A comprehensive text analysis of lecture slides to generate concept maps. (December 2017)

• Record Type: Journal Article
• Authors: Atapattu, Thushari; Falkner, Katrina; Falkner, Nickolas
• Is Part Of: Computers & education. Volume 115(2017)
• Page Start: 96
• Page End: 113
• Publication Date: 2017-12
• Subjects: improving classroom teaching -- Media in education -- Pedagogical issues -- Teaching/learning strategies; Education -- Data processing -- Periodicals; Education -- Periodicals; Computers -- Periodicals; Computer-Assisted Instruction -- Periodicals; Éducation -- Informatique -- Périodiques; Electronic journals; 370.285
• DOI: 10.1016/j.compedu.2017.08.001
• Languages: English
• ISSNs: 0360-1315

Abstract

Abstract: Current instructional methods widely support verbal learning through linear and sequential teaching materials, focusing on isolated pieces of information. However, an important aspect of learning design is to facilitate students in identifying relationships between information. The transformation of linearity in teaching resources into integrated network models such as concept maps facilitates effective knowledge organisation by constructing relationships between new and existing knowledge. However, the manual construction of concept maps from teaching materials places an additional workload on the academics involved. Consequently, this research investigates the effectiveness of automated approaches in extracting concept maps from lecture slides and the suitability of auto-generated concept maps as a pedagogical tool. We develop a set of Natural Language Processing (NLP) algorithms to support concept-relation-concept triple extraction to form concept maps. Structural and graph-based features are utilised to rank the triples according to their importance. The natural layout of the lecture slides is incorporated to organise the triples in a hierarchy, facilitating highly integrated structure. Our evaluation studies identify promising results, with several case studies demonstrating a statistically significant correlation (rs > 0.455) between auto-generated concept maps and human experts' judgment. Auto-generated concept maps were rated from 'good' to 'very good' by the academics on evaluation factors such as coverage, accuracy, and suitability as a pedagogical tool. Thus, auto-generated concept maps from this research can be utilised as a positive alternative to the manual construction of expert concept maps and further, it is possible to utilise these maps for a wider range of applications including knowledge organisation and reflective visualisation of course contents. Our research contributes to bridging the gap between linearity in teaching materials and the necessity of creating integrated network models from teaching resources. Highlights: Investigate the effectiveness of auto-generated concept maps from lecture slides. Explore suitability and accuracy of auto-generated concept maps as a pedagogical tool. Develop NLP algorithms to extract concept-relation-concept triples from text. Bridge the gap between linearity in teaching materials and necessity of network models.