Inferring on numerical extracted answers from question answering systems

Loading...
Thumbnail Image
Date
2008-06
Authors
Nadi, Farhad
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Every day question answering systems and their undeniable usefulness become more and more obvious. There are several difficulties related to these systems that researchers all around the world are trying to solve or sometimes improve. One of these difficulties is related to the situations when the question answering systems are not able to extract a concise and salient answer to a given question. A long descriptive answer instead of a concise one can affect the efficiency of the system. One way to improve it could be by providing shorter answers from the existed answers that convey the same meaning. Sometimes applying inference on the retrieved answers for a question yields to a correct and exact answer. This research focuses on producing a concise answer from the set of retrieved answers given by any question answering system. We particularly focus on the situation when the correct answer is a numerical value. The proposed method processes the retrieved text through several modules. The proposed algorithm constructs a type of data structures called frames for numerical values that are within the answer set. Each frame that holds information about the context will pass a unification process right after a filtering process. A sentence will be produced from the frames as the concise answer to the question. The evaluation of the proposed method shows that in 60% of the cases the results are perfect.
Description
Keywords
Numerical extracted , Answering systems
Citation