Constructing Grading Information System for Words´ Difficulty using a Supervised Learning Method

A primary factor in the design of an e-learning or e-testing system lies in the proper grading of a given set of vocabulary or phrases. This paper proposes an automatic system that can learn to do the fundamental word´s difficulty grading once taught by a domain expert. As a comparison, two approaches are used to analyze the system: the former is a rule-based system which requires a complete knowledge base to support a good scoring system, and the latter is an instance-based grader. Analytic and simulation results showed that rule-based reasoning acts much like uses the concept of absolute accuracy. Instance-based inferring method uses relative accuracy to identify varied difficult levels in words. Since scoring a word´s difficulty via absolute accuracy is not always applicable, the grading system accomplished the task by the instance-based approach. After checking for consistency by using the joint entrance examination center (JEEC) testing data in Taiwan, we summarize as follows: 1) The information system has the capability to answer the difficulty level of a word. 2) The system can persistently professionalize itself via the supervised learning approach. It can act more like a true domain experts when providing more training instances.