| dc.description.abstract | The development of tools to improve the education quality is completely essential and
beneficial for a range of institutes or universities. Yet, almost no tool seems to automatically
and effectively categorize student’s opinions, either qualitatively or quantitatively. There have
been some scholarly works that deal with labelling the reviews by using sentimental analysis.
However, no research mentioned any possibility of automating such a system by developing
a chatbot. Practically, not many universities have a readily-available chatbot for this purpose
either. A specific example is that International University, Vietnam administers an official
Facebook page where students may express their points of view, opinions, or any ideas that
they would like to share with others and with the university about their studies or education in
general. The fanpage is quite active with 20 interactions/comments per post on average, which
means that the amount of student opinion data is enormous. It is hoped that, by applying a
chatbot to this platform, students’ opinions will be better categorized and the information
between the students and the schools may appear to be conveniently informative.
With the primary goal of developing a chatbot for International University, Vietnam
fanpage, this thesis first investigates the available literature on the subject of sentimental
analysis, then builds a chatbot using Python language. The chatbot has special features, such
as, sentimental analysis, and automatically answer questions based on machine learning
methods and string-matching techniques. The sentimental analysis feature of the chatbot will
be trained on the dataset retrieved from the “IU Confessions” Facebook page, Foody review
file. A dataset of frequent questions and answers at IU will be employed to build a set of
rules for automatically-answered questions. Some experiments are carried out to evaluate the
chatbot. As a result, answering time for each question is less than one second, the achieved
accuracy of detecting between comment and question is about 80%, and the accuracy of
sentimental prediction is about | en_US |
