ASPECT-BASED SENTIMENT ANALYSIS WITH CATEGORY ENHANCED WORD EMBEDDING AND DEPENDENCY PARSER FOR INDONESIAN REVIEWS

• Main Author: Iswara, Kevin
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/43550
• Institution: Institut Teknologi Bandung

Determining the right online shop can be done by reading reviews about the shop. Aspect-based sentiment analysis can be used to get public opinions from an online shop by getting conclusions from existing reviews. This final project adapts the research conducted by Wang and Liu (2015) and Cahyadi (2018) in hope to get good results on the aspect-based sentiment analysis on the online shop domain for Indonesian reviews that have been conducted by Fachrina and Widyantoro (2017) and Ilmania , et al (2018). In this final project, aspect-based sentiment analysis is divided into two modules: category classification module and sentiment polarity classification module. The first module, category classification module will use a neural network with additional input category enhanced word embeddin. The result from this module is a probability of a category in the review. The second module, sentiment polarity classification module will use a convolutional neural network with a one-vs-all strategy with additional input is word embedding with dependency parser. The result from the second module is a sentiment value of the review for the related category. The data used in this final project is obtained by crawling online shop site. The data obtained amounted to 4,187 and manually labeled based on product categories, shipping, suitability, responsiveness, price, and others. With the comparison of training and test data of eight to two, the F1-score results for the first and second modules are 0.9926 and 0.9976.