{"id":148541,"date":"2019-07-10T21:27:52","date_gmt":"2019-07-10T13:27:52","guid":{"rendered":"https:\/\/www.geek.com\/featured\/-1795116\/"},"modified":"2019-07-10T21:27:52","modified_gmt":"2019-07-10T13:27:52","slug":"neural-network-trained-to-spot-spoilers","status":"publish","type":"post","link":"https:\/\/itteacheritfreelance.hk\/test\/wordpress\/2019\/07\/10\/neural-network-trained-to-spot-spoilers\/","title":{"rendered":"Neural Network Trained to Spot Spoilers"},"content":{"rendered":"
\n\t\t\t\t\t<\/p>\n
Hush, now\u2014spoilers (via BBC\/Doctor Who)<\/div>\n<\/figure>\n

Researchers after my own heart have developed an AI-based system that can flag spoilers in online reviews of books and TV shows.<\/span><\/p>\n

A team from the University of California San Diego will present their findings<\/a> at the upcoming Association for Computational Linguistics annual meeting in Italy.<\/span><\/p>\n

“Spoilers are everywhere on the Internet, and are very common on social media,” senior study author Ndapa Nakashole, computer science professor at UC San Diego, said in a statement<\/a>. “As Internet users, we understand the pain of spoilers, and how they can ruin one’s experience.”<\/span><\/p>\n

Amen!<\/span><\/p>\n

Some reviewers are kind enough to splash a spoiler-alert warning across their articles, before too much is revealed. But not everyone understands the crushing disappointment of prematurely knowing important plot details.<\/span><\/p>\n

The folks at UCSD do, though. So they invented SpoilerNet\u2014a neural network that automatically detects spoilers.<\/span><\/p>\n

To train and test their system, the team collected more than 1.3 million book reviews annotated with spoiler tags from Amazon-owned social cataloging site Goodreads.<\/span><\/p>\n

SpoilerNet has a knack for spotting literary leaks, performing at an impressive rate of 89 to 92 percent accuracy.<\/span><\/p>\n

It did struggle, however, with a dataset of more than 16,000 single-sentence summaries of about 880 TV shows, detecting spoilers with only 74 to 80 percent accuracy.<\/span><\/p>\n

Most of the errors, according to UC San Diego, came from the system getting distracted by words\u2014like “murder” or “killed”\u2014that are usually loaded and revelatory.<\/span><\/p>\n

“To our knowledge, this is the first dataset with spoiler annotations at this scale and at such a fine-grained granularity,” PhD student and first author Mengting Wan said.<\/span><\/p>\n

Spoilers date back to the early days of the Internet, when online etiquette insisted they could and should be avoided\u2014or at least marked by an obvious warning.<\/span><\/p>\n

But caution is often thrown to the wind, and many unwitting readers have had literature, films, and television programs “spoiled.”<\/span><\/p>\n

UC San Diego’s project is deeper than simply saving people from the threat of unwanted information. Researchers aim to better understand how people write spoilers, and what kind of linguistic patterns and common knowledge mark a sentence as such.<\/span><\/p>\n

Spoiler alert: Different users have different standards for tagging content; what’s important to one may be mundane to another. And that’s going to make calibrating a neural network quite challenging.<\/span><\/p>\n

Moving forward, the researchers expect their dataset could be used to train algorithms to detect spoilers on different platforms.<\/span><\/p>\n

More on Geek.com:<\/span><\/i><\/p>\n