engleski

Predicting Dependency of Approval Rating Change from Twitter Activity and Sentiment Analysis

In recent years, multi-agent systems have been augmented with alternative social network data like Twitter for preforming target inference. To this extent, public figures tracked by agent system usually have key social value in social networks. This social value can additionally drift in public approval based on social network communication. To test the connection in public approval and social communication, we perform analysis of presidential Twitter account activity during the two-year period 2017–2018. Sentiment analysis was used on the processed tweets in order to test such a data set’s predictability in gaining an insight into a 7, 14 and 21 days (1, 2 and 3 weeks) significant presidential job approval rating change. To this extent, five different supervised machine learning algorithms are used: Random Forest, Xgboost, AdaBoost, AdaBag and ExtraTrees. Results indicate that voter approval rating has slight future predictability based on Twitter activity and emotional sentiment analysis possibly indicating consistency with the human nature of positive news and outcomes resonating with people for a much shorter period than negative ones.

Twitter ; NRC sentiment ; Emotional sentiment ; Supervised learning ; Social networks ; Approval rating

nije evidentirano