COMET: Weak-tie hypothesis in complex digital networks

Mikko Laitinen, Masoud Fatemi, and Mehrdad Salimi present a method for tracing the structures of social media networks that can be applied to any social media platform. They also discuss the usefulness of a network perspective for linguistic research by examining how network structures affect the spread of recent lexical innovations.

Salimi, Fatemi, and Laitinen introduce their massive Bluesky dataset that grows in real time through continuous data streaming. The dataset is among the largest and most versatile language datasets currently available, as it includes the posts of around nine million users together with all interactional ties between them, and it continues to grow by approximately 1.5 billion words every month.

Irene Taipale presents her research on linguistic Americanization and its possible countercurrents in data drawn from Finnish social media users in relation to the language use of their social networks. Taipale analyzes the forms kind of / kinda | sort of / sorta, and toward | towards, which reflect intriguing and partly opposing paths of change in English.

More information about the upcoming presentations can be found from the abstracts below. See you in Vancouver!

Laitinen_Fatemi_Salimi_abstract_Methods_XIX

Salimi_Fatemi_Laitinen_abstract_Methods_XIX

Taipale_abstract_Methods_XIX

Masoud Fatemi, Mikko Laitinen and Pasi Fränti have been awarded the Best Paper prize at ISKE2025 conference in China! The title of the paper is “Clustering Digital Ego Networks by Tie Strength: A Scalable, Platform-independent Method”.

Abstract:

This study presents a scalable method to classify online social networks based on tie strength. Utilizing ego networks from Twitter, we applied four measurable features—interaction strength, relative interaction strength, social similarity, and outlier ratio—to cluster over 8,000 networks into four categories: weak, moderately-weak, moderately-strong, and strong ties. Our approach is not platform-dependent and overcomes the limitations of previous methods that relied on fixed thresholds or manual labeling. The results reveal regional and gender-based differences in tie strength patterns: Nordic users tend to form weaker ties, while users in Australia, the UK, and the US are more likely to build stronger-tie networks. Male users dominate across all tie categories, while female and uncategorized users are more common in weaker networks. The findings can support research in online social behavior, content delivery, and information diffusion.