Course recommendation systems (back-end)

Abstract

Recommender systems have been widely used in different domains such as e-commerce and streaming services to provide personalized recommendations to its users. The underlying concept to recommend item is mainly based on the user and item interactions and the learned embeddings of these interactions. It is important to consider other types of information available such as the different modalities of items such as visual, textual and acoustic. Basic recommender systems mainly focus on user-item interactions, but it is insufficient to solely focus on them as they are unable to capture hidden dependencies between users and item modalities. This Final Year Project (FYP) focuses on implementing the model Muli-Modal Self Supervised Learning Recommender System. The state-of-the-art model investigates the visual, textual and acoustic modalities of item and the effects that it has on the accuracy of the recommendations when integrating these modalities into user and item embeddings. The model also aims to alleviate problems in recommendation systems such as data scarcity through self-supervised learning techniques such as Generative Adversarial Networks. Four public datasets (TikTok, Amazon-Baby, Amazon-Sports, All-Recipes) are used to evaluate the performance of the model by using common evaluation metrics such as recall, precision and NDCG. Results obtained are compared to other baseline papers and shows how the model outperforms existing data. Lastly, the implemented model can be translated into other types of recommendations such as course recommendations for NTU’s students and teaching staff. It could provide personalized course recommendations to students to improve their academic performance.