The intelligent tutoring system field is concerned with ways of personalizing to the student. Wang and Heffernan introduced the Student Skill model and showed that it was reliably better than the Knowledge Tracing (KT) model in predictive accuracies. One limitation of their work is that they only investigated one particular way of personalizing, which individualizes all four KT parameters simultaneously. But it may be better if we just use some of the parameters to personalize the model. More generally, we want to address the research question: What are the most important features to personalize? In this work, we systematically explored all 16 possible ways of incorporating student features into the model. We found that prior and slip are the two most important features to individualize, and the best model is the one with all four parameters individualized. Additionally, the one parameter that can be dropped without any hurt to performance is guess.
%0 Book Section
%1 citeulike:13207645
%A Gu, Junjie
%A Wang, Yutao
%A Heffernan, NeilT
%B Intelligent Tutoring Systems
%D 2014
%E Trausan-Matu, Stefan
%E Boyer, KristyElizabeth
%E Crosby, Martha
%E Panourgia, Kitty
%I Springer International Publishing
%K knowledge-tracing student-model
%P 647--648
%R 10.1007/978-3-319-07221-0_92
%T Personalizing Knowledge Tracing: Should We Individualize Slip, Guess, Prior or Learn Rate?
%U http://dx.doi.org/10.1007/978-3-319-07221-0_92
%V 8474
%X The intelligent tutoring system field is concerned with ways of personalizing to the student. Wang and Heffernan introduced the Student Skill model and showed that it was reliably better than the Knowledge Tracing (KT) model in predictive accuracies. One limitation of their work is that they only investigated one particular way of personalizing, which individualizes all four KT parameters simultaneously. But it may be better if we just use some of the parameters to personalize the model. More generally, we want to address the research question: What are the most important features to personalize? In this work, we systematically explored all 16 possible ways of incorporating student features into the model. We found that prior and slip are the two most important features to individualize, and the best model is the one with all four parameters individualized. Additionally, the one parameter that can be dropped without any hurt to performance is guess.
@incollection{citeulike:13207645,
abstract = {{The intelligent tutoring system field is concerned with ways of personalizing to the student. Wang and Heffernan introduced the Student Skill model and showed that it was reliably better than the Knowledge Tracing (KT) model in predictive accuracies. One limitation of their work is that they only investigated one particular way of personalizing, which individualizes all four KT parameters simultaneously. But it may be better if we just use some of the parameters to personalize the model. More generally, we want to address the research question: What are the most important features to personalize? In this work, we systematically explored all 16 possible ways of incorporating student features into the model. We found that prior and slip are the two most important features to individualize, and the best model is the one with all four parameters individualized. Additionally, the one parameter that can be dropped without any hurt to performance is guess.}},
added-at = {2017-11-15T17:02:25.000+0100},
author = {Gu, Junjie and Wang, Yutao and Heffernan, NeilT},
biburl = {https://www.bibsonomy.org/bibtex/2f4deade3fb0315b19364e0e4b2b51988/brusilovsky},
booktitle = {Intelligent Tutoring Systems},
citeulike-article-id = {13207645},
citeulike-linkout-0 = {http://dx.doi.org/10.1007/978-3-319-07221-0_92},
citeulike-linkout-1 = {http://link.springer.com/chapter/10.1007/978-3-319-07221-0_92},
doi = {10.1007/978-3-319-07221-0_92},
editor = {Trausan-Matu, Stefan and Boyer, KristyElizabeth and Crosby, Martha and Panourgia, Kitty},
interhash = {7e4a8b790dc1eb6f48241a326bc0f3ed},
intrahash = {f4deade3fb0315b19364e0e4b2b51988},
keywords = {knowledge-tracing student-model},
pages = {647--648},
posted-at = {2014-06-02 18:55:22},
priority = {3},
publisher = {Springer International Publishing},
series = {Lecture Notes in Computer Science},
timestamp = {2019-12-28T05:23:51.000+0100},
title = {{Personalizing Knowledge Tracing: Should We Individualize Slip, Guess, Prior or Learn Rate?}},
url = {http://dx.doi.org/10.1007/978-3-319-07221-0_92},
volume = 8474,
year = 2014
}