Improving recommender systems with machine learning and social media

This thesis studies the opportunity to utilize posts from social media in recommender systems. Recommender systems are used to help users find content they are interested in when there are many alternatives, and is very common in streaming services. One of the problems of recommender systems is called the cold-start problem, where the system does not have enough information about an item or a user to make accurate recommendations. We want to mitigate this problem by gathering information from Twitter. Twitter is an extensive source of information about real-time events, but because of considerable noise it can be difficult to find the information relevant for a recommender system. In this thesis, we present a system capable of gathering and organizing information from Twitter in a way that makes it usable for a movie recommender system. This is done by filtering tweets about new movies through a classifier based on an artificial neural network that sorts movie-related tweets from non-movie-related tweets. After finding relevant tweets, our system can tell movie titles apart by clustering the data in such a way that clusters are assigned a movie title and a sentiment. The solution gathers data from an unstructured source and organizes the data in such a way that makes it usable by a recommender system. The solution is evaluated against previous research concerning both clustering, sentiment analysis and classification. The results indicate that social media and Twitter in particular is a useful, extensive source than can be used to mitigate the cold-start problem of recommender systems