Mining Models for Automated Quality Assessment of Learning Objects

Cristian Cechinel (Federal University of Pelotas, Brazil)

Sandro da Silva Camargo (Federal University of Pampa, Brazil)

Miguel-Ángel Sicilia (University of Alcalá, Spain)

Salvador Sánchez-Alonso (University of Alcalá, Spain)

Abstract: The present paper presents the results of an alternative approach for automatically evaluating quality inside learning object repositories that considers lower-level measures of the resources as possible indicators of quality. It is known that current repositories face a difficult situation, as their amount of resources tends to increase more rapidly than the number of evaluations provided by the community of users and experts. Alternative approaches for automatically assessing quality can relieve human-work and provide temporary quality information before more time and consuming evaluation is performed. We propose a methodology to automatically generate quality information about learning resources inside repositories with Artificial Neural Networks models. For that, we considered 34 low-level measures as possible indicators of quality and we used available evaluative metadata inside two world recognized repositories (MERLOT and Connexions) as baseline information for the establishment of classes of quality. The preliminary findings point out the feasibility of such an approach and can be used as a starting point in the process of automatically generating internal quality information about learning objects inside repositories.

Keywords: artificial neural networks, learning object repositories, quality assessment, ranking mechanisms, ratings

Categories: L.0, L.1.2, L.3.2