XRec: Large Language Models for Explainable Recommendation (2406.02377v2)

Abstract: Recommender systems help users navigate information overload by providing personalized recommendations aligned with their preferences. Collaborative Filtering (CF) is a widely adopted approach, but while advanced techniques like graph neural networks (GNNs) and self-supervised learning (SSL) have enhanced CF models for better user representations, they often lack the ability to provide explanations for the recommended items. Explainable recommendations aim to address this gap by offering transparency and insights into the recommendation decision-making process, enhancing users' understanding. This work leverages the language capabilities of LLMs to push the boundaries of explainable recommender systems. We introduce a model-agnostic framework called XRec, which enables LLMs to provide comprehensive explanations for user behaviors in recommender systems. By integrating collaborative signals and designing a lightweight collaborative adaptor, the framework empowers LLMs to understand complex patterns in user-item interactions and gain a deeper understanding of user preferences. Our extensive experiments demonstrate the effectiveness of XRec, showcasing its ability to generate comprehensive and meaningful explanations that outperform baseline approaches in explainable recommender systems. We open-source our model implementation at https://github.com/HKUDS/XRec.

An Overview of the XRec Framework for Explainable Recommendation

The paper presents a framework named XRec, which focuses on enhancing the explainability of recommender systems by leveraging the language capabilities of LLMs. The research emphasizes the importance of providing clear, textual explanations for recommendations, addressing a critical gap in traditional recommender systems that often lack transparency.

Recommender systems, particularly those based on Collaborative Filtering (CF) methodologies, face challenges in explaining the rationale behind suggesting specific items to users. While advancements in deep learning, such as Graph Neural Networks (GNNs) and Self-Supervised Learning (SSL), have improved the representational quality of such systems, they rarely incorporate mechanisms for explainability. XRec seeks to integrate the high-level understanding capabilities of LLMs with CF methodologies to generate meaningful explanations.

Methodology

XRec is designed as a model-agnostic framework that creates a synergy between graph-based collaborative filtering and LLMs. It incorporates several key components:

1. Collaborative Relation Tokenizer: This uses GNNs to convert user-item interactions into latent embeddings reflecting high-order collaborative signals. A LightGCN architecture is employed to transmit these interactions through message passing mechanisms.
2. Collaborative Information Adapter: Given the distinct semantic domains represented by user-item interactions and text-based semantics, a Mixture of Experts (MoE) model adapts and aligns these embeddings for injection into LLMs. This adapter ensures that the embeddings are seamlessly integrated into the LLM structure.
3. Unifying CF with LLM: LLMs are trained with prompts that include special tokens reserved for the adapted collaborative embeddings. This methodology allows LLMs to draw on both interaction patterns and natural language context, facilitating robust scenario adaptation, including zero-shot recommendations.

Experimental Results

The authors assess XRec's capabilities using datasets from Amazon, Yelp, and Google Reviews. Results highlight the framework's effectiveness in producing unique, accurate explanations with notable improvements over baselines such as Att2Seq and PEPLER.

• Metrics related to semantic similarity (BERTScore, GPTScore) demonstrate enhanced performance of XRec in generating coherent explanations.
• XRec's Unique Sentence Ratio (USR) of nearly 1 indicates highly personalized outputs, a novel achievement in the field of explainable recommendation systems.

Additionally, the research involves an exhaustive evaluation of the system's resilience against data sparsity. Tests show that XRec maintains its capability to deliver personalized and contextually appropriate explanations even when dealing with novel users items, thus addressing common pitfalls such as the cold-start problem.

Implications and Future Directions

XRec's ability to link collaborative filtering insights with the interpretive strength of LLMs opens new pathways for developing user-centric, transparent recommender systems. The approach does not just optimize recommendations but also aligns them with user understanding, potentially increasing user trust and system acceptability.

Future developments might focus on integrating multimodal data inputs to enrich the recommendation context further. While XRec relies predominantly on textual and graph data, incorporating visual data could yield more holistic user understanding and satisfaction in recommendations.

Overall, this framework provides a foundational model for marrying the computational rigor of collaborative filtering with the narrative clarity of LLMs, setting the stage for a new standard in transparent, explainable AI systems.