Djoko Adi Widodo, FT P Teknik Elektro (2022) Sentiment Analysis of Twitter Media for Public Reaction Identification on COVID-19 Monitoring System using Hybrid Feature Extraction Method. International Journal of Intelligent Systems and Applications in Engineering, 11 (1). pp. 92-99. ISSN 2147-6799

	PDF - Published Version Download (608kB)
	PDF - Published Version Download (2MB)
	PDF - Published Version Download (1MB)

Abstract

Several strategies were implemented to prevent COVID-19 spread. However, these steps were not effectively implemented in the community due to low public awareness and lack of discipline in daily life and this indicated a potential threat of continuous exposure to the virus. It was also observed that the pandemic greatly affected other areas besides the health sector ranging from the social, political, religious, and economic aspects to the resilience of the people. These can be observed through direct observation of the community or activities of the people on social media, especially in relation to the socio-economic aspect. Therefore, this research was conducted using social media, specifically Twitter, via the Twitter API to obtain data related to COVID-19 pandemic in Indonesia. In this research, a sentiment analysis method was developed in this research to identify public opinion related to the spread of the COVID�19 virus and its social impact on society. This was achieved using the TF-IDF and Lexical methods for feature extraction and Naive Bayes for classification. This research used a dataset obtained through Twitter using the keyword "COVID-19" in Indonesian and manually labelled using 5 categories of reactions i.e., fear, angry, love, sad, and happy. The prediction accuracy values showed that the proposed TFBS method had a higher accuracy value of 0.85 compared to other methods. The performance was also evaluated by calculating the precision, recall, and F-score values for each extraction method used, and the proposed TFBS method was observed to have the highest values while ME had the lowest.

Actions (login required)

View Item