Hochschulschrift (Dissertation)

Open University of the Netherlands (OUNL), Heerlen, The Netherlands, 2009

ISBN: 978-90-79447-30-5

URL: http://hdl.handle.net/1820/2107

Accreditation of prior learning (APL, in the Netherlands referred to as EVC) enables institutions to offer personalized learning arrangements which take into account the prior learning of individuals. Unfortunately, this placement process of learners is time-and cost-intensive and cannot be scaled up easily to meet higher demands. Much work needs to be undertaken even before the potential learner makes a commitment to enter a study. Not only the assessment of an APL portfolio is time-consuming; literature shows that learners are unsure what to put in their portfolios for the APL procedure. Technological support may help reduce time and costs involved in this process and to improve the quality of the APL process. In this thesis we explore the application of advanced text mining and statistical natural language processing to offer a solution for these procedures which can be used in traditional APL procedures and informal learning networks. Based on the identification of three different situations which depend on the availability of data from learners we focus on the most complicated case where no (meta)data about the learners and the target learning content exists. A prototype model of a placement web-service for lifelong learning is proposed that is able to function as a supporting service by pre-analyzing documents in the APL procedure and by assisting in deciding whether these documents are relevant or irrelevant for the target course or study programme. This service employs in its core a reduced vector space model similar to Latent Semantic Analysis (LSA). While LSA uses very large corpora, in this thesis we evaluate the use of dimensionality reduction methods with smaller domain specific corpora. For this purpose we empirically evaluate the use of dimensionality reduction on the basis of two exemplary small corpora. We demonstrate that a combination of filtering strategies, the use of multiple criteria and dimension reduction that takes into account the variance accounted can help maximize performance. Based on these finding we have conducted a study with real learner data. Data is collected in a psychology course of the Open University of the Netherlands. The results of this study show that the application of dimensionality reduction techniques for APL procedures is a promising supporting method to decide about relevancy of learner portfolios. Based on these results we introduce two technological artifacts that have been developed during the project and which allow us to evaluate the approach in several different environments and to extend our approach on the different placement support situations introduced at the beginning of the thesis.

This Dissertation is licensed under a Creative Commons License .