Text mining technologies applied to free-text answers of students in e-assessment: an experimental study in Greek

Abstract

Educational text mining is a rapidly growing field, thanks to the adoption of current probabilistic and machine learning algorithms. The current study focuses on student e-assessment through open-ended questions that require answers in the form of free text (student essays). Their analysis and evaluation are resource-demanding tasks for the instructor, even when supported by modern e-learning platforms. Topic modelling through the Latent Dirichlet Allocation algorithm is employed in an experimental setup, aiming to (a) extract meaningful topics from the body of pooled student answers (interpretable in the educational context of the course), (b) align the extracted topics to the ‘native’ internal structure of the body of texts, and (c) produce recommendations for the teacher in the form of alternative (meaningful) restructurings of the e-assessment units and consequently of the course content units. Quantitative and qualitative evaluation of the extracted topic models yield positive results for the first two aims, while at the same time, and regarding the third aim, the extracted topic models directly recommend for the teacher possible restructurings of the course content. These recommendations are of practical use for the teacher, especially when he/she seeks to restructure a course, either by shrinking or by expansion (fewer or more internal units). In conclusion, topic modelling opens a spectrum of possibilities for the teacher interested to explore ways to improve the structure and organization of his/her course.