Different Student Model #28
Description
Now that we have hit our GBL deliverable, I would like to discuss future Student Modeling.
I have a feeling that our current student model, using the knowledge component scores to advance through the tree, might not work well once we get into the incremental update sections, or, as a student masters all knowledge components before reaching the end of the tree (We only have 14 of these, and many more tree nodes).
What if we used each node as its own BKT like concept? Then, when a student completes a problem tagged as "Node 7", we update Node 7's p_learned and things will progress properly.
-(Kevin answers himself) Well, then we wouldn't really have granular(ish) understanding of the student. Like, we wouldn't be able to tell if they know how to combine unknowns or simplify addens. Furthermore, how do we tell if they actually know how to solve that type of problem?
Revised Question: What if we continued using knowledge concepts to calculate tree pre-requisites and augmented this with a p_learned for each equation type? For example, unlocking the [x=4_3] node, having no prerequisites, would function as before. However, instead of its confidence being directly computed based on simplify addens, we would add an additional parameter based on the actual times a student solved equations of this form. I imagine something like: If a student solves this type of equation optimially, with no hints, +2 updates, if they solve it in optimal_steps + 4 or less w/o using bottom out hint, +1 update, more than that, or with bottom out hint, +.5 update. Nothing else *should_ change, assuming unlocking the nodes rely on the computed confidences of their predecessors.