The Question Answering System of Indonesia's History Using Dynamic Memory Networks (DMN) Model

Abstract

The history of Indonesia which is quite long causes difficulty for our people in obtaining information about the history of Indonesia. In order to obtain information, people still need to seek from many books or documents on the history of Indonesia. Such a way is considered less efficient, thus a question answering system is considered necessary so that the information can be obtained quickly and efficiently. Questions on the topic of history have a tendency on the factoid question type so the type of question in this research is factoid. This research uses the Dynamic Memory Networks (DMN) model to obtain answers to the given questions. The parameter of the tested DMN model is learning rate, iteration, and episodes. This study uses 0.0005; 0.005; 0.05 as the value of learning rate, 1563; 3125; 6250 as the value of the number of iteration, and 3, 4, 5 as the value of the number of episodes. The dataset used in this research is 500 questions with a context in the form of single sentences and 500 questions with a context in the form of compound sentences which are taken from Wikipedia. The highest accuracy results are obtained by using the learning rate value of 0.005, iteration of 6250, and episodes of 5 on the dataset with the context in the form of single sentences amounted to 56% whereas the dataset with the context in the form of compound sentences amounted to 38.6%.