Movie collaborative filtering

Abstract

The growth of the internet has resulted in an excessive need of systems that personalize information for their users. Information Filtering and Information Retritwal have Jone a lot in this area but are still not enough especially with the ever growing internet. Likewise, personalized web-agents are still a little too far in existence. Instead the emerging recommender systems seem to be doing part of the job but are still developing. Traditional movie recommender systems shifted from suggesting the most popular movies and the highest rated movies to techniques such as Collaborative Filtering and Content-Based Filtering. In such techniques, recommendations are given based on past user ratings and by finding similar users that have rated more or less the same way in the past. This resulted in giving a personal touch depending on the likes of the user. In this dissertation we propose a hybrid system between Collaborative Filtering and Content-Based Filtering in order to exploit the advantages of each. Users of the system log on with their account on a centralized website and give integral ratings in the range of 1-5 for movies they have seen. Based on these ratings the system then attempts at finding other movies the user will probably enjoy watching based on three main factors: similar users, movie cast (particularly actors and actresses) and movie genre which should add to the accuracy of the predictions. The main movie dataset used was that of Netflix for the Netflix prize. The dataset consists of 17770 movies which were parsed and matched with those of IMDB in order to get other considered important data such as genre and cast that were not present in the Netflix dataset.