An Experimental Study of Building User Profiles for Movie Recommender System

Abstract

User preferences and interests are usually represented in user profile, which is a crucial part in recommender systems. The more accurate user profile we can build, the better recommendations we can provide for the user. However, challenges in building and representing user profile exist. To overcome these challenges, in this paper, we focus on building user profile and its representation. We compare the different approaches in building user profile and optimization its weights. The study is conducted through experimental work using a real-world dataset, in which top recommendations based on user interests are provided. In addition, we extend the dataset by adding more features with the purposes of comparing its effects on recommendation results. The results show that user profile features and their weights have a great impact on recommendation results. More features we know about an item, more accurate user profile based on user interests we can build. Furthermore, properly optimized weights of the item features can increase the accuracy of recommendations.