Question answering system for the travel domain

Abstract

A Question Answering (QA) system backed by a comprehensive and up-to-date knowledge base would be appropriate for travellers to satisfy their information needs. In this paper, a complete QA system is presented. It has two main phases: question identification (Expected Answer Type (EAT) identification) and searching the knowledge base (KB) to find the answer to the classified question. In QA systems, identification of the EAT of a question imposes some constraints when determining the possible answer. This paper presents the first study on semantic classification of questions into EATs in the travel domain. A new two-level taxonomy for the travel domain is introduced, along with a dataset annotated with the same. A machine learning approach is used for question identification, which gives very promising results even with the use of syntactic and semantic features. A rule-based approach is used for searching the KB to find the answer. An ontology serves as the KB of the QA system which is traversed using a Simple Protocol and RDF Query Language (SPARQL) query generated through the rule-based approach.