Tweet Analytics for Political Position Estimation

Abstract

An increasingly popular method of predicting trends and forecasting voting outcomes is to create a prediction model based on alignment of publicly available social media content produced by voters with voting behaviours. This paper aims to analyse whether Twitter data extracted from local Dublin City Council members’ Twitter accounts in comparison with corresponding Councillor and Motion data from the Council- Tracker.ie website can be interpreted and used to create a model to predict the voting outcome of local Dublin City Council votes to pass motions. The aim was to explore whether through utilising machine learning techniques along with natural language processing techniques, reliable and data driven predictions can be generated for policy-making proposals brought forward. The acquired experimental results suggest that the approach used was marginally adequate in supporting the proposed hypothesis, although some interesting results were derived. Of the models analysed the Decision Tree model produced the most accurate results with an accuracy score of 0.71 (baseline: 0.63). Analysis of the models and an ablation study showed that the features derived from tweet texts and motion texts along with overall properties of a Councillor’s twitter account were the most powerful indicators. The behaviour of a tweet, such as its acquired number of favorites or retweets, were not indicative of the results in both the random forest model and decision tree model