I would like to start my master's thesis soon. The topic is "the use of collaborative filtering for creating recommendations and predictions of learning performance".
I have a dataset that consists of the label of learning units and corresponding ratings between 0 and 1. My idea is to use a model-based collaborative system, more precisely, SVD (Singular Value Decomposition), to highlight, depending on how bad the user was, which future learning units would also perform badly. In this case, scores towards 1 are called very bad and towards 0 are called very good.
The query could be done like this, I enter the results of the 3 out of 200 ratings in my prediction function (e.g. surprise.SVD in python) and see what the prediction is for the 4th rating. And I update this every time when there are new ratings to get more and more accurate predictions.
So I wanted to ask if you recommend this way at all, or would this be rather a bad solution? What other alternatives could be considered?
I would be very happy about any answer and our help.