On the dynamic prediction of a president's net satisfaction survey: A test of Twitter accuracy

Abstract

Twitter as a social media platform, contains data that are considered noisy, unstructured, and dynamic in nature. Different studies have been conducted with the use of Twitter in forecasting outcomes such as Box Office, Stock Market, and Electoral Outcomes. In the Philippines, one prominent topic that is always in the news is President Rodrigo Roa Duterte due to his different perspective in serving the country. Surveys that count for the President's net satisfaction are conducted by different organizations across the Philippines, an example of which is the Social Weather Stations (SWS) which conducts quarterly surveys in different social classes in the country. This research aims to test Twitter posts (tweets) as an alternative basis in predicting SWS's result of the president's net satisfaction survey using the Partial Least Squares Regression. To determine whether Twitter sentiment on the President's satisfaction performance can be a basis for predicting outcomes of SWS the dynamics of its data is examined through a chaos analysis using the largest Lyapunov exponent (LLE). The LLE test therefore is used to determine whether the sentiment data obtained from Twitter possess a chaotic or nonchaotic behavior. Chaos test showed positive results for the Twitter variables except for the 2nd quarter of the Number of Negative Tweets, thus determining that for the whole year of 2017, all variables show a chaotic behavior. LLEs in the last quarter of 2017 tend to have a higher exponent during the month of November when the ASEAN summit was held in the Philippines. Evidently, the sentiment Twitter data possess a chaotic behavior as identified by positive LLEs. Partial Least Squares was able to accommodate the dynamic nature of the data and was able to create a forecasting model for the President Net Satisfaction Survey of SWS.