Guiding Users by Dynamically Generating Questions in a Chatbot System
Abstract
Chatbots can efficiently support users in finding relevant answers in complex information domains. Chatbot aggregate data from different sources and provide information in an interactive dialog. In a conversation, chatbots mime a human expert trying to provide information in well-consumable pieces and try to guide users towards predicted information needs. One challenge for chatbots consists in generating questions if a user input is ambiguous or incomplete. Computing good counter-questions requires an understanding of the user's intentions and a good structuring of the data the chatbot uses of providing to needed information. In this work we present a solution for generating clarification-questions based on dynamic data collections applying semantic clustering and flexible questions trees, We optimize and evaluate our approach for a chatbot tailored for answering questions related to services offered by the Local Public Administration. We show that our approach efficiently helps users to find the relevant information in a natural conversation avoiding long lists for potentially interesting search results. The approach is based on an data enrichment and knowledge extraction pipeline that enables the adaptation of the components to different knowledge sources and the specific requirements of new domains.