Suggestion Mining and Knowledge Construction from Thai Television Program Reviews

Authors

ผศ.ดร.ปราโมทย์ ลือนาม, น.ส.กานดา แผ่วัฒนากุล

Published

Proceedings of the International MultiConference of Engineers and Computer Scientists 2013

Abstract

In today’s fast-changing business world, businesses must be able to understand customers’ thoughts more effectively and efficiently in responding to customer’s needs and expectations.

The Internet allows people to share their thoughts, opinions and suggestions regarding purchased products or services in the form of online reviews. Especially, suggestions are valuable to businesses and are not supposed to be ignored.

However, the enormous amount of information is mixed up with suggestion, facts, and opinions and unstructured text expression in the reviews make them difficult to be used by businesses.

This study examines the characteristics of suggestions and proposes a suggestion mining framework for classifying suggestion sentences in customer reviews. Two main processes were employed in this study:
(1) constructing knowledge based and
(2) classifying each suggestion in the reviews as either a’suggestion’or’non-suggestion’sentence.

The data used in the experiments are television watcher reviews obtained from several sources including the Thai Public Broadcasting Service (TPBS) website, facebook. com, and other related websites. These reviews are subjective opinions of television watchers regarding programs and broadcasting services. The data set consists of 2,561 sentences.

In conjunction with linear SVM classification, we use a knowledge based approach that is a combination of keyword selection with specific Part-of-Speech, association wordlists, and domain specific wordlists. The SVM classifier has obtained precision, recall and F-Measure of 0.83, 0.94 and 0.88, respectively. Results show that overall accuracy of the SVM classifier is satisfactory.

Suggestion Mining and Knowledge Construction from Thai Television Program Reviews. Hong Kong, the International MultiConference of Engineers and Computer Scientists 2013 (195-200).