New Neural Net Detects Spoilers, Blocks Them

Spoiler alert: This article contains spoilers about new technology that’s going to block spoilers. 

Most of us have probably had a sullen look fall upon our countenance as we discover that some article — or maybe even headline — has spoiled the major plot points of a piece of media we’re excited to read or watch. But that heartbreaking moment when you realize that you already know what’s going to happen in some movie, TV show or book before you’ve seen or read it may be a worry of the past thanks to a new neural net trained to detect and block spoilers.


News of the new neural net, which comes via io9, should delight fans who can’t stand the idea of being in the know before they want to be; this is because it’s already quite effective and will likely only become better in time. The AI tool, developed by researchers at UC San Diego, has been dubbed SpoilerNet, and is reportedly 74 to 80% accurate at spotting plot-ruiners in articles about TV shows and 89 to 92% accurate at spotting spoilers in book reviews. No word on films, but it seems reasonable to guess that it’ll be on par with the TV numbers.

Training SpoilerNet to detect plot points wasn’t a straightforward task for the researchers who engineered the tool though, as nuanced language is hard and telling an algorithm what is and isn’t a spoiler is a pretty nuanced affair. But thanks to the Goodreads database, which provided 1.3 million book reviews marked with spoiler warnings, as well as 16,000 single-sentence spoiler reviews of TV shows (not from Goodreads), SpoilerNet was able to be trained to detect and block language that contained spoilers.

This task was difficult because language is often nebulous and varied — for example, the researchers point out in their report of the SpoilerNet tool that the word “green” could simply be a color in one review, whereas it could be the name of a character in another. They also note in the report that SpoilerNet would get hung up on words like “murder” or “killed,” even if they weren’t necessarily part of spoilers. But the researchers were able to deal with these issues by training SpoilerNet to look for spoilers that were hidden behind “view spoiler” links in articles and then determine the linguistic patterns generally used when discussing a spoiler.


As usually occurs with neural nets, previously unknown patterns became apparent once SpoilerNet started picking out spoilers. For example, it turns out that sentences containing spoilers “tend to clump together in the latter part of review.” So until you can get SpoilerNet on your computer — the researchers want to turn it into a browser extension — you may want to avoid spoilers by just bypassing the second half of reviews.

What do you think of SpoilerNet? Would you use it if it actually becomes a browser extension? Let us know in the comments!

Images: Dean Strelau

Top Stories
More by Matthew Hart
Trending Topics