Do you flee social media like the plague so as not to read that Stranger Things review full of previews? Do you risk beating up your friend who loves to write posts with all the details about Avengers Endgame?
The diagnosis is clear: you have spoiler syndrome. Fear not, the University of California San Diego is working on your mental health already under strain.
Those good guys have developed an artificial intelligence that can tell you where spoilers are online, so you can avoid them.
“Spoilers are everywhere online, and social media especially is full of them. As users as well as researchers we understand the discomfort and how it can ruin the enjoyment of a book or a TV series,” says Ndapa Nakashole, professor of computer science and author of the study.
Some sites allow users to report things to others that might reveal details. Sometimes we come across the gentle premise “warning: contains spoilers”. Unfortunately this doesn't always happen.
This is why SpoilerNet was born, an anti spoiler AI that hunts for damned advances around and signals them as to avoid.
Theoretically it is a way to better understand what kind of linguistic paths and knowledge is used in the drafting of a text, but it is a blabla, come on. The truth is, researchers in San Diego broke spoilers and have the power to end this! Yes!
The team will present its comments to annual meeting of the Association for Computational Linguistics in Florence. The tool developed by the researchers could be used to build a browser extension. I foresee it, I imagine it, I hope so.
To train and test SpoilerNet, the UC San Diego team searched through spoiler databases. WARNING: CONTAINS SPOILERS – They found none. For this they had to compile one, putting together almost one and a half million reviews of books and series. The help of Goodreads is providential, a social network that allows readers and viewers to track and share their impressions (a bit like Anobii).
“To our knowledge, this is the first spoiler database on this scale and with this accuracy,” says Mengting Wan, Ph.D. in computational sciences and first author of the paper.
Curiosity
Researchers note that spoilers are often concentrated in the last part of the reviews. Apart from this assumption there are still things to be fine-tuned. First, it is not easy to make theAI the different styles (many) used by those who create spoilers. Second, on a semantic level there are words that change meaning as the context changes (example: “Black” is a color, but it can also be the name of a character).
Beyond these improvable aspects, SpoilerNet already performs quite well as an anti-spoiler: its accuracy is between 89% and 92% for books, and between 74% and 80% for series.
