transfer 23(1) » Rezeptions- und Wirkungsforschung

How emotional reactions to topic fatigue manifest themselves in social media

A text mining approach

This explorative research investigates emotional reactions to topic fatigue in Facebook over April-May 2017 in the UK. 154.106 comments were collected, of which 1.963 were manually labeled with 7 emotions. A word-level supervised machine learning technique was applied using three classifier algorithms (i.e., Naïve Bayes, Random Forest and SVM) to predict emotional reactions to topic fatigue compared to a non-fatigue topic. TF-IDF was used as weighting schema to extract the keywords.

Findings indicate that the share of comments with embarrassment, fear, frustration and sarcasm are higher for topic fatigue, while sadness and anger are higher for the non-fatigue topic. Over time, the reactions to topic fatigue appear constant while in the non-topic fatigue topic emotion levels tend to fluctuate.