Kdd2022 𞓜 unified session recommendation system based on knowledge enhancement prompt learning

Thesis link ：https://arxiv.org/abs/2206.09363

Overview of the paper ： Dialogue recommendation system （CRS） It aims to recommend high-quality products to users through interactive dialogue , It usually includes two modules: recommendation and dialogue . Existing work shares knowledge and representation between the two modules , Or design strategies to align the semantics of both . However , These methods still rely on different models to implement the two modules separately , Make it difficult for them to integrate seamlessly . This paper is based on knowledge enhanced prompt learning, unified modeling dialogue and recommendation , Specific prompts are designed to stimulate the pre training model to complete different tasks . To be specific , We have added word or entity representation of semantic fusion to the prompt , To provide relevant context and background . Besides , We also use the generated reply template as part of the recommendation task prompt , Thus, the information interaction between the two tasks is further strengthened . Experiments on two public datasets demonstrate the effectiveness of the proposed method .