Methodology

This chapter describes the construction used for our community detection technique. We first show how to build the four user-user similarity graphs from the extracted Instagram sample. Each graph has the same set of nodes, the users, but the weighted edges in each measure the similarity based on posted content, hashtag use, image use and followership respectively. These four graphs are then combined into a single graph by connecting the versions of the same user by a clique whose edge weights are derived using a random walk strategy. The resulting graph is then embedded conventionally, and communities found using DBSCAN. Topic models are then applied to the discovered communities.