Tamburrini et al. Twitter users change word usage according to conversation-partner social identity. Soc Net. 40 (2015) 84-89

Summary of
Twitter users change word usage according to conversation-partner social identity
Tamburrini et al. Soc Net. 40 (2015) 84-89. Full text here (unfortunately it is not free, yet).

Methods
Snowball sampling for determining which Twitter users' data to download, capturing messages with an '@' sign, discarding retweets. 
Modularity maximization algorithm for graph partitioning (community detection)
Text similarity measures: Euclidean and Jaccard
Dataset of messages split into inter- and intra-community groups, then balanced by randomly discarding those from the larger set until both sets had the same cardinality.
Bootstrap sampling with set union

Results
The basic result is demonstrating that users change their word usage according to audience, and the community of the audience; specifically, the linguistic patterns used for inter-community messages differs from those of intra-community messages. This may seem intuitive or obvious to anyone who associates with multiple social circles--e.g., I talk differently to my grandparents than I do to my friends, and ditto for co-workers. Further, they show that the degree of isolation of a community is proportional to the magnitude of linguistic difference between speakers.
The paper addresses the issue of social identity, and the results add evidence to the two well-known social-psychological theories of Social Accommodation and Convergence. The former states that a speaker will alter their word usage to suite that of the audience, while the latter states that more isolated communities will differ more from the rest of the population.