Empirical study of the behavior of several Recommender System methods on SAPO Videos

Abstract

In the last years, the internet became an indispensable tool for any company or internet user, which led to a huge amount of information being at every internet user's disposal. This information overload became a pressing problem making the user unable to keep track of his own interests. To solve this issue, recommender systems are developed to automatically suggest items to users that may fit their interests. The most popular strategies for predicting user preferences are: 1) Collaborative filtering, 2) Content-based filtering, 3) Social based filtering, 4) Social tagging filtering, 5) Knowledge-based filtering, 6) hybrid filtering, 7) context-aware filtering and 8)time-aware filtering. This thesis aims to do an empirical study regarding recommender systems strategies for the Sapo Videos website. The motivation for this work lays with assessing which is the best strategy for the proposed problem, that leads to finding the best tool and evaluation metrics. There are a lot of different tools and metrics to implement and evaluate this kind of strategies finding the best one will point out that best strategy. To accomplish this study it will be necessary to survey different recommendation tools, collect and prepare the data to be used on the experimental plataform that will be develop with some of the tools surveyed. In the end of each run of the experiment the data will analyzed using offline evaluation metrics that most suit the problem. Considering a growing number of platforms of online videos, this kind of recommendation systems also offers companies a great competitive advantage. It provides to its users personalized recommendations and also promotes their products.