Natural Language Query Recommendation in Conversation Systems

Shimei Pan, James Shaw

In this paper, we address a critical problem in conversation systems: limited input interpretation capabilities. When an interpretation error occurs, users often get stuck and cannot recover due to a lack of guidance from the system. To solve this problem, we present a hybrid natural language query recommendation framework that combines natural language generation with query retrieval. When receiving a problematic user query, our system dynamically recommends valid queries that are most relevant to the current user request so that the user can revise his request accordingly. Compared with existing methods, our approach offers two main contributions: first, improving query recommendation quality by combining query generation with query retrieval; second, adapting generated recommendations dynamically so that they are syntactically and lexically consistent with the original user input. Our evaluation results demonstrate the effectiveness of this approach.