Knowledge Prompt-tuning for Sequential Recommendation
This work addresses the challenge of enhancing recommendation accuracy for users in domains like books, music, and movies by mitigating information loss and noise from knowledge integration, though it is incremental in combining existing techniques.
The paper tackles the problem of sequential recommendation by integrating both general and domain knowledge to capture fine-grained user preferences, resulting in improvements of up to 40.65% in NDCG@5 and 36.42% in HR@5 on benchmark datasets compared to PLM-based methods.
Pre-trained language models (PLMs) have demonstrated strong performance in sequential recommendation (SR), which are utilized to extract general knowledge. However, existing methods still lack domain knowledge and struggle to capture users' fine-grained preferences. Meanwhile, many traditional SR methods improve this issue by integrating side information while suffering from information loss. To summarize, we believe that a good recommendation system should utilize both general and domain knowledge simultaneously. Therefore, we introduce an external knowledge base and propose Knowledge Prompt-tuning for Sequential Recommendation (\textbf{KP4SR}). Specifically, we construct a set of relationship templates and transform a structured knowledge graph (KG) into knowledge prompts to solve the problem of the semantic gap. However, knowledge prompts disrupt the original data structure and introduce a significant amount of noise. We further construct a knowledge tree and propose a knowledge tree mask, which restores the data structure in a mask matrix form, thus mitigating the noise problem. We evaluate KP4SR on three real-world datasets, and experimental results show that our approach outperforms state-of-the-art methods on multiple evaluation metrics. Specifically, compared with PLM-based methods, our method improves NDCG@5 and HR@5 by \textcolor{red}{40.65\%} and \textcolor{red}{36.42\%} on the books dataset, \textcolor{red}{11.17\%} and \textcolor{red}{11.47\%} on the music dataset, and \textcolor{red}{22.17\%} and \textcolor{red}{19.14\%} on the movies dataset, respectively. Our code is publicly available at the link: \href{https://github.com/zhaijianyang/KP4SR}{\textcolor{blue}{https://github.com/zhaijianyang/KP4SR}.}