Analyzing Stance and Topic of E-Cigarette Conversations on Twitter: Case Study in Indonesia

摘要

To control the use of e-cigarette, Indonesia plan to establish a regulation that embodies all the concerns, sentiments, and opinions of public This study aims to identify public opinions in social media Twitter by classifying tweets into group in favor or against e-cigarette and explore dominant topics of each group. This research obtained 15,373 tweets between June 2019 – May 2020 that is classified into 4 labels: Against, Favor, Neutral, and Irrelevant. The best model was selected with specification: 3 features (Count, Unigram, and Bigram), Logistic Regression algorithm, and three-stage classification pipeline ($mathrm{F}1-ext{score}=0.807$). As for topic modelling, corpus Against and Favor are used to retrieve dominant topics. We chose Non-negative Matrix Factorization algorithm with $mathrm{k}=6$ and achieve high coherence scores, which are 0.962004 for corpus Against and 0.999736 for corpus Favor.