PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Contextualized question answering
Luka Bradesko, Lorand Dali, Blaz Fortuna, Dunja Mladenić, Inna Novalija and Bostjan Pajntar
In: 32nd International Conference on Information Technology Interfaces, 21 - 24 June 2010, Cavtat, Dubrovnik, Croatia.

Abstract

he paper describes a system which enables accurate and easy-to-use contextualized question answering and it provides document overview functionalities. The possibility of asking natural language questions enables a friendly interaction for the user.The contextualization is achieved by using an ontology. The answers are provided based on a domain specific document collection of choice. The approach consists of several phases as follows: data preparation, data enhancement, data indexing and handling questions. Every module uses state of the art technologies that are shown to work in a complex pipeline to make available question answering on top of a given document repository with the context of ontologies, such as Cyc, ASFA and WordNet. The functioning of the proposed approach is demonstrated on English document collections on Aquatic Sciences and Fisheries — ASFA, using Cyc ontology, ASFA thesaurus as domain specific ontology and WordNet as general ontology. Experimental evaluation has shown that the usage of ontologies increases the number of answers retrieved by about 60%. However, the number of answers that are actually correct increases by only 40% when using ontologies.

EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
Learning/Statistics & Optimisation
Information Retrieval & Textual Information Access
ID Code:7485
Deposited By:Jan Rupnik
Deposited On:17 March 2011