While the initial motivation may seem to be derived from the popular “Netflix prize problem”, recommendation systems find their use and application in many other fields such as “Business-to-customer” in E-commerce, sports game results prediction, gene expression prediction, online learning recommender systems, etc.
The types of recommendation systems approaches include
-
Collaborative filtering: Collaborative filtering predicts suggestions for a particular by observing and processing the preferences of many users (collaborating). It works on the assumption that if person A has the same preference as person B on a product, A is more likely to have B’s opinion on a different product as well
-
Content-based filtering: It recommends other items similar to what the user has previously liked/preferred. Future predictions are based on their previous actions or explicit feedback
-
Hybrid approaches: This approach combines collaborative filtering, content-based filtering, and other approaches
Collaborative filtering is the basis of the Recommender System used in this project, and it predicts suggestions for a particular user by observing and processing the preferences of many users (collaborating). It can be achieved by applying Matrix Factorization, where a sparse matrix contains the user ratings for movies. Predicting the unknown matrix entry, which relates to user-item ratings, would result in a new suggestion of a movie, item, or product to the user.
The usage of recent technology in recommendation systems has been documented in . provides a thorough review of some highly notable works on deep learning-based recommender systems. This paper served as motivation to delve into Neural Architecture and Deep Learning. Deep learning effectively captures the non-linear and non-trivial user-item relationships, and enables the codification of more complex abstractions as data representations in the higher layers, hence gaining popularity over conventional recommendation systems.
Matrix completion techniques are explored in this project to build a recommendation system. The method was implemented on the MovieLens dataset using Keras to create a flexible model that can predict the possible ratings the user can give to unwatched movies and suggest the best among them.