TY - GEN
T1 - Simplicity of Positive Reviews and Diversity of Negative Reviews in Hotel Reputation
AU - Hirokawa, Sachio
AU - Hashimoto, Kiyota
PY - 2018/7/2
Y1 - 2018/7/2
N2 - User's review on products and services is valuable information for both users and providers. The present paper conducted a polarity estimation of 73,589 reviews on hotels in Europe. Users rated one to five points for seven aspects (Value, Rooms, Location, Cleanliness, Check-in, Service, Business, Overall). In this paper, we predicted the polarity (positive/negative) of each aspect by using a machine learning method, SVM (Support Vector Machine), and feature selection, with more than 4 points being positive and less than 3 being negative. As a result, positive reviews with respect to six aspects, other than Business, were able to achieve 74% prediction performance (F-measure) with only 20 feature words. On the other hand, for negative reviews, optimal prediction performance could not be obtained unless almost all words were used, and on average F-measure was only 27%. The results indicate that positive reviews are simple, meanwhile negative reviews are diverse and hard to predict mechanically.
AB - User's review on products and services is valuable information for both users and providers. The present paper conducted a polarity estimation of 73,589 reviews on hotels in Europe. Users rated one to five points for seven aspects (Value, Rooms, Location, Cleanliness, Check-in, Service, Business, Overall). In this paper, we predicted the polarity (positive/negative) of each aspect by using a machine learning method, SVM (Support Vector Machine), and feature selection, with more than 4 points being positive and less than 3 being negative. As a result, positive reviews with respect to six aspects, other than Business, were able to achieve 74% prediction performance (F-measure) with only 20 feature words. On the other hand, for negative reviews, optimal prediction performance could not be obtained unless almost all words were used, and on average F-measure was only 27%. The results indicate that positive reviews are simple, meanwhile negative reviews are diverse and hard to predict mechanically.
UR - http://www.scopus.com/inward/record.url?scp=85065058690&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85065058690&partnerID=8YFLogxK
U2 - 10.1109/iSAI-NLP.2018.8692973
DO - 10.1109/iSAI-NLP.2018.8692973
M3 - Conference contribution
AN - SCOPUS:85065058690
T3 - 2018 International Joint Symposium on Artificial Intelligence and Natural Language Processing, iSAI-NLP 2018 - Proceedings
BT - 2018 International Joint Symposium on Artificial Intelligence and Natural Language Processing, iSAI-NLP 2018 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2018 International Joint Symposium on Artificial Intelligence and Natural Language Processing, iSAI-NLP 2018
Y2 - 15 November 2018 through 17 November 2018
ER -