Stories can have tremendous power -- not only useful for entertainment, theycan activate our interests and mobilize our actions. The degree to which astory resonates with its audience may be in part reflected in the emotionaljourney it takes the audience upon. In this paper, we use machine learningmethods to construct emotional arcs in movies, calculate families of arcs, anddemonstrate the ability for certain arcs to predict audience engagement. Thesystem is applied to Hollywood films and high quality shorts found on the web.We begin by using deep convolutional neural networks for audio and visualsentiment analysis. These models are trained on both new and existinglarge-scale datasets, after which they can be used to compute separate audioand visual emotional arcs. We then crowdsource annotations for 30-second videoclips extracted from highs and lows in the arcs in order to assess themicro-level precision of the system, with precision measured in terms ofagreement in polarity between the system's predictions and annotators' ratings.These annotations are also used to combine the audio and visual predictions.Next, we look at macro-level characterizations of movies by investigatingwhether there exist `universal shapes' of emotional arcs. In particular, wedevelop a clustering approach to discover distinct classes of emotional arcs.Finally, we show on a sample corpus of short web videos that certain emotionalarcs are statistically significant predictors of the number of comments a videoreceives. These results suggest that the emotional arcs learned by our approachsuccessfully represent macroscopic aspects of a video story that drive audienceengagement. Such machine understanding could be used to predict audiencereactions to video stories, ultimately improving our ability as storytellers tocommunicate with each other.
展开▼
