| dc.description.abstract | As the volume of online information increases, recommender systems have been an effective strategy to overcome information overload by giving selective recommendations based on certain criteria such as user ratings and user interactions. Recommender systems are utilized in a variety of fields, with common examples being music recommendations and product recommendations on E-Commerce websites. These systems are usually constructed using either collaborative filtering, content-based filtering, or both. The most traditional way of developing a collaborative filtering recommender system is using matrix factorization, which works by decomposing a user-item interaction matrix into the product of two lower dimensionality rectangular matrix. However, as new technologies appear, matrix factorization is often replaced by other algorithms that could perform better than in a recommendation system. In recent years, deep learning has garnered considerable interest in many research fields such as computer vision and natural language processing. These successes are made possible by deep learning algorithms’ outstanding ability to learn feature representations non-linearly. The influence of deep learning is also prevalent in recommender systems, as demonstrated by its effectiveness when applied to information retrieval and recommender research. This research project performs an analysis and implementation on variants of two deep learning algorithms, autoencoder and restricted Boltzmann machines, and how they perform in recommender systems compared to matrix factorization. | |
