This paper uses text analysis to measure cognitive engagement, a key indicator of cognitive presence. It's well worth a careful read. It identifies two dimensions of cognitive engagement: new information added, and relevance (ie., staying on topic). These are measured using neural networks against Prince's (1981) "taxonomy that can be used to hand-code discourse text for identifying given and new information." The utility here is that if the level of new information drops, the platform should "encourage learners to think from different perspectives," while if it drifts off topic, it should "remind learners of what is being discussed." There's a lot of new thinking here that could be developed further and I can easily imagine this sort of analysis being added to an automated content moderation tool.