Deep Learning-Based Course Recommendations Using Sentence Embeddings and User Information for Learning Platforms

Online learning platforms have gained immense popularity in recent years, offering learners a wide range of course options. However, the vast number of available courses can make it difficult for learners to find the most relevant and suitable courses for their needs and interests. To address this challenge, a deep learning-based approach for personalized course recommendations is proposed, leveraging sentence embeddings and user information to enhance the relevance and accuracy of recommendations. The method utilizes the Multilingual Universal Sentence Encoder (MUSE) to generate dense vector representations of course content, effectively capturing semantic relationships between courses. These embeddings are then combined with user-specific features, such as level of education, year of birth, and geographical location, to create comprehensive user profiles. Two state-of-the-art recommendation models, Convolutional Sequence Embedding (Caser) and Session-Based Recommendations with Recurrent Neural Networks (GRU4Rec), are employed and trained on the edX dataset to predict the most relevant courses for each user. Experimental results demonstrate that the Caser model, incorporating embeddings and user information, achieves a Precision@5 of 0.190, Recall@5 of 0.897, and a Mean Average Precision (MAP) of 0.635. The GRU4Rec model, with embeddings and user information, obtains a Precision@5 of 0.185, Recall@5 of 0.876, and a MAP of 0.611. These findings show the improved performance of the models, resulting in more accurate and personalized course recommendations for users.