Parallelization issues of domain specific question answering system on cell B.E. processors

Abstract

A question answering system is an information retrieval application which allows users to directly obtain appropriate answers to a question. In order to deal with an explosive growth of information over internet and increased number of processing stages in answer retrieval, time and processing hardware required by question answering system has increased. The need of hardware is currently served by connecting thousands of computers in cluster. But faster and less complex alternatives can be found as a multi-core processor. This paper presents a pioneer work by identifying major issues involved in porting a general question answering framework on a cell processor and their possible solutions. The work is evaluated by porting the indexing algorithm of our biomedical question answering system, INDOC (Internet Doctor) on cell processors. © 2009 Springer Berlin Heidelberg.