I have a question regarding a research article titles “Modeling Individualization in a Bayesian Networks Implementation of Knowledge Tracing“. I am trying to create a bayesian network for the model shown in this paper.
As per my understanding there is a parent node called
prior knowledge, which has three child nodes namely
slip rate, and
learn rate. These three nodes have a common child called
question node which has two states called ‘correct’ and ‘incorrect’, depending on whether the answer to question is correct or not.
I have another viewpoint, which relates more to the figure 1 from the article, as shown below. In this view, there are three nodes.
Student node, which is specific to a student and governs the prior knowledge parameter. A
knowledge node (K) which has two states determining the knowledge/skill is obtained or not. A
question node (Q) which again has two states, related to whether the question is answered correctly or not. Transition from
Q is governed by the guess and slip rates, i.e. even if a student has the knowledge they can slip the question (answer it wrong) and despite being no skill they may answer it correctly (guess correctly).
I am making an educational video game and have no prior practical experience with the Bayesian networks. My game has 5 levels, each level has a quiz in the end. I will ask a question in the beginning of each level to gauge their prior knowledge, so that I do not have to assign a random or same value for the prior knowledge parameter for all the students. I am planning to assign a value of 0.5 to each of the guess, slip and learn rates in the beginning. As the student answers the first question I need to re-adjust the values of the guess, slip and learn rates. I will then use these to adjust the game play to show more or better hints, and basically adjust the game to the level of the student. However, I am stuck right now and despite reading a lot I am not able to figure out how to go about this.
PS: I have made the game in unity and planning to use the infer.net framework for running Bayesian inference.
I just realized that guess rate can be expressed as
P(~K|Q) and slip rate as
P(K|~Q), but what is learn rate then, how to quantify it?
Below is a screenshot of the game menu and quiz: