In an article published in the journal Scientific Reports, researchers proposed a web mining- and machine learning (ML)-based classification approach for effective learning object classification in e-learning systems to maximize the reusability of learning objects.
Background
The reusability and sharing of learning objects have been significantly influenced by the extensive massive open online course availability and rapid distance-learning systems' development. These learning objects are primarily crucial educational resources that cater to the specific learning requirements of both students and teachers across different disciplines. However, the efficiency of the existing e-learning environments in using these learning objects effectively for reusability remains a significant challenge despite the increasing proliferation of these resources.
Several e-learning platforms do not have a systematic approach for learning object classification, which leads to a suboptimal organization of these resources and limited prospects for their multiple reuses and uses in various educational contexts. The use of ML techniques can maximize the reusability of learning objects and transform the e-learning system landscape, fostering authentic resource sharing and expanding opportunities for learners to explore these materials easily.
The proposed ML-based approach
In this study, researchers proposed a comprehensive web mining-driven and ML-based classification approach for classifying learning objects to significantly improve resource discovery and facilitate efficient knowledge dissemination in the e-learning environment. The objective was to enable learners to easily access the relevant information that aligns with their learning interests and goals.
They employed advanced ML algorithms to address the complex problem of learning object classification in e-learning systems. These techniques can effectively extract insights and patterns from substantial amounts of metadata of learning objects. Thus, learning objects can be efficiently categorized and organized based on their educational objectives, topic relevance, and unique characteristics by leveraging the abilities of ML.
Additionally, researchers also proposed an innovative approach that involves mining important information from the web to improve the classification of learning objects. Web mining techniques enable extracting valuable insights and data from online sources and complement the metadata-based classification process. Thus, a more holistic understanding and effective categorization of learning objects can be achieved by incorporating web mining.
The study consisted of two key phases, including the first phase, where the metadata was extracted from learning objects using web exploration algorithms, specifically feature selection techniques, to identify the most important features and eliminate the redundant features, and the second phase where ML algorithms were employed to categorize learning objects based on their specific forms of similarity.
The first phase substantially reduces the dataset's dimensionality and allows the development of practical and useful models, while the second phase accurately classifies objects by measuring their similarity using Euclidean distance metrics. Thus, the proposed comprehensive theoretical model/integrated approach for learning objects' classification allows systematic structuring and grouping of learning objects to ensure their seamless integration into different educational contexts.
Significance of the study
Researchers evaluated the effectiveness of learning objects through different ML techniques by performing a series of experimental studies on three real-world datasets. The datasets were obtained from Blackboard, a well-recognized digital learning platform possessing core learning management features; Schoology, a platform that provides essential tools to educators for student collaboration, communication, and lesson design; and Moodle, an open-source and free learning platform with precise solutions for educational requirements.
The diverse set of collected data encompassed practice, presentation, and conceptual models of learning objects across different categories. Specifically, using datasets from prominent companies enriches the potential of this study to comprehensively investigate learning object classification using ML techniques. Classification accuracy, F1-measure, recall, and precision were utilized as performance indicators during evaluation.
Results demonstrated that support vector machine (SVM) consistently attained exceptional accuracy in all three datasets. For instance, SVM achieved 90.03% precision, 87.5% F1-score, and 53.04% recall in the Moodle dataset, which indicated the effectiveness of SVM in accurately identifying important learning objects and minimizing false positives.
Random forest excelled in the Blackboard dataset by achieving the highest F1-score of 90.21%. This ML technique maintained a balance between recall and precision in this dataset and effectively classified learning objects with different attributes. Although naive Bayes displayed slightly poor performance compared to random forest and SVM, the technique still showed reasonable effectiveness and remained competitive in classifying learning objects.
Moreover, integrating the fuzzy C-means algorithm with random forest or SVM models led to significant improvements in several cases. For instance, SVM + fuzzy C-means achieved 92.07% precision, 91.4% F1-score, and 58.81% recall in the Moodle dataset, which indicated the feasibility of combining fuzzy clustering techniques to increase classification accuracy and the number of learning objects identified correctly.
Overall, the findings of this study demonstrated that the proposed ML-based classification approach yields promising and efficient outcomes for enhancing learning object reusability.