| Summary: | 碩士 === 國立臺灣大學 === 電信工程學研究所 === 106 === Peer assessment has become a popular teaching method in recent years.
Students learn from feedbacks to improve their learning. They watch others’
work to learn from different perspectives, and turn their thoughts into
feedback words. As a result, the quality of feedback is an important factor
in peer assessment. This research gives an adaptive peer review matching
algorithm based on student profiles.
Student profiles are used to determine an appropriate matching between
authors and reviewers. Author profiles include features defined by a student’s
learning status while reviewer profiles are features for a reviewer’s behavior.
The two profiles are combined to predict the matching usefulness between
an author and a reviewer.
To assign peer reviews, our algorithm takes two steps. The first step is
“predict matching usefulness” which uses trained models with data from
KaiGon, an online peer assessment system. The other step is “find an
optimized matching assignment” which includes a constraint optimization
process that finds an optimal solution. In addition, we also present an
assigning method specifically for groups of students with certain needs.
The contribution of this research is that we propose multiple features for
student profiles to predict matching usefulness, and compare difference
between results of feed forward neural network and random forest models.
In addition, we simplify the optimization framework for expertise matching
for peer assignment, then alter the method to optimize matching needs for
different groups of students.
|
|---|
