When Ghana Speaks: Using Deep Learning Approaches to Study Public Sentiment on LGBTQ+ in Ghana

Peter Annor Mensah, Regional Institute for Population Studies (RIPS), University of Ghana
Patience Serwaa Bonsu, Regional Institute for Population Studies
Emmanuel Tetteh Amponsah, Regional Institute for Population Studies
Fidelia A. A. Dake, Regional Institute For Population Studies, University Of Ghana
Donatus Yaw Atiglo, University Of Ghana
Charles Asabere, University of Ghana, Regional Institute for Population Studies
Emmanuel Olamijuwon, University of St Andrews

The discourse on criminalizing same-sex relationships and the activities of the LGBTQ+ community in Ghana has received considerable local and international attention with the passage of the Anti-LGBTQ Bill by parliament. Given the renewed interest in this discourse, vis-a-vis the insufficient research to understand public sentiment on the topic, we sought to understand the nuanced sentiments from social media regarding LGBTQ+ in Ghana. We performed a sentiment analysis of 26,302 YouTube comments extracted from videos on LGBTQ+ in Ghana. The analyses were performed on two levels: the first level employs the “syuzhet” lexicon to conduct separate analyses on public sentiment regarding the anti-LGBTQ+ Bill, and the general phenomenon of LGBTQ+ in Ghana; and the second level adopts deep learning approaches, including Deep Neural Networks, Convolutional Neural Networks, Multi-Layer Perceptron, and Long Short-Term Memory. Additionally, we used the topic modelling approach (Latent Dirichlet Allocation) to identify emerging themes from the comments. The preliminary findings from the first level analysis using the syuzhet lexicon suggest predominantly negative sentiments toward LGBTQ+ persons and rights in Ghana. The results from the topic modelling analysis reveal themes bordering on religious, political and cultural positions, and little on human rights perspective, in LGBTQ discourse in Ghana.

Keywords: Inequality, Disadvantage and Discrimination, Gender Dynamics, Computational social science methods, Big data

See extended abstract.