Cross-Sample Community Detection and Sentiment Analysis: South African Twitter

This study investigates the persistent communities on South African Twitter across 24 datasets that have been collected since 2014. It also analyses the sentiment of the identified communities towards each other, as well as the sentiment the community shares with itself. To perform this analysis, 24 datasets were aggregated, cleaned and an iterative approach to community detection was used to accurately map the South African communities. The procedure identified 18 communities, 15 of which were found to be persistent across all datasets. The overall sentiment calculated across the dataset resulted in 16.1% tweets classified as neutral, 41.6% classified as positive and 40.3% of tweets classified as negative. Sentiment was aggregated to community level to investigate the polarity of interactions between these communities.