A Review of Knowledge-Grounded Dialogue Systems

With the rapid development of computer science and natural language processing technologies, the development of dialogue systems attracts researchers' wide interest, and how to build natural and engaging conversations becomes an urgent problem in great demand. Traditional dialogue systems that solely rely on dialogue context have exposed a series of severe shortages such as tend to generate numerously uninformative responses and meaningless replies. To this end, the principle of knowledge-grounded dialogue systems is proposed, which provides new views on the evolution and research of dialogue systems. However, it's challenging for knowledge-grounded dialogue systems to select appropriate and accurate knowledge information from the knowledge base and integrate the selected knowledge information and the dialogue history into the responses correctly. This paper introduces the definition and development of knowledge-grounded dialogue systems, lists several common datasets, summarizes current main methods and model structures based on the division of the main structures of knowledge-grounded dialogue systems, finally discusses the shortcomings of previous work, and provides views on future research trends.