What if we collect 1000 meta posts and 1000 non-meta posts, vectorize them using TFIDF, and build a classifier eg. LogisticRegression to differentiate between meta and non-meta posts with a (hopefully) 90% accuracy?

Might actually try this out just for fun if I'm bored