A novel recommender system

• Main Author: Leong, Xin Yun
• Other Authors: Liu Siyuan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/175017
• Institution: Nanyang Technological University

Conversational recommender systems (CRS) are an emerging trend in the realm of recommendation systems. It aims to deliver outstanding recommendations through dynamic and context-aware interactions. Typically, a CRS consists of two key components: a recommendation component responsible for providing users with recommendations and a conversation component in charge of generating responses to users. The effectiveness and success of a CRS lies in its ability to seamlessly engage users in conversations while extracting invaluable insights into their preferences through the dialogue history. This paper introduces a novel conversation recommender system framework. In our proposed framework, we introduce a novel three-pronged approach to reduce the semantic gap between the semantic spaces of knowledge graph and pretrained language model. Additionally, we leveraged on the success of UniCRS [12] and incorporated its prompt-based learning approach into our framework to assist pretrained language models in adapting to downstream recommendation and conversation tasks without having to finetune the parameters of pretrained language models. The proposed algorithm is trained and evaluated on the ReDial dataset. The findings presented in this paper contributes to the ongoing discourse on the evolution of recommender systems and their role in delivering personalised content within our interconnected digital landscape.