Learning from Explanations: Multi-aspect based Age-restricted Rating Prediction in Long Scripts

Anonymous

Abstract: In the Motion Picture Association of America (MPAA), reviewers watch the entire film to determine the age-restricted category (MPAA rating) of the movie and provide the explanatory feedback for rating decision. As such human expert system is a time-consuming and non-scalable process, this paper proposes to develop a machine review system named MARS that automatically predicts the MPAA ratings of movie scripts. Specifically, in MARS, we first explore the use of the well-studied multi-aspect classification as machine-provided explanations, then leverage them to better learn the target rating prediction models. We demonstrate MARS outperforms various baselines by around 10 points in terms of F1 score, detecting severe contents with multi-aspect view.

Paper Type: short