Extend the Knowledge Tracing Framework using Partial Credit as Performance

In an ITS, students typically have two types of performance to a problem: correct and incorrect, and all the other information such as how many hints the student sees in this question and how many attempts he/she does to get the correct answer is ignored. Feng and Heffernen (2010) showed that we can predict better by accounting for problem solving behavior as well as correctness. By introduce continuous performance nodes into knowledge tracing model, we are able to represent partial credit – which is computed using all the information in a student’s response to a question. In this paper, we present the algorithm to compute partial credit, as well as the modified Knowledge Tracing (KT) model using partial credit as performance. We compared the two KT model with different types of performance node. The result shows that partial credit reliably improves the knowledge tracing model in predicting both the partial credit performance and the binary performance.