Eureka! Institutional Repository

Recommendation systems are widely used today by a range of platforms for product and service recommendations to consumers, customers and subscribers. Recommendation systems’ goals include increasing sales, understanding the market, as well as the consumers, and increasing the experience quality of platforms and customer satisfaction. Recommendation systems have come to play an important role in an era of product and service abundance, with the goal of filtering products and services and increasing recommendation quality to consumers, depending on their interests. Notable recommendation systems can be found in platforms such as Google’s YouTube, large e-shops such as Amazon and E-Bay, social platforms in the likes of Facebook and Instagram and music platforms, for example, Spotify and Soundcloud. Meanwhile, libraries like numpy, pandas and scikit-learn can be used to implement recommendation algorithms in software. In this thesis we study the machine-learning library, PyTorch. PyTorch is a new addition to the deep learning framework. At a time when data become increasingly larger, usage of computer clusters for the parallel distributed processing of data becomes a growing common trend. In this thesis we analytically present and implement recommendation system algorithms, specifically the collaborative filtering kind, with the use of PyTorch. We study and train models of these algorithms, in a simple manner, as well as in a parallel and parallel distributed way, with the use of processes, and conclude by evaluating these algorithms in terms of accuracy and correctness.