Information-focused conversational AI Emory University at the Alexa Prize 2017 challenge
We describe an information-oriented conversational AI system, EmersonBot, developed for the Alexa Prize 2017 competition. The main goal of the system was informing users about current events, and answering their questions, while maintaining a fluent conversation. The main innovations of Emersonbot include the development of a federated multi-source information retrieval system that is aware of the conversation context; a general dialogue management system which uses machine learning to analyze the user responses, complemented with rule-based conversational logic component, specifically tailored for the Alexa Prize competition rules. This report provides a detailed description of the system, and includes extensive analysis of the contributions of the various components to user satisfaction and engagement with the system. Our results show a steady and substantial improvement in user ratings and user engagement over the period of the semi-finals as the system was refined, especially in the last weeks as entity-oriented search and recommendation features were introduced. We conclude the report by outlining promising research directions, which could improve the satisfaction of the users, while also advancing the state of the art in conversation-oriented search.