Don’t be Spoiled by Your Friends: Detecting Spoilers in TV Program Tweets

Abstract

Smartphones provide a convenient mechanism for accessing the Internet. As a consequence, smartphones have led to the rapid growth of Social Networking Services (SNSs) such as Twitter, and have become a major platform for SNSs. Nowadays, people are able to casually check SNS messages posted by their friends and followers via their smartphones. As a consequence, people can be exposed to spoilers of TV programs in which they are interested. To date, there are two previous works that explore the detection of spoilers in texts, not SNS: a keyword matching method and a machine-learning method that is based on Latent Dirichlet Allocation (LDA). However, the keyword-matching method is impractical and evaluates most tweets as spoilers, meaning that this method has poor recall performance. The other method, based on LDA, works poorly on short segments of text such as those found on Twitter, thereby evaluating most tweets as non-spoilers. We extract four features that are significant in the classification of spoiler tweets by analyzing tweets pertaining to a TV audition program (”Dancing with the Stars”). We demonstrate the drawbacks of previous works and compare those earlier works with our method. Our method achieves a 17% recall increase over the method based on keyword matching and an even greater increase compared to the machine-learning method based on LDA.