Serious suggestion: Train a text classifier on texts from the Cultural Revolution and similar historical episodes. When you start to get matches from things written in your own time, your society is in trouble.
-
I’m guessing the most useful training texts would be in Chinese — any thoughts on working with translations? ML models are likely to loss accuracy based on low-quality translation. At first glance, this might be the bottleneck for this idea.
I assumed the project would use translations, unless of course it was comparing the same society at two different times. I don't know if it would work. All I do know is that text classifiers can pick up more than you'd expect.
