Content-Based Recommender System using Word Embeddings for Pedagogical Resources

Recommender Systems are important systems operating within a system to ensure how certain types of data are managed on the internet. These systems help users with overwhelming data and provide a better user navigation experience. This paper presents a content-based recommender system for online resources using deep learning. We have included some deep learning techniques to allow a good semantic understanding of educational resources. However, we have used a pre-Trained word2vec model owned by Google for the following three reasons: (1) Google is reliable; (2) the content of the Google news dataset is close to the content of the shared articles dataset; and (3) training a word2vec model is time-consuming and domain-independent itself. We have also used techniques for dimensionality reduction like t-distributed Stochastic Neighbor Embedding and Principal Component Analysis to reduce the dimensions of users and items vectors. Our approach aims to ameliorate the recommendations' accuracy and better satisfy the requirements of users. The results obtained when we tested our system are encouraging.