Scalable and Practical Recommender System in Massive Open Online Courses (MOOCs)

Massive Open Online Courses (MOOCs) have witnessed a surge in popularity among learners and providers, prompting researchers to explore the enhancement of MOOC utilisation through recommendation systems. However, while existing efforts primarily focus on designing recommender systems capable of providing recommendations across diverse areas of MOOCs, less attention is given to essential characteristics of a robust recommender system, such as practicality and scalability. This literature gap leads to a lack of practical and scalable recommender systems capable of effectively handling increasing volumes of data. Furthermore, current systems leveraging neural networks and deep learning for user behavioural data-based recommendations face challenges in accommodating new data points without retraining. Consequently, scalable and practical recommendation algorithms for MOOCs remain limited in the current literature.

To address these shortcomings, this research aims to contribute to the development of scalable and practical recommender systems for MOOCs by utilising demographic and behavioural data of learners, alongside the unique characteristics of MOOCs, for personalised recommendations. In line with these objectives, the study introduced the Novel online Recommender system for MOOCs (NoR-MOOCs) specifically designed for course recommendation. Preliminary studies on the COCO dataset demonstrate the algorithm's promising performance. NoR-MOOCs constructs user profiles based on course ratings provided by users, and its performance evaluation employs predictive and classification accuracy metrics (RMSE, ROC, precision, recall and coverage). A comparison with traditional K-Means and collaborative Filtering algorithms validates the effectiveness of NoR-MOOCs.

Additionally, considering the abundant user behavioural and demographic data available in MOOCs, a Novel online Multi-attribute-based Recommendation algorithm for MOOCs (NoM-MOOCs) is designed. This algorithm creates user profiles based on their behavioural and demographic characteristics. Extensive experiments on the Edx and CAN datasets demonstrate the algorithm's significant out-performance compared to the Kernal Mapping recommender system (KMR) in terms of predictive and classification accuracy metrics. Moreover, the analysis of behavioural data from MOOCs logs contributes to forming valuable learning pathways for learners. Existing literature has extensively explored the design and implementation of learning pathway recommendation systems, with such recommender systems being intricately designed using neural networks and deep learning. To this end, a fast online learning path recommender system that incorporates demographic and behavioural data of learners to recommend learning paths is designed. This system identifies similar learners by leveraging demographics and behavioural data and recommends pats paths taken by successful students to struggling students with similar behavioural and demographic profiles. To evaluate the results of this proposed algorithm and the potential impact of the learning path recommender system on learners, an interview with the MOOC professor is conducted.

Overall, this research strives to contribute to the advancement of scalable and practical recommender systems for MOOCs, enriching the learning experiences of users and promoting the effective utilisation of MOOCs platforms.