Extracting Time-evolving Latent Skills from Examination Time Series

Examination results are used to judge whether an examinee possesses the desired latent skills. In order to grasp the skills, it is important to find which skills a question item contains. The relationship between items and skills may be represented by what we call a Q-matrix. Recent studies have been attempting to extract a Q-matrix with non-negative matrix factorization (NMF) from a set of examinees’ test scores. However, they did not consider the time-evloving nature of latent skills. In order to comprehend the learning effects in the educational process, it is significant to study how the distribution of examinees’ latent skills changes over time. In this paper, we propose novel methods for extracting both a Q-matrix and time-evolving latent skills from examination time series, simultaneously.