Information Retrieval in Restricted Domain for ChatterBots

This paper presents a restricted domain approach for handling information in the context of regulation search. Information Retrieval (IR) using a chatterbot as front-end is especially complex due to the interaction with Natural Language. The problem becomes more complicated when the IR is for a Restricted Domain: there is no statistical error compensation. The slang in the documentation usually does not match the common language usage, and dialogs tend to be informal, making people tend to perform incomplete questions to the system. PTAH is a chatterbot developed to interact with students, administrative employees, professors and many other people of the university community. Cultural diversity and big age differences present an interesting challenge for the Natural Language Processing area, since many approaches tend to overcome only one of the previous aspects. This paper presents the chatterbot PTAH, and specifically depicts an IR approach that intends to improve the quality of the answers. Results are evaluated with traditional precision and recall metrics, and an additional one to assess the quality of the heuristics involved in the retrieval process. Statistics indicate that it is possible to apply a specific combination of simple and traditional heuristics with good results.