Target-guided knowledge-aware recommendation dialogue system: An empirical investigation

The target-guided recommendation dialogue system aims to make high-quality recommendations through interactive conversations proactively and naturally. Existing methods still struggle to incorporate background knowledge for coherent response generation, and to recommend appropriate items with respect to dialogue context, user preference and recommendation target. In this paper, we investigate the problem of target-guided knowledge-aware recommendation dialogue and design a dialogue generation system to alleviate the above-mentioned issues. Specifically, we employ pre-trained language models with multi-task learning to jointly learn response generation and goal prediction towards the target. We also present a knowledge-preserving encoding strategy to maintain the facts in background knowledge. Extensive experiments on two benchmark datasets show that our system significantly outperforms various competitive models in terms of both automatic and manual evaluations. We further provide analysis and discussions to demonstrate that our system is effective in leveraging both related knowledge and planned goals to generate fluent, informative and coherent responses towards the target of recommendation.