This paper presents a novel approach for recommending movies based on weighted Character Graphs. This approach proposes a dedicated crawler that gathers movie screenplays and a methodology of character graphs generation that contains all the necessary information needed for the representation of movie plots. A representative vector is extracted for each graph and used along with user ratings, as an input for a gradient boosting algorithm to predict movie ratings. The proposed method is tested on a publicly available MovieLens dataset and it was experimentally shown that it outperforms the fundamental collaborative filtering recommendation algorithms.