A New Tool Designed To Detect Spoilers Proves How Impossible It Is To Be Unspoiled

No one likes an errant spoiler showing up while reading a movie review, but what if there were a tool designed to detect those spoilers? That's what a group of researchers at the University of California, San Diego set out to make with their AI tool SpoilerNet, capable of accurately identifying sentences that are likely to give away major plot points. But in testing out their new spoiler tool, the researchers found that SpoilerNet had its job stacked against it.

A University of California, San Diego study (via io9) has developed "an AI-based system that can flag spoilers in online reviews of books and TV shows." But in the process, they have inadvertently revealed what is so impossible about spoiler culture: there is no defined parameter for what constitutes a spoiler.

To train the tool, the researchers created a dataset by collecting "more than 1.3 million book reviews annotated with spoiler tags by book reviewers. The tags encompass sentences that include spoilers and hide them behind a 'view spoiler' link in the text." Through this data, collected from Goodreads, researchers were able to train SpoilerNet to achieve an 89-92% accuracy, though even this accuracy was somewhat arbitrary. According to the study, there's no universal language to constitute a spoiler:

Researchers found that spoiler sentences tend to clump together in the latter part of reviews. But they also found that different users had different standards to tag spoilers, and neural networks needed to be carefully calibrated to take this into account.

In addition, the same word may have different semantic meanings in different contexts. For example, "green" is just a color in one book review, but it can be the name of an important character and a signal for spoilers in another book.

For example, take the word "snap": a word that in any Avengers: Endgame review, might be flagged as a spoiler. But it's a word that could be used in plenty of other contexts — snap judgment, snap decision. It's not a foolproof tool by any means. Perhaps the best comparison could be a "filter" on a social media platform that blocks out a certain word or phrase, but doesn't prevent negative information being spread on your feed. While SpoilerNet could eventually be refined and turned into a browser plug-in, it highlights a larger issue with spoiler culture: there's no ironclad definition for what constitutes a spoiler. At this point, human instinct is still probably the best way detect spoilers and common human decency toward other movie and TV lovers.