Personalized Prompt Learning for Explainable Recommendation (2202.07371v2)

Abstract: Providing user-understandable explanations to justify recommendations could help users better understand the recommended items, increase the system's ease of use, and gain users' trust. A typical approach to realize it is natural language generation. However, previous works mostly adopt recurrent neural networks to meet the ends, leaving the potentially more effective pre-trained Transformer models under-explored. In fact, user and item IDs, as important identifiers in recommender systems, are inherently in different semantic space as words that pre-trained models were already trained on. Thus, how to effectively fuse IDs into such models becomes a critical issue. Inspired by recent advancement in prompt learning, we come up with two solutions: find alternative words to represent IDs (called discrete prompt learning), and directly input ID vectors to a pre-trained model (termed continuous prompt learning). In the latter case, ID vectors are randomly initialized but the model is trained in advance on large corpora, so they are actually in different learning stages. To bridge the gap, we further propose two training strategies: sequential tuning and recommendation as regularization. Extensive experiments show that our continuous prompt learning approach equipped with the training strategies consistently outperforms strong baselines on three datasets of explainable recommendation.

• The paper introduces personalized prompt learning using both discrete and continuous approaches to generate high-quality, user-specific explanations in recommendation systems.
• It employs sequential tuning and recommendation as regularization to effectively bridge the gap between continuous prompts and fully-trained Transformer models, improving metrics like BLEU and ROUGE.
• Extensive experiments on TripAdvisor, Amazon, and Yelp demonstrate significant enhancements in explainability and user trust, paving the way for more personalized AI applications.

Personalized Prompt Learning for Explainable Recommendation

The paper "Personalized Prompt Learning for Explainable Recommendation" by Lei Li, Yongfeng Zhang, and Li Chen explores the application of prompt learning in the domain of explainable recommendations using pre-trained Transformer models. The authors aim to address the challenge of integrating user and item IDs into Transformer-based models for generating natural language explanations of recommendations, which can enhance user understanding and trust in recommendation systems.

The focus of this work is on two main approaches for integrating IDs with pre-trained models: discrete prompt learning and continuous prompt learning. Discrete prompt learning converts user and item IDs into domain-specific words such as item features, which are compatible with the vocabulary of pre-trained LLMs. Continuous prompt learning, on the other hand, uses vector-based representations of IDs that are directly input into the Transformer models. This approach is advantageous as it retains more information about the IDs compared to discrete prompts.

One key innovation is the incorporation of sequential tuning and recommendation as regularization to bridge the gap between continuous prompts and fully-trained Transformer models. Sequential tuning involves a two-phase training approach that first optimizes the continuous prompt parameters with the model frozen, followed by a joint optimization phase. This technique ensures that the pre-trained model's linguistic capabilities are effectively utilized in generating explanations. Recommendation as regularization incorporates rating prediction tasks to aid in learning continuous prompts, enhancing the personalized aspect of the generated explanations.

The authors conduct extensive experiments on three datasets—TripAdvisor, Amazon, and Yelp—demonstrating the efficacy of their proposed methods. Their continuous prompt learning strategy, particularly when combined with sequential tuning, consistently outperforms state-of-the-art baselines in terms of both text quality (BLEU and ROUGE scores) and explainability (feature matching and diversity metrics). The results show that the proposed methods generate high-quality, personalized explanations that align with user and item characteristics while leveraging the rich language understanding of pre-trained models.

The implications of this research are significant for advancing the integration of pre-trained LLMs into recommender systems. The personalized explanations have practical benefits in increasing user satisfaction and trust in recommendations. The work also hints at broader applications in personalized dialogue systems and other language generation tasks needing personalization, and it sets the stage for future research into multi-modal recommendation explanations and cross-lingual systems.

Future research directions outlined in the paper include addressing potential biases in generated explanations, exploring other forms of prompts for better model interpretability, and expanding the scope to multi-modal and cross-lingual recommendation tasks. The methodological advancements presented in this work pave the way for more user-centric and explainable AI systems in the field of recommendations and beyond.