Evaluation of C Programming Exercises by the Automatic Reports Grading Support System
Hiroyuki A. Watanabe · Journal of JSEE · 2009
In this research, the degree of similarity between reports is calculated by the one pair comparison method and the TF·IDF method, the z-scores are calculated by all words and some keywords in a report. The training data for the neural network is made by these values, and the automatic grading support system, which evaluates reports in 5 steps, has been developed. The reports of the C programming exercises submitted by e-mail could predict the score within an error of 10% by using this system, compared to manual operation. Furthermore, the office hours of 3 TAs could be reduced to 0, and contributed to improving the quality of writing reports for the students.