ASPECT-BASED SENTIMENT CLASSIFICATION FOR INDONESIAN RESTAURANT REVIEW

• Main Author: SONORA - NIM : 13514019, WIEGA
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/31594
• Institution: Institut Teknologi Bandung

Some websites provide information about restaurant including reviews. Reviews submitted by visitors take form of long sentences consisting many aspects. Users need to read all the reviews to conclude restaurant’s quality based on certain aspect. Thus, in this final project we find best method to classify sentiment based on aspect for Indonesian restaurant reviews so that the number of positive reviews can be counted to represent restaurant’s quality on such aspect. <br /> <br /> <br /> <br /> <br /> Sentiment classification on this final project is done for each review per aspect. The process consists of gathering and filtering data, preprocessing, feature extraction, and aspect-sentiment classification. Ascpets used in this final project are service, price, food and beverage, ambience, and place. Classification of sentiment based on aspect is done with approach of Binary Relevance for multi-label classification thus we create 5 different classifier for all aspects. Machine learning is implemented using Support Vector Machine, Random Forest, and Logistic Regression. <br /> <br /> <br /> <br /> <br /> Performance evaluation for each method is done using 999 data reviews that are obtained from Zomato and resulting machine learning method with Logistic Regression for service, price, ambience and place with accuracy value of 0.807, 0.798, 0.780, and 0.704 respectively. For food and beverage, the selected method is by using SVM with accuracy value of 0.753. <br />