Incidence of specific semantic characteristics on the performance of recommender systems of audiovisual content

This paper presents a comparison of different simple methods of recommending audiovisual content. The recommendations are performed based on the interest the user had on specific semantic characteristics of previous seen content, such as the actors, directors, writers or genres, or the average IMDB rating of the content, or a combination of them. The tests are based on a subset of MovieLens 1M database and information about the content has been collected from IMDB. Finally, results are compared with the ones obtained by semantic filtering in a previous work.