Extraction of Poor Learning Items with Automatic Labeling in Fill-in-the-blank Test

In this paper, we propose a method to automate the systematic analysis of grading results in fill-in-the-blank tests. The proposed method evaluates appropriateness for labels from learning histories of learners to find the most suitable labels for blanks. A set of the most suitable labels to the blanks is used to create a graph which shows a variation of understanding levels of the learners. The method applies cluster analysis to the graphs in order to classify the learners with understanding levels for programming. It is implied that the proposed method classifies learners into 6 categories according to the understanding levels for programming.