrics, and backed by ablation studies showing the value of stochastic baselines. The work provides a new fidelity benchmark for explainable recommendations and releases code and tools to enable broader evaluation and adoption in practice.

Abstract

Explanation fidelity, which measures how accurately an explanation reflects a model's true reasoning, remains critically underexplored in recommender systems. We introduce SPINRec (Stochastic Path Integration for Neural Recommender Explanations), a model-agnostic approach that adapts path-integration techniques to the sparse and implicit nature of recommendation data. To overcome the limitations of prior methods, SPINRec employs stochastic baseline sampling: instead of integrating from a fixed or unrealistic baseline, it samples multiple plausible user profiles from the empirical data distribution and selects the most faithful attribution path. This design captures the influence of both observed and unobserved interactions, yielding more stable and personalized explanations. We conduct the most comprehensive fidelity evaluation to date across three models (MF, VAE, NCF), three datasets (ML1M, Yahoo! Music, Pinterest), and a suite of counterfactual metrics, including AUC-based perturbation curves and fixed-length diagnostics. SPINRec consistently outperforms all baselines, establishing a new benchmark for faithful explainability in recommendation. Code and evaluation tools are publicly available at https://github.com/DeltaLabTLV/SPINRec.

Figures (2)

• Figure 1: Illustration of Counterfactual Fidelity. SPINRec identifies items in the user's history most responsible for recommending "The Lion King". When these items are masked, the recommendation's rank drastically drops, demonstrating explanation fidelity.
• Figure 2: Fidelity (INS) vs. number of baseline samples ($\kappa$) for NCF on ML1M. Gains plateau after $\kappa{=}10$.