In this course project, we selected an existing tutor, identified and addressed interface issue that influence both user interaction and learning. We conducted Cognitive Task Analysis with several users, and analyzed log data from DataShop generated by previous experiments. We then tried to improve interaction with a more natural input functionality.
As Chemistry is a subject that values quantitative analysis and experiments, Stoichiometry is an important section in this domain. Apart from its significance, we also choose this domain because there is an existing tutor interface, and a relatively large amount of log data.
Findings from CTA
Using think alouds, we identified common strategies and errors in solving stoichiometry problems. We also discovered iteraction issues with the tutor interface. Furthermore, we hypothesized that the interface problems are also influencing learning. Thus, we tried to dig into the transaction data, to see if that is the case.
Findings from Data Analysis
First, we took a look at the learning curves of students' Knowledge Components (KCs). There are several smooth declining learning curves ("good" learning curves). However, there are also a large amount of "low and flat" learning curves, indicating that students are not learning those KCs. By the names of the KCs, we can see that they involve interactions with the tutor interface.
We then exported student transaction data from DataShop, and analyzed time spent on each type of interaction using pivot table. We found that the interaction "updateComboBox" is taking about a quarter of the time, and does not lead to learning.
The main feature of the rules is it evaluates student input as 3 parts: quantity, unit and substance, since they are three different KCs. It also allows variations of input, such as space / no space between quantity and unit.
Click here to check out the tutor. (Works best on a desktop screen)