This article combines the study of large-scale social media data with social network theory in sociolinguistics. Given that the purpose of social media is to form networks and communities, big data from social media applications could have substantial potential in deepening the understanding of the role of networks in language variation and change. The study first presents an algorithmic method suitable for directed-graph ego networks in computer-mediated communication. This method measures network strength and enables us to enrich social media data with a network parameter that indexes how strongly (or loosely) people in a network are connected to each other. We then use a large dataset of c. 4.8 billion words from nearly four thousand networks to study how network strength conditions linguistic change. The results show that online networks are highly similar to traditional offline networks, a finding that enables fixing a major methodological limitation in the study of weak ties, namely that the method is less suited for studying socially and geographically mobile individuals. This finding makes it possible to apply the theory of social networks in sociolinguistics to very large digital networks in social media.