Spoiler detection in TV program tweets.

•It shows why we chose the support vector machine (SVM) for spoiler detection compared with five famous classifiers.•It explains that the verification of the features proposed in the previous work and how those features were selected and utilized to enable the SVM to classify spoilers from TV program tweets.•It shows that our method can work with semi-supervised learning (self-training) to reduce the amount of labeled data in an experiment that displays a similar level of performance.•We collected 2014 World-Cup tweets and conducted an experiment to prove that our method works in general.