Authors: Ruowei Liu*, University of Georgia, Xiaobai Yao, University of Georgia
Topics: Geographic Information Science and Systems, Spatial Analysis & Modeling, Quantitative Methods
Keywords: Presidential Election, Prediction, Sentiment Analysis, Twitter, Political Science
Session Type: Paper
Start / End Time: 2:35 PM / 4:15 PM
Room: Marriott Ballroom Salon 1, Marriott, Lobby Level
Presentation File: No File Uploaded
Predicting presidential election has recently been an attractive topic that was discussed in many disciplines (such as political science, computer science, etc.). Traditionally, political scientists have proposed several prediction models for forecasting presidential elections. These models mainly treat popularity or preference of the two candidates in the poll survey and the economic growth at the national level as the predictive factors of the presidential election. However, spatially or temporally dense polling has always been expensive. In the recent decade, the exponential growth of social media has drawn attention of researchers from various disciplines. Existing studies showed that online twitter information has the potential to reflect and mirror the offline political landscape. Twitter sentiment has been extensively analyzed to predict the elections around the world. However, most of the studies correlated twitter sentiment directly and solely with the election results or just polling preference which cannot be regarded as a prediction. This study uses Twitter data to replace polling data and develops an innovative approach to predict the 2016 U.S. presidential election at the county level in Georgia. First, we collected Twitter data from September 26th to November 8th by keywords related to the election. Then we analyze the sentiment score for each tweet using Stanford CoreNLP. Thirdly we establish the relationship between the predicted voting results and twitter sentiment as well as the economic growth based on the trial-heat model. Finally, by comparing the predicted voting results with the actual one in Georgia, we evaluate the accuracy of our model.