IMPROVING DIALOGUE GENERATION VIA PROACTIVELY QUERYING GROUNDED KNOWLEDGE

Recent advances in pre-trained language models have significantly improved neural response generation. Further, an intelligent dialogue system should be able to give accurate responses that meet the needs of users. However, using appropriate knowledge has so far been proved difficult, because faced with a mass of relevant knowledge, the model not only needs to accurately retrieve the target information, but also integrate the information into dialogue response. In this paper, we propose a novel knowledge-based dialogue system which integrates the strength of a transformer-based generator and a knowledge retriever capable of proactively constructing queries for accurate information. Experiments show that conversational systems that leverage knowledge integrator could generate more informative and human-like responses than strong baseline systems.