Description15 minute conference talk. Abstract:
Visual Algorithm Simulation (VAS) exercise is an interactive applica-
tion which teaches an algorithm or a data structure. The exercise shows
the student a visual representation of a data structure with initial data.
The student imitates the execution of the algorithm by interacting with
the visual representation. The student’s solution is graded automatically.
A misconception about the algorithm being learned can manifest itself as
systematic errors which can be modelled as a new algorithm. We are de-
veloping an intelligent tutoring component for VAS exercises: we assume
that the exercises could further support learning by giving constructive
feedback based on automatically detected, well-known misconceptions.
In a recent master’s thesis, a dataset of 1430 Build-heap VAS exercise
submissions was analysed manually by viewing student’s steps in each
submission. Hypothetical misconceptions were found in manual analysis,
and they were modeled as algorithms. The submissions are then automatically classified based on the misconceptions found using these algorithmic
The main result extends the set of known misconceptions of the Build-
heap VAS exercise. 52% of the submissions were correct, 17% were misconceptions and the rest 31% had a logical explanation. However, 95% of
submissions classified as misconception have multiple explanations with
heap size of 10.
We have ongoing doctoral research to develop the misconception detection and validation methods. Furthermore, we aim to find misconceptions
in multiple VAS exercises, such the ones featuring Quicksort and Dijkstra’s algorithm.
|18 marrask. 2021
|AI in Learning: Shaping the Future