Towards Efficient Item Calibration in Adaptive Testing

Reliable student models are vital for the correct functioning of Intelligent Tutoring Systems. This means that diagnosis tools used to update the student models must be also reliable. Through adaptive testing, student knowledge can be inferred. The tests are based on a psychometric theory, the Item Response Theory. In this theory, each question has a function assigned that is essential for determining student knowledge. These functions must be previously inferred by means of calibration techniques that use non-adaptive student test sessions. The problem is that, in general, calibration algorithms require huge sets of sessions. In this paper, we present an efficient calibration technique that just requires a reduced set of prior sessions.