A Novel Adaptation Model for E-Learning Recommender Systems Based on Student’s Learning Style
In recent years, a substantial increase has been witnessed in the use of online learning resources by learn- ers. However, owing to an information overload, many find it difficult to retrieve appropriate learning resources for meeting learning requirements. Most of the existing systems for e-learning make use of a “one-size-fits-all” approach, thus providing all learners with the same content. Whilst recommender systems have scored notable success in the e-commerce domain, they still suffer from drawbacks in terms of making the right recommendations for learning resources. This can be attributed to the differences among learners’ preferences such as varying learning styles, knowledge levels and sequential learning patterns. Hence, to identify the needs of an individual student, e-learning systems that can build profiles of student preferences are required. In addition, changing students’ preferences and multidimensional attributes of the course content are not fully considered simultaneously. It is by failing to review these issues that existing recommendation algorithms often give inaccurate recommendations. This thesis focuses on student learning styles, with the aim of dynamically tailoring the learning process and course content to meet individual needs. The proposed Ubiquitous LEARNing (ULEARN) system is an adaptive e-learning recommender system geared towards providing a personalised learning environ- ment, which ensures that course learning objects are in line with the learner’s adaptive profile. This thesis delivers four main contributions: First, an innovative algorithm which dynamically reduces the number of questions in the Felder-Silverman Learning Styles (FSLSM) questionnaire for the purpose of initialising student profiles has been proposed. The second contribution comprises examining the accuracy of various similarity metrics so as to select the most suitable similarity measurements for learning objects recommendation algorithm. The third contribution includes an Enhanced Collaboration Filtering (ECF) algorithm and an Enhanced Content-Based Filtering (ECBF) algorithm, which solves the issues of cold-start and data sparsity in- herent to the traditional Collaborative Filtering (CF) and the traditional Content-based Filtering (CBF), respectively. Moreover, these two new algorithms have been combined to create a new Enhanced Hybrid Filtering (EHF) algorithm that recommends highly accurate personalised learning objects on the basis of the stu- dents’ learning styles. The fourth contribution is a new algorithm that tracks patterns of student learning behaviours and dynam- ically adapts the student learning style accordingly. The ULEARN recommendation system was implemented with Visual Studio in C++ and Windows Pre- sentation Foundation (WPF) for the development of the Graphical User Interface (GUI). The experimental results revealed that the proposed algorithms have achieved significant improvements in student’s profile adaptation and learning objects recommendation in contrast with strong benchmark models. Further find- ings from experiments indicated that ULEARN can provide relevant learning object recommendations based on students’ learning styles with the overall students’ satisfaction at almost 90%. Furthermore, the results showed that the proposed system is capable of mitigating the problems data sparsity and cold-start, thereby improving the accuracy and reliability of recommendation of the learning object. All in all, the ULEARN system is competent enough to support educational institutions in recommending personalised course content, improving students’ performance as well as promoting student engagement.
- PhD