Temporal Analysis of Social Concerns on African Social Media: Insights From Topics, Themes, and Sentiments(Journal Article)
Understanding social media discussions about African development is crucial for policy- makers. This study analyses trends in these discussions by (1) reviewing data collection methods, (2) extracting key topics, themes, and sentiments, and (3) applying temporal anal- ysis to observe trends and predict future trends. We compiled a dataset with 22,036 records from Twitter (X) and YouTube. Using BERTopic and Llama for topic extraction, we identified unique topics in six themes: poverty, hunger, education, employment, health, and security. Sentiment analysis was conducted with validation from human-annotated labels. By temporal analysis, we observed and predicted future trends for social concerns using Prophet model. We achieved a coherence score of 0.65 C-v for 304 topics, with 0.82 Kappa agreement. Llama2 outperformed BERTScore in theme extraction, with ”Poverty” scoring an F1 of 0.89, followed by ”Health” (0.76). Llama had a precision of 0.72 and balanced accuracy of 0.55 in senti- ment analysis Temporal analysis showed steady trends in poverty and security, with growing interest in health, education, and jobs. Positive sentiment peaked in 2020 around youth lead- ership, while governance and corruption remained stable. Findings inform policy decisions on Africa’s development, and future work will explore key social entities within posts.
Authoured by: Harriet Sibitenda , Awa Diattara, Assitan Traore, Elke Rundesteiner, Cheikh Ba
Academic units: Faculty of Science
Departments: Computer Science and Information Systems