Large Scale Online Readability Assessment

Author(s): Varkevisser, R.A. (2016)

Abstract:
The internet is the new primary resource for information gathering. However, if you are a child, most of this information is either too difficult to read and/or understand, or impossible to find. This research focus on the readability aspect and provides a technical and theoretical foundation for the creation of a large scale child feedback readability assessment system. This research is divided into two steps. The first step measuring the scientific validity of digitizing the Cloze-test readability assessment method. The second step examining methods to automate the construction process of such a test. In the first step a user-test performed with primary school children proved the validity of a digitized version of the Cloze-test, showing significant correlations between the offline and online versions that were tested among the participants of the user-test. In the second step several automated methods were created, including TF.IDF- and NLP approaches that tried to emulate the human process of creating a Cloze-test. Performances of these methods varied and a combination of methods was shown to be most effective. Additionally, participants in a user-test gave conflicting reports when asked to identify the automated test in a mix of both automated- and human-made Cloze-tests.

Document(s):

Varkevisser_MA_EEMCS.pdf