Analysis of Public Health on Facebook Posts in Ghana Using Topic Modelling (Preprint)

Abstract

Social Media, a pervasive communication medium, offers a vast opportunity for people of varied background to talk about common things of interest including public health issues. This communication tool is crucial for the flow of information between public-health authority and the general public for the purpose of health education. The social media platform is however, inundated with voluminous unstructured data as 1.13 billion people use Facebook daily and 90 million tweets are generated on daily basis. Part of this huge data is the over 5 million Ghanaians Facebook users who create and share very large information every day.

Thus, the purpose of this study is to discover public health related topics using Latent Dirichlet Allocation (LDA). These topics are literally hidden in huge social media data. So, we seek to meaningfully organize this shapeless Data for some insights which could optimize the operation of public health in Ghana.

The study used Natural Language Processing to achieve the stated objective. FacePager-API was used to fetch 3,546 posts from three Facebook web pages over a six-year period. The R4.0.5 programming language was used in an experiment. Topic Modelling technique was effectively utilized in the experiment.

Five public-health related topics were discovered from the 3,546 posts using Latent-Dirichlet-Allocation (LDA) algorithm. The LDA model registered an accuracy level of 99%.

The huge generated social media data can be harnessed for the optimization of public health operations in Ghana. The study concluded that the results of the study could be useful in disease surveillance and vaccination operations. It is therefore recommended that the findings could be used to strategize how to proactively expel misinformation about a pandemic.