Cross-graph prompt enhanced learning for personalized recommendation reason generation

Providing users with easily understandable recommendation explanations is crucial. Recently, transformer-based large language models have been increasingly used by researchers for generating recommendation explanations owing to their excellent language modeling capabilities. However, existing methods overlook high-order interactions between users and items and the semantic gap between user IDs and words, making it difficult to accurately understand users’ true preferences. Motivated by recent advances in prompt-enhanced learning, we present a Personalized Explainable Recommendation Algorithm based on Cross-Attentive Graph Propagation (CAGP). Additionally, CAGP utilizes graph structures for rating prediction to bridge the semantic gap and incorporates a large model module with personalized prompts to enhance the quality of explanations. We conducted extensive experiments on three interpretable datasets, consistently outperforming state-of-the-art baselines, proving that CAGP can generate clear, informative, and highly personalized explanations. The code for the paper will be released on GitHub after the study is accepted.