Modeling techniques for adaptive practice systems

https://is.muni.cz/th/255651/fi_d/thesis.pdf

We study the challenges associated with the development of adaptive practice systems, particularly with the focus on the description of methods, which allow running such intelligent systems which give answers to questions occurring during development and automatize tasks usually done by human experts.

First, we describe various techniques which help to create a domain model that describes the knowledge domain on which the system focuses. Some of these techniques allow building a domain model automatically from the learners’ data, some of them combine the datadriven approach with input from a human expert and some process data and give experts an insight into the domain and foundations for further decisions.

Next, we depict several learner models which allow the system to adapt to every user and provide individualized content.

Finally, we deal with response times as extra information to the frequently used correctness of answers. We study how to combine these two types of information and how the addition of the response time benefits the models and the analysis. To evaluate these techniques, we use both simulated data and datasets from various real-world systems.

To study response times, we developed the MatMat system focused on the basics of mathematics; that is the domain where response time plays a significant role in measuring the learner’s success.

• Use of time information in models behind adaptive system for building flency in mathematics (Rihak, 2015)
  • Use MatMat (an adaptive practice system of arithmetic operations which guide children from basic work with numbers (e.g. counting objects) to mastery of basic mathematical operations)
  • MatMat dataset (11,344 children, 1491 items, 493,340 answers): data_description.md
• Properties and applications of wrong answers in online education system (Pelanek & Rihak, 2016)
• Choosing a student model for a real world application (Rihak & Pelanek, 2016)