Analysing Features of Lecture Slides and past Exam Paper Materials - Towards Automatic Associating E-materials for Self-revision

Abstract

Digital materials not only provide opportunities as enablers of e-learning development, but also create a new challenge. The current e-materials provided on a course website are individually designed for learning in classrooms rather than for revision. In order to enable the capability of e-materials to support a students revision, we need an efficient system to associate related pieces of different e-materials. In this case, the features of each item of e-material, including the structure and the technical terms they contain, need to be studied and applied in order to calculate the similarity between relevant e-materials. Even though difficulties regarding technical term extraction and the similarities between two text documents have been widely discussed, empirical experiments for particular types of e-learning materials (for instance, lecture slides and past exam papers) are still rare. In this paper, we propose a framework and relatedness model for associating lecture slides and past exam paper materials to support revision based on Natural Language Processing (NLP) techniques. We compare and evaluate the efficiency of different combinations of three weighted schemes, term frequency (TF), inverse document frequency (IDF), and term location (TL), for calculating the relatedness score. The experiments were conducted on 30 lectures (900 slides) and 3 past exam papers (12 pages) of a data structures course at the authors institution. The findings indicate the appropriate features for calculating the relatedness score between lecture slides and past exam papers.