Agent-assisted collaborative learning

Abstract

This work started off as an inquisitive question in our minds relating to whether students enrolling in large e-Learning courses can be helped to remain committed in the course that they enrolled in. This thesis has been a journey of discovery and learning which started in 2016. Right from the very start a number of processes were investigated in order to better understand the problem at hand and offer a possible solution to the matter. To achieve this aim, the study had to move out of the fringes of a purely technical solution and delve into other disciplines. Although the proposal being offered through this work is purely a technical one, it is based on different disciplines such as psychology of learning, philosophy. The historical and economic importance of learning was also explored, but the research was placed in appendices at the end of the document in order not to detract from the main line of the work: that of finding a way forward to assist students in completing their on-line commitments. In the Literature Overview chapter the creation of an e-learning ecosystem was discussed. Then the research moved into the way people learn and what makes them hold focus better. These phenomena were initially examined so as to lay a foundation on which the rest of the document would rest. Namely that people need to be involved in learning by giving them the means to follow up on the whole process, to scaffold through their learning and by making connections between facts that have been added to their repertoire. In essence this makes learning a holistic experience. It was argued that many of the main-stay e-learning platforms do not offer such an experience. Material production is excellent, but largely the student is left on his own. Collaboration between the different actors or agents in the learning process is necessary. So once the missing link was identified the research focused onto how one could fill in the gap and assume agent collaboration in learning. The focus then moved to examine various technologies that exist in different contexts but found to be lacking in education. An approach very similar to that of a recommendation system has been introduced. Just as buyers on a retail site are profiled and offered suggestions, students can be profiled in the same manner using their their progress and behaviour as input. The outcome of which would be that of suggesting areas that are of concern to the student. Another step in the learning process has also been put forward to complement recommendation. That of explanation. Many a time artificial intelligence algorithms work well, but leave little clues as to how they arrived to their conclusions. Consequently, excluding the human from the loop. The solution proposed in this thesis was not that of creating new algorithms that keep tabs on their internal workings. But that of annotating data to facilitate understanding, for the human actor, as he goes through the learning path. To achieve a better automated explanation the research departed from the rigid regimen of relational databases and moved to a more novel approach of representing data as a network or graph. In this way data could be linked, added and modified at will, adding flexibility and freedom to the underlying structures. Students, and even teachers, could then follow on the ever growing database as their own personal knowledge base. The traversal through the knowledge graph would then explain the “why” certain facts are linked together or else possibly highlight missing relationships in the knowledge. Also leaving open the possibility of exploration through the knowledge base. An artificial student performance data-set was used, closely reflecting that of a real class. This was used as a basis for profiling students. The data was labelled, and classified using a KNN algorithm. Students’ performance was then input back into the algorithm to profile the unknown performance metric with known groups in the data-set. After proper classification the student was then directed onto a pre-built knowledge graph containing the answers needed to improve knowledge. Through this study it has been shown that one can, with effort, set up a system that is able to follow users through their learning journey. This has been achieved by closely imitating the way humans expect knowledge to be presented. Rather than taking the human out of his context, technology was moved into the realm of the human.