Here's a fun project idea. Take as input tweets, articles, etc, from various politicians and think tanks. Then generate a mapping of who is probably in a Signal chat with one another, and at what point in time.
I could imagine if the model was very good and well done, to even generate names for the chats, in a UI where clicking into it could show a graph of involvement, ideas likely shared, and approximate timelines. Perhaps clicking into the ideas could lead to details on the history / corruption of the idea, etc.
People do this already, but the signal (heh) detected by that kind of analysis isn't consumption of shared media. That's weak‚ and near the noise floor. The much stronger component is just dissemination of deliberate talking points, which happens all the time. What you need to add is the ability to discriminate the spinsters from the "real people", which is basically impossible.
I could imagine if the model was very good and well done, to even generate names for the chats, in a UI where clicking into it could show a graph of involvement, ideas likely shared, and approximate timelines. Perhaps clicking into the ideas could lead to details on the history / corruption of the idea, etc.