A Restricted Domain Question Answering System - IPS KnowledgeMaker

Abstract

Background: Question Answering System had been studied for many years. It already became a core component part for Chatbot area. The implementation of a Question Answering System can be divided into three different stage – Question Classification, Document Retrieval and Answer Extraction.

Problem: One of the difficulty of building a Question Answering System is the selection of features for training the classifier model in the answer type prediction section because it directly affect the selection of final answer.

Conducted Study: After reading a number of research studies and report of Question Answering System, the architecture for different system are similar to each other. The machine learning algorithm for training the classifier is the main focus point of this project. Therefore, the training method like Naïve Bayes algorithm is being learnt. Also, the Term Frequency – Inverse Document Frequency algorithm for ranking the passages is also being studies.

Main Result: It is successfully implement a web-based Question Answering System and perform well in typical question such as “Who”, “Where” and “When” questions. It also achieves a good accuracy on answer type prediction. It can predict the correct answer type for 63% testing questions. However, most of the time, the system extract wrong answer from the document if the prediction is failed.

Conclusion: The development in Question Answering Area is still a popular topic. There are lots of different research studies on different type of Question Answering System. In this project, it will focus on the restricted domain Question Answering system in Web-based (with the help of Search Engine). The approach has its own limitation and advantage on different type of questions.