Another piece of work I spent a good deal of time on over the past year at TRSS — a paper on community detection of hashtags on networks: Automated Hashtag Hierarchy Generation Using Community Detection and the Shannon Diversity Index, with Applications to Twitter and Parler.
The main takeaway is that we introduced some novel metadata-based entropy derivations to associate with hashtags, in order to group hashtags into like communities. Essentially, we give each hashtag several entropy scores along different axes of information — other-hashtag co-occurrence, time of day of posting, number of users posting about it, and so on.
My major contributions to this paper are the word entropy metric, the community detection quality heuristic for comparing community detection algorithms, and collecting, cleaning and preparing the Parler data.
Sibling projects from the same stretch at TRSS: the xView3 challenge and the anomix package.