Emotion is fundamental to human experience and impacts our daily activities and decision-making processes where, e.g., the affective state of a user influences whether or not she decides to consume a recommended item - movie, book, product or service. However, information retrieval and recommendation tasks have largely ignored emotion as a source of user context, in part because emotion is difficult to measure and easy to misunderstand. In this paper we explore the role of emotions in short films and propose an approach that automatically extracts affective context from user comments associated to short films available in YouTube, as an alternative to explicit human annotations. We go beyond the traditional polarity detection (i.e., positive/negative), and extract for each film four opposing pairs of primary emotions: joy-sadness, anger-fear, trust-disgust, and anticipation-surprise. Finally, in our empirical evaluation, we show how the affective context extracted automatically can be leveraged for emotion-aware film recommendation.