In recent years, Twitter has become a highly popular form of social media.Twitter provides a platform for users to post short messages for followers to read inan on-or off-line fashion. Twitter is used in a variety of ways, from posting aboutpersonal daily life, to keeping up to date with current events.This thesis aims to find a reliable pipeline to analyse and visualize hottest topics(or trends) that people are talking about on Twitter during a period of time. Topicmodel is used to cluster Twitter messages and identify topic words, then topic wordscombined with the tweets’ influences are graphically represented by visualizationsoftware to reflect the trend under the topic. However, two limitations of Twittermessages prevent normal topic model tools from being applied their full potentials:Twitter messages are short and and colloquial. Twitter message provides little usefulinformation for the topic model to work properly. Thus, we proposed an poolingschema to enhance the performance of a topic model on Twitter data. Meanwhile, toidentify a reliable pipeline to do the task, we compared different methodologiesduring the process. We compared performance with and without pooling schema inthe data sampling step, performance with and without TF*IDF in the data processingstep; and finally compare performance of Latent Dirichlet allocation (LDA) withCorrelated Topic Models (CTM) to identify a topic. The results show thatLDA-TF*IDF with pooling schema is the most accurate model to identify Twittertrend.