A comparison of two methods for the topical clustering of social media posts

While traditional search engines can often provide useful knowledge, some information that is more temporal and contextual can be found more easily or exclusively on social media. However, the number of posts on many social media platforms is too large to manage without a method for clustering posts into distinct topics. In this paper, we compare two approaches for the topical clustering of Twitter posts to discover various subjects that users are talking about. The first approach is GeoContext, our tool for clustering a social media stream into topics that uses a unique approach for calculating the similarity between posts. We compare GeoContext against Latent Dirichlet Allocation (LDA), a commonly used topic modeling algorithm. LDA has been used as a basis in several different approaches for clustering social media posts.