Identifying Policy Agenda Sub-Topics in Political Tweets based on Community Detection

The explosive use of twitter in the political landscape presents new avenues for tracking political conversations at federal and state level. Tweets are used by state and federal government bodies to present citizens with information about future and present policies. It is also used by political candidates to express their views on policy changes, laws and to campaign for legislative body elections, the most recent example being the 2016 US presidential elections. In this paper, we use supervised learning, textual semantic similarity and community detection techniques to find actively discussed policy agenda sub-topics among political tweets within a certain time period. Specifically, we target tweets pertaining to major policy agendas published by state representatives in US, to try and discern the major policy sub-topics that they address using their twitter accounts. Using our method, we demonstrate how we achieve a high accuracy in terms of Topic Recall and Order Recall, by comparing the output of our proposed method with sub-topic annotations done by domain experts.