Helping conversational agents to find informative responses: query expansion methods for chatterbots

Chatterbots are conversational agents engaging a natural language-based interaction with web site users. In e-commerce, however, simple "chatting" which is mainly entertaining the user is not sufficient. Instead, the agent needs to be cooperative by trying to provide relevant information about products and/or the conditions of purchasing, usually retrieved from product databases and manuals. In this paper we explore ways to extract information out of the on-going dialogue for the automatic generation of queries to application data sources. By evaluating the result set we identify and eventually apply so called "query expansion" mechanisms for improving the quality of the results.