Cross-repository aggregation of educational resources Articles uri icon

authors

  • MOURIÑO GARCIA, MARCOS ANTONIO
  • Pérez Rodríguez, Roberto
  • ANIDO RIFON, LUIS
  • Fernández Iglesias, Manuel J.
  • Darriba Bilbao, Víctor M.

publication date

  • February 2018

start page

  • 31

end page

  • 49

volume

  • 117

International Standard Serial Number (ISSN)

  • 0360-1315

Electronic International Standard Serial Number (EISSN)

  • 1873-782X

abstract

  • The proliferation of educational resource repositories promoted the development of aggregators to facilitate interoperability, that is, a unified access that would allow users to fetch a given resource independently of its origin. The CROERA system is a repository aggregator that provides access to educational resources independently of the classification taxonomy utilized in the hosting repository. For that, an automated classification algorithm is trained using the information extracted from the metadata of a collection of educational resources hosted in different repositories, which in turn depends on the classification taxonomy used in each case. Then, every resource will be automatically classified on demand independently of the original classification scheme. As a consequence, resources can be retrieved independently of the original taxonomy utilized using any taxonomy supported by the aggregator, and exploratory searches can be made without a previous taxonomy mapping. This approach overcomes one of the recurring problems in taxonomy mapping, namely the one-to-none matching situation. To evaluate the performance of this proposal two methods were applied. Resource classification in categories existing in all repositories was automatically evaluated, obtaining maximum performance values of 84% (F1 score), 87.8% (area under the receiver operator characteristic curve), 86% (area under the precision-recall curve) and 75.1% (Cohen's kappa). In the case of resources not belonging to one of the common categories, human inspection was used as a reference to compute classification performance. In this case, maximum performance values obtained were respectively 69.8%, 73.8%, 75% and 54.3%. These results demonstrate the potential of this approach as a tool to facilitate resource classification, for example to provide a preliminary classification that would require just minor corrections from human classifiers.

keywords

  • architectures for educational technology systems; cooperative/collaborative learning; distributed learning environments; media in education