Personalized Prompt Learning for Explainable Recommendation

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