ANALYSIS OF PUBLIC OPINION ON TRANSPORTATION IN BANDUNG AND JAKARTA IN TWITTER USING BIDIRECTIONAL ENCODER REPRESENTATIONS FROM TRANSFORMERS MODEL

• Main Author: Pratama, Dionisius
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/74369
• Institution: Institut Teknologi Bandung

Transportation has been one of the main challenges for people living in urban areas, especially in big cities. Handling transportation problems traditionally is no longer considered suitable due to the increasingly large and complex data, which calls for an intelligent transportation system. One source of data that can be used to is social media (Twitter), in which the development of user-generated content can improve the management of existing transportation systems. In this study, IndoBERT, as a state-of-the-art language model in natural language processing tasks, is used to do sentiment analysis on Indonesian tweets about public transportation to have a better understanding of tweet context. There are three classes of sentiments used: positive, negative, and neutral. Experimental results show that IndoBERT performs better than traditional machine learning algorithms, with the best combination of hyperparameter tuning results in accuracy of 90.9% which generalizes best to the dataset. To further enhance the sentiment analysis results, a mechanism of text analysis, consisting of topic modelling with topic clustering, topic-sentiment mapping, and a search for words related to keywords is performed to widen insights provided by data. Tweet data is provided by Continuum Data Indonesia and NoLimit Indonesia with a total of 23,200 tweets.