Algorithms and Applications of Learned Student Models for Adaptive Learning Systems

点击翻译按钮获取中文摘要

In the last few years, a number of innovative applications have emerged to guide a user through a study process that can lead to an adequate preparation for an exam, a diploma, a new job, or simply a self-paced, self-motivated learning agenda. Such applications are designed to guide the user into a large body of learning material, according to specific learning objectives, and to the user's personal needs. They take full advantage of the one-on-one, personalized tutoring that a computer can offer. These applications rely on a detailed cognitive skills assessment that determines what the user has mastered and what topics should be the focus over the next hours, or the next days of study. The current research program's primary objective is to develop probabilistic models for skills assessment that are automatically built from test data. We aim to construct skill assessment models solely from small data sets, yet with fine grained representations of skills. The approach we advocate, item-to-item models, was shown to outperform current approaches in terms of predicting a user's success to unanswered items. However, it also raises issues such as model update and the mapping of items to skills. This project addresses these issues, in addition to looking at new means to improve the predictive performance of existing models.