TY  - JOUR
T1  - Comparing Techniques for Sentiment Analysis in Cosmetic
Industry from Thai Reviews Videos
AU - Nuanmeesri, Sumitra AU - Kadmateekarun, Preedawon AU - Meesad, Phayung 
JO  - Journal of Engineering and Applied Sciences
VL  - 12
IS  - 2
SP  - 397
EP  - 403
PY  - 2017
DA  - 2001/08/19
SN  - 1816-949x
DO  - jeasci.2017.397.403
UR  - https://makhillpublications.co/view-article.php?doi=jeasci.2017.397.403
KW  - Sentiment analysis
KW  -thais reviews videos
KW  -naive bayes
KW  -support vector machine
KW  -natural language
AB  - The cosmetic industry has been in a very high marketing competition by advertising through various
media to promote sales and build up images of the products. In addition, consumers can access information
through a search engine finding that there are up to 45.3 billion video clips and movies of beauty online in
social network such as YouTube. Consumers are able to share messages, voices, pictures and video clips and
movies through these media swiftly. There are both content and opinions indicating “like” (Positive) or
“dislike” (Negative) on the products. These opinions can be brought to conduct the sentiment analysis on the
products. This research focuses on automatic sentiment analysis. Therefore, this study aims at the automatic
sentiment analysis which is a part of natural language processing of the cosmetic product, lipstick. The research
methodology consists of the following steps. Firstly on the collection of data in Thai language such as
criticisms on lipstick from YouTube to separate audio signals; secondly, the audio tracks conversion into texts
to cut up into words in transcription. Next, the machine learning technique consisting of Naive Bayes (NB) and
Support Vector Machine (SVM) to be used for the analysis of the consumer’s sentiment towards the lipstick.
Finally, The measurement for the efficiency and the comparative study to find result from those different
techniques. As a result, the Support Vector Machine (SVM) Technique is found to offer the best result with
the accuracy value at 85.17%.
ER  -