Automated Answer Extraction for Reading Comprehension System Based on Matching Approach

Abstract

One of Reading Comprehension (RC) tasks is inspired by the Information Extraction (IE) application to extract a set of features from a natural language text. Since reading comprehension tests were created to judge the reading ability of humans, there are challenges in using language understanding systems to extract information from comprehension stories. The questions and answer keys already exist in the story. The challenge how use the language understanding system to find automatic an answer for questions. The main target in this study is to review the matching approach of Natural Language Processing techniques to extract information from reading comprehension; the information would be able to answer the WH questions of reading comprehension texts. The matching approach decomposed the story sentences and questions into a container of words that were augmented with additional automated linguistic processing and then the answering engine stage is applied to the matching process after representing the information into a bag of words. Because the answer to a question must come from the given document, the story structure has to be examined in the context of responding to test questions. On WH questions, the experiment tested 15 children’s stories that contained 262 sentences (with an average of 18 sentences per story) and 75 WH questions. The result achieved was 67.3% Hum sent accuracy of the correct answer on the questions pertaining to the children’s stories.