Transformer Embeddings for the Implementation of a Semantic MOOC Recommender System

Semantic web technologies aim to interconnect objects using descriptors that identify their characteristics and improve information retrieval. In Massive Open Online Courses (MOOCs), the learning contents semantic links and their pertinence for learner needs are essential to motivate learners and reduce dropout rates. This work aims to develop a semantic recommender system prototype that uses external data sources on learners to identify their knowledge, skills, and interests to recommend MOOCs. First, this paper traces the knowledge-based recommendation of MOOCs and its data representation in the literature. Afterward, it explores a dataset of a MOOC platform to extract MOOCs content-related data, and a dataset of a professional social media to extract knowledge-based characteristics of a learner from his/her social profile. For the implementation of the semantic recommender, three different techniques: TF-IDF, GloVe, and USE models are used to compute the semantic similarity of learner needs and MOOCs content. The semantic MOOC recommender system implementation showed the advantages of using different data sources and the increased performance of a MOOC recommender system based on transformer models like the USE model.