Beyond Factual Correctness: Mitigating Preference-Inconsistent Explanations in Explainable Recommendation
arXiv:2603.03080v1 Announce Type: new Abstract: LLM-based explainable recommenders can produce fluent explanations that are factually correct, yet still justify items using attributes that conflict with a user's historical preferences. Such preference-inconsistent explanations yield logically valid but unconvincing reasoning and are largely missed by standard hallucination or faithfulness metrics. We formalize this failure mode and propose PURE, a preference-aware reasoning framework following a select-then-generate paradigm. Instead of only improving generation, PURE intervenes in evidence selection, it selects a compact set of multi-hop item-centric reasoning paths that are both factually grounded and aligned with user preference structure, guided by user intent, specificity, and diversity to suppress generic, weakly personalized evidence. The selected evidence is then injected into LLM generation via structure-aware prompting that preserves relational constraints. To measure prefer
arXiv:2603.03080v1 Announce Type: new Abstract: LLM-based explainable recommenders can produce fluent explanations that are factually correct, yet still justify items using attributes that conflict with a user's historical preferences. Such preference-inconsistent explanations yield logically valid but unconvincing reasoning and are largely missed by standard hallucination or faithfulness metrics. We formalize this failure mode and propose PURE, a preference-aware reasoning framework following a select-then-generate paradigm. Instead of only improving generation, PURE intervenes in evidence selection, it selects a compact set of multi-hop item-centric reasoning paths that are both factually grounded and aligned with user preference structure, guided by user intent, specificity, and diversity to suppress generic, weakly personalized evidence. The selected evidence is then injected into LLM generation via structure-aware prompting that preserves relational constraints. To measure preference inconsistency, we introduce a feature-level, user-centric evaluation metric that reveals misalignment overlooked by factuality-based measures. Experiments on three real-world datasets show that PURE consistently reduces preference-inconsistent explanations and factual hallucinations while maintaining competitive recommendation accuracy, explanation quality, and inference efficiency. These results highlight that trustworthy explanations require not only factual correctness but also justification aligned with user preferences.
Executive Summary
This article presents PURE, a novel preference-aware reasoning framework for explainable recommendation systems. PURE addresses the issue of preference-inconsistent explanations in LLM-based recommenders by intervening in evidence selection and generation. By selecting a compact set of item-centric reasoning paths aligned with user preferences, PURE suppresses generic and weakly personalized evidence. The framework is evaluated on three real-world datasets, demonstrating improved preference consistency and factual correctness while maintaining competitive recommendation accuracy and explanation quality. The article highlights the importance of justification alignment with user preferences for trustworthy explanations. The proposed framework and evaluation metric have significant implications for the development of more transparent and user-centric recommendation systems.
Key Points
- ▸ PURE addresses preference-inconsistent explanations in LLM-based recommenders.
- ▸ The framework intervenes in evidence selection and generation to align with user preferences.
- ▸ PURE is evaluated on three real-world datasets with improved preference consistency and factual correctness.
Merits
Strength in addressing preference-inconsistent explanations
PURE effectively addresses a significant issue in LLM-based recommenders, providing a novel approach to improving explanation quality and alignment with user preferences.
Improved factual correctness and recommendation accuracy
The framework maintains competitive recommendation accuracy and explanation quality while reducing factual hallucinations and preference-inconsistent explanations.
Demerits
Limitation in scalability
The framework's performance may be impacted by the complexity of the evidence selection process, potentially affecting scalability for large datasets.
Dependence on user intent and specificity
PURE relies on accurate user intent and specificity, which may be challenging to obtain, particularly in cases where user preferences are ambiguous or unknown.
Expert Commentary
The article presents a significant contribution to the field of explainable recommendation systems, addressing a critical issue in LLM-based recommenders. PURE's focus on preference-aware reasoning and evidence selection is a crucial step towards developing more transparent and user-centric recommendation systems. While the framework has demonstrated promising results, further research is needed to address potential scalability and dependency issues. The article's implications for policy and regulatory frameworks are significant, highlighting the need for a more user-centric approach to AI system design.
Recommendations
- ✓ Future research should focus on addressing scalability and dependency issues in PURE, as well as exploring applications in high-stakes domains.
- ✓ Policy and regulatory frameworks should prioritize user-centric design and transparency in AI systems, incorporating the principles outlined in this article.