Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Deep Learning on Title and Content Features to Tackle Clickbait

linkedin.com

Post reply on HN
Print view View original

11–20 of 31 posts

Re: Deep Learning on Title and Content Features to Tackle Clickbait

#11

Post by Cthulhu_ »

grabcocque wrote: post #8

The problem is, we don't have a clear definition of what clickbait is. nouninformal (on the Internet) content whose main purpose is to attract attention and encourage visitors to click on a link to a particular web page. But that's basically everything on the web.

A definition would need a bit more fleshing out, mostly about the (lack of) actual content; a long-winded page (not just text) that eventually leads to the core, which could be summarised in one line, even the article title itself. (like 'peanut butter is made out of peanuts' instead of "you'll NEVER guess this ONE SECRET peanut butter ingredient!"

Re: Deep Learning on Title and Content Features to Tackle Clickbait

#12

Post by BooglyWoo »

grabcocque wrote: post #8

The problem is, we don't have a clear definition of what clickbait is. nouninformal (on the Internet) content whose main purpose is to attract attention and encourage visitors to click on a link to a particular web page. But that's basically everything on the web.

I'm not sure that quite addresses the problem here.

After all there is no clear definition of 'what dogs look like' (in the sense of a collection of logical rules), but deep learning models excel at detecting them, when provided with enough positive examples.

If it's possible for humans to agree on whether a given article is clickbait or not, we should be able to put together an adequate dataset for training a system to classify them too. From the linked article I am unable to discern how the training dataset was labelled.

In other words, the fact that 'clickbait' is a nebulous concept shouldn't preclude machine learning from being able to detect it.

Re: Deep Learning on Title and Content Features to Tackle Clickbait

#15

Post by trevyn »

grabcocque wrote: post #8

The problem is, we don't have a clear definition of what clickbait is. nouninformal (on the Internet) content whose main purpose is to attract attention and encourage visitors to click on a link to a particular web page. But that's basically everything on the web.

The core of the definition is subtly wrapped in "main purpose" -- once the attention is attracted and the link is clicked, the clickbait's job is done. So the content of the article will be lower quality and less intellectually satisfying than non-clickbait articles.

For example, if you charted "interest on clicking this link text" vs "satisfaction with article after reading", I think clickbait would be clearly in the high interest vs low satisfaction quadrant.

Re: Deep Learning on Title and Content Features to Tackle Clickbait

#16

Post by hyperpape »

grabcocque wrote: post #8

The problem is, we don't have a clear definition of what clickbait is. nouninformal (on the Internet) content whose main purpose is to attract attention and encourage visitors to click on a link to a particular web page. But that's basically everything on the web.

I would define clickbait as articles which intentionally try to disguise what you'll get out of reading them. The information is banal, but the headline makes it out to be revolutionary or shocking.

You might disagree with the details of the formulation, but I think there's pretty broad agreement that something similar is going on with clickbait.

Re: Deep Learning on Title and Content Features to Tackle Clickbait

#17

Post by hyperpape »

mtgx wrote: post #10
baxuz wrote: post #6

You could add any title that's formulated as an imperative. "You won't believe..." "Guess which..." "You should..." Also titles that are formulated as a simple subject - predicate - object sentence: "XY considered anti-pattern" "Trump is right" "Hitler did nothing wrong" "Drunk girl shows tits" "Homeopathy is the future of medicine" Same works if formulated as a question: "Is Trump right?" "Has Hitler done nothing wr…

Not all clickbait headlines are written like that. For instance: "Russia hacked US power grid" doesn't have any of those, and yet it was a completely clickbait/sensationalist/borderline fake news headline from WashPost. How is AI going to deal with those ? https://theintercept.com/2016/12/31/russia-hysteria-infects-...

That wasn't clickbait. Arguably it was worse. "You'll never guess what happens when she starts to sing!" isn't likely to contribute to increased military tensions between nuclear powers.

To put it as a triviality: just because two things are bad doesn't mean they have to be bad in the same way.

I also wouldn't classify that story as "fake news"[0]. Those were things like "Revealed: Obama says Clinton would be terrible president", or "Revealed: Trump under investigation by European Court for Human Rights". Those were straightforward false claims, with zero actual sourcing, by people who knew they were lying. This Washington Post article was shitty reporting, using thin sources, that fit a currently popular hysteria. And it was completely inaccurate. But the authors didn't sit down and say "what can we make up." They got some sources and didn't do any due diligence, because it was too hard to pass up on such a juicy story.

I'm not wedded to the idea that these articles aren't fake news, but I'm confident it doesn't make sense to call them clickbait.

[0] Of course, this relies on the idea that fake news doesn't just mean "news that is wrong", which has been with us forever, but more about a social media driven trend within the past year or two.

Re: Deep Learning on Title and Content Features to Tackle Clickbait

#18

Post by samirahmed »

hikkigaya wrote: post #13

All I see is that the author uses deep learning to distinguish post published by Buzzfeed, clickhole, upworthy and stopclickbaitofficial v.s. the other pages?

yes - i am skeptical on how generalizable the final model is - given lots of features (numerical and text) are closely linked to same domain.

Re: Deep Learning on Title and Content Features to Tackle Clickbait

#20

Post by richdougherty »

grabcocque wrote: post #8

The problem is, we don't have a clear definition of what clickbait is. nouninformal (on the Internet) content whose main purpose is to attract attention and encourage visitors to click on a link to a particular web page. But that's basically everything on the web.

I guess it's an inherently fuzzy concept, so quite a good fit for machine learning.

But my definition of clickbait is any link I follow where I feel like I've been tricked into the click. The link looked interesting, but I feel regret once I see the actual content.

Post reply on HN