As traditional recommendation techniques suffer from scalability problems resulting in poor-quality recommendations, they cannot be effectively used on big data. With the immense amount of emerging online learning resources nowadays, it has become harder for users to find and select their preferred content. Similarly, course recommendation systems also face an information overload problem. Most recommendation systems are created based on their own learning management systems and can only be used with those systems. Furthermore, the storage and processing of these systems cannot be updated, which makes them unsuitable for real-world problems, because data is continuously changing and emerging. Focusing on the aforementioned problem, in this study, we propose a novel online recommender system, namely, MCR-C-FGM. It runs on clusters and is trained with a fit-generator method which uses the Apache platform to distribute the processing of large datasets along with a clustering model created by a Deep Neural Network and Long Short-Term Memory. The network is trained with the fit-generator method. The test results with real MOOCs data from Harvard University and MIT, which were published in edX, show a high precision rate of 75%, an accuracy rate of 76%, and a recall rate of 78% in the evaluation processes. The time efficiency during the training process improves by 35% compared to the non-clustering model. Moreover, the MCR-C-FGM is capable of being scaled out, which allows it to efficiently support big data.
Keywords
Big Data, Course Recommendation, Deep Learn- ing, MOOCs.
ECTI TRANSACTIONS ON COMPUTER INFORMATION TECHNOLOGY