Spoiler alert: machine learning approaches to detect social media posts with revelatory information

Spoilers---critical plot information about works of fiction that \"spoil\" a viewer's enjoyment---have prompted elaborate conventions on social media to allow readers to insulate themselves from spoilers. However, these solutions depend on the conscientiousness and rigor of Internet posters and are thus an imperfect system. We create an automatic alternative that could alert users when a piece of text contains a spoiler. An automated spoiler detector serves not only as an additional protection against spoilers, but it also contributes to important problems in computational linguistics. We develop a new dataset of spoilers gathered from social media and create automatic classifiers using machine learning techniques. After establishing baseline performance using lexical features, we develop metadata-based features that substantially improve performance on the spoiler detection task.