An Analysis of Interactions Within and Between Extreme Right Communities in Social Media

This article studied a selection of right-wing extremist (RWE) groups on Twitter. The authors looked at particular language-based networks as case studies, collecting Twitter data for groups across eight countries. The authors identified international linkages between groups with interactions, being defined by accounts that referenced and/or reciprocally mentioned each other. The results were unweighted and generated two network representations of the accounts from the eight country sets used for analysis. The identified communities of RWE accounts were analyzed within local interaction networks where locality was considered in terms of nationality and linguistic proximity. Matrix factorization techniques were found to be most suitable for topic analysis of the particular data sets and mapping was generated between the detected communities and their associated topics. The authors present an analysis of the activity of selected RWE groups using network representations based on reciprocal follower and interaction activity, in addition to topic analysis of their corresponding tweets.

This article will be of particular use to researchers and practitioners who aim to identify international relationships between groups across varying geopolitical boundaries. A variant of the OSLOM algorithm from the work of Lancichinetti and Fortunato was used to generate a data set, identifying stable “consensus communities”. These communities were then ranked by stability. The authors create undirected interaction networks from their expanded language specific data sets and used reciprocal edges (filtered <5 to bring out the strongest linkages). In order to describe the communities, the tweet content was analyzed for the purpose of generating interpretable descriptions for the detected communities and to discover latent topics associated with interaction between the member accounts. Their outputs included a “profile document” for each node consisting of an aggregation of their corresponding tweets, from which a tokenized representation was produced. Although a certain awareness exists between accounts based on follower relationships, the study found that mentions, retweets, and interactions indicate stronger relationships in cases where linguistic and geographical proximity are highly influential.