Abstract
It is very important to extract the Twitter trends since it reflects the personal view over 645 million of its users. We examine the effectiveness of two topic modeling techniques i.e., standard Latent Dirichlet Allocation (LDA) and semantic-based Joint Multi-grain Topic-Sentiment (JMTS) in Twitter trends extraction. In addition, we also examine the frequent phrase method. Our finding reveals that JMTS significantly outperforms frequent phrase method and LDA by 54% and 24%, respectively.