TY - GEN
T1 - Aspect Category Detection on Indonesian E-commerce Mobile Application Review
AU - Nasiri, Denanir F.
AU - Budi, Indra
N1 - Funding Information:
The authors acknowledge the PIT 9 research grant NKB-0010/UN2.R3.1/HKP.05.00/2019 from Directorate Research and Community Services, Universitas Indonesia.
Publisher Copyright:
© 2019 IEEE.
PY - 2019/11
Y1 - 2019/11
N2 - E-commerce platform has a great influence on the growth of digital economy in Indonesia. This promising sector creates a fierce competition between e-commerce platforms. User reviews can be utilized to discover useful information for both developers and users. Developers use them for enhancing their application, while users take them as a consideration for using the application. We perform aspect category detection to retrieve the important aspects in reviews. We gather and analyze any aspect categories related to e-commerce and find that new potential aspects can be obtained from mobile application reviews, such as promos and payment. For identifying the aspects, we employ two different approaches, one-vs-all and single model, for 3748 annotated reviews. In one-vs-all we compare Naive Bayes, SVM, and the effect on using class-weights in SVM, since the distribution in the dataset is imbalanced. Meanwhile, we implement neural networks architecture for single model. We compare CNN to GRU+CNN architecture. GRU+CNN successfully achieves overall best result on both example-based and label-based performance metrics for multi-label task in our dataset. In example-based we use Hamming score to calculate the accuracy where GRU+CNN gets 71%, and it obtains 78% samples average F1-score for labelbased metric.
AB - E-commerce platform has a great influence on the growth of digital economy in Indonesia. This promising sector creates a fierce competition between e-commerce platforms. User reviews can be utilized to discover useful information for both developers and users. Developers use them for enhancing their application, while users take them as a consideration for using the application. We perform aspect category detection to retrieve the important aspects in reviews. We gather and analyze any aspect categories related to e-commerce and find that new potential aspects can be obtained from mobile application reviews, such as promos and payment. For identifying the aspects, we employ two different approaches, one-vs-all and single model, for 3748 annotated reviews. In one-vs-all we compare Naive Bayes, SVM, and the effect on using class-weights in SVM, since the distribution in the dataset is imbalanced. Meanwhile, we implement neural networks architecture for single model. We compare CNN to GRU+CNN architecture. GRU+CNN successfully achieves overall best result on both example-based and label-based performance metrics for multi-label task in our dataset. In example-based we use Hamming score to calculate the accuracy where GRU+CNN gets 71%, and it obtains 78% samples average F1-score for labelbased metric.
KW - application review
KW - aspect detection
KW - e-commerce
KW - Indonesian review
KW - multi-label
UR - http://www.scopus.com/inward/record.url?scp=85085568478&partnerID=8YFLogxK
U2 - 10.1109/ICoDSE48700.2019.9092619
DO - 10.1109/ICoDSE48700.2019.9092619
M3 - Conference contribution
AN - SCOPUS:85085568478
T3 - Proceedings of 2019 International Conference on Data and Software Engineering, ICoDSE 2019
BT - Proceedings of 2019 International Conference on Data and Software Engineering, ICoDSE 2019
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2019 International Conference on Data and Software Engineering, ICoDSE 2019
Y2 - 13 November 2019 through 14 November 2019
ER -