Development of an Intelligent Adaptive Computerized Testing System Using Semantic Networks to Represent Itempool

Abstract

Adaptive testing is a computerized testing system which, for the purpose of raising the precision of the measurement of achievement, selects test items based upon examinee responses. However, in conventional adaptive testing the examinee levels of achievement can be defined in only a single dimension. This paper reports on an adaptive testing system which also evaluates the structure of examinee levels of understanding. This structure is represented by a semantic network to which the test items correspond, while the examinee model is described as an overlay to a domain network model. As a result of applying system to real testing situations we were able to more accurately evaluate the structure of the examinee's level of understanding. Our experiments also showed that the algorithm we have developed is effective in selecting items for this system.