A Hybrid Question Answering System to Support Physics Learning

Fong, Kwong Seng (2017) A Hybrid Question Answering System to Support Physics Learning. Masters thesis, Universiti Malaysia Sarawak.

PDF Fong Kwong ft.pdf Download (2MB)

Abstract

Question Answering (QA) systems are applications that provide the accurate answers in response to the user’s natural language questions. Many literatures have reported that the QA system is a suitable learning platform for active and unsupervised learning. The ultimate goal of this study is to propose an automated QA system using a hybrid approach, a combination of the knowledge-based approach and text-based approach, to obtain the most relevant answers to many types of Physics-related questions. For the sake of brevity, the study focuses on solving Malaysian secondary school levels of Form 4 and Form 5 Physics questions covering Electricity, Waves, Electromagnetism and Thermodynamics. In this study, different language models (unigram, bigram and trigram) and information retrieval approaches (Vector Space Model (VSM), Latent Semantics Indexing (LSI) and Latent Dirichlet Allocation (LDA)) are investigated and evaluated to construct the proposed system. LSI performs better than LDA. Besides that, VSM can help to improve the performance. In conclusion, the proposed QA system is able to perform beyond the random baseline and solve up to 53 out of 80 of the different Physics questions with Mean Reciprocal Rank (MRR) of 47.23% using trigram LSI5 + VSM. The overall process of the model is to select the top five candidate answers by using LSI in the answer extraction, and then they are rearranged by VSM in the answers rearrangement.

Actions (For repository members only: login required)

View Item