A Reciprocity-Based Method for Improving Community Detection on Directed Friendship Networks

This paper studies community detection on directed friendship networks. We propose a novel pre-processing method, Friendship Reciprocal Inclusivity-Emphasizing Never Does Backtrack Random Walk (FRIENDBRW), which is an extension of existing community detection algorithms to enhance results for sociological networks of friendship relations. This method is embarrassingly parallelizable and is motivated by the importance of information cycles and the reciprocal nature of friendships in forming communities. The algorithm can be adjusted to change how inclusive detected friend groups are, as measured by the prevalence of nonreciprocal edges. This scalable method is compared with the Louvain algorithm and the communities detected by our algorithm are more tightly-knit and realistically sized compared to empirical studies.