An effective approach for relevant paragraph retrieval in Question Answering systems

Abstract

Paragraph retrieval is a substantial task in Question Answering (QA) systems. It represents the extraction of data from a huge data collection, which have the probability to contain an answer to a question and is a significantly important intermediary step between a user of a QA system and the answers. It is barely impossible to analyse a large collection of data extensively in quick time and because of the vast nature of the background data, it is very important to narrow down the search space from where an answer can be looked for. In this paper, we address the information extraction step and present an effectively designed model for the relevant paragraph retrieval, which is efficient in terms of execution time as well. The model deals with the structure and the organization of information, performs an in-depth analysis of user's question, and presents a priority based searching capability for retrieving paragraphs according to the necessity which will be both effective and time efficient. Experiments were carried out and compared against similar systems based on the data of the document retrieval task of Text REtrieval Conference (TREC) 2005. We also tested our methodology against the data set from TREC 2007. The system performance was measured in terms of various parameters such as R-precision, Recall, and Mean Average Precision. A satisfactory result achieved by our approach establishes its competency for getting integrated into any realtime QA systems.