Despite the fact that we live in a world subsumed by varying iterations of the exact same superhero movie — sometimes repackaged just enough to be called a sequel or a reboot — spoilers continue to piss off people. Sure, the good guys mostly always win, but how they win and to what extent they “save the world” is the mystery that keeps us coming back to see Spider-Man do his spider-thing again and again (I’m not even sure which installment we’re on now because I stopped caring after Tobey went emo).
The point is, our collective movie-going experience is as much defined by already knowing the ending of a movie because we’ve seen something eerily similar to it just a few months before, as it is our collective freak out when someone spills the beans on how Jon Snow dies. (Very late spoiler: He doesn’t really die.)
This tragedy, exacerbated by an increasingly connected world and a content cycle that sleeps for no person, is the basis for a new type of A.I. tool developed by researchers at the University of California, San Diego, designed specifically to help flag spoilers in online reviews of books and TV shows.
“Some websites allow people to manually tag their posts with tags that serve as ‘spoiler ahead’ warning signs,” says Mengting Wan, the lead author of the study around the A.I. “But people don’t always use these tags. So naturally, we wanted to develop an A.I.-powered tool to automatically detect spoilers and complement existing approaches.” A key part of this, Wan continues, was trying to figure out answers to questions such as how people write spoilers, and what kind of linguistic patterns and knowledge make a sentence a ‘spoiler’ in the first place.
To find out more, the researchers collected approximately 1.3 million book reviews from Goodreads containing spoiler tags. “This large-scale dataset basically allowed us to develop modern ‘data-hungry’ deep learning models for this problem,” says Wan. “Using our new dataset, we trained a deep neural network model to capture the spoiler patterns, and applied this model on new posts to flag spoiler sentences.”
Based on their findings, the researchers found that spoiler sentences tend to clump together toward the end of a review. “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,” per Science Daily.
Bad news (or should I say, spoiler alert) for those who only consume video content, however: According to Wan, currently the A.I. tool only works on text articles. There is, though, a silver lining. “For video reviews, if we can transfer audios into texts using speech-recognition techniques, then the A.I. tool can also be applied in this scenario,” she explains.
While this tool may seem somewhat superfluous, since spoilers aren’t nearly as soul crushing as everyone seems to believe — that said, I still haven’t seen the final episode of The Sopranos, so please don’t fuck it up for me — per Wan and her team, there’s really no better solution to mitigate spoilers than having a robot do it for you. This is especially true since, she reiterates, most people don’t bother to use spoiler tags even when they’re available. “Another possible solution is crowdsourcing,” says Wan. “Consumers can report reviews that reveal critical plot details, but this method may have scalability issues: It’s relatively difficult to engage sufficient consumers to do so in a timely fashion. Thus, in this work, we try to approach the problem from a different angle: Let A.I. help us to detect spoilers. In terms of applications, such an A.I.-powered tool can not only help people avoid reading spoilers, but also ease authors’ writing process and help them flag spoilers.”
If you happen to be extremely sensitive about spoilers, until this software becomes widely available, you’ll just have to avoid them the old-fashioned way: Encasing your head in a block of cement and hurling yourself into the Atlantic Ocean until you’ve had a chance to see the movie you don’t want spoiled. Works like a charm.