Comment by eva1984 | Hacker News Reader

Comment by eva1984 | Hacker News Reader

Parent Full thread

No. Google is fooled by fake news too.

http://www.cbsnews.com/news/googles-top-search-result-for-fi...

I have some problem with those claims that offer little to none details. Use the above link as an example, how do we teach machine to verify this news? First, we need to teach them to formulate a question, in this case, 'Who has a larger number of popular vote [as of now]?', then it needs to find its way to verify the number from reliable sources, possibly need to struggle between conflicted sources as well.

This is no trivial, if we solve this, we might as well solve the problem of understanding human language as a whole.

The mentioning of Princeton students' Hackthon project is really offbeat, if the author really bothers to look into what that project actually is, nothing more than an API mashup.

https://github.com/anantdgoel/HackPrincetonF16/blob/master/b...

Disclaimer: I worked for a major news app, and the effort to battle fake news or low quality non-trustworthy news is ever growing. While application of machine learning is promising, human, especially experienced human is still the real workhorse for this problem

Edit: Typo.

You're focusing on the content.

Look at the source. 70news.wordpress.com should not be at the top of anybody's list of reliable news sources.

Look at the topic (e.g. from simple keywords). Look at the references (urls) mentioned in the article. Look at the timeliness.

There is plenty of information available to machine learning algorithms, without even needing to touch the natural language aspect.

> You're focusing on the content.

That's the only thing you can focus on. Why should someone be punished and labeled as fake if they're on a wordpress.com subdomain? Domain is just out, even spammy or weird looking domains may end up being popular / culturally fit. You would be preventing good journalistic folks from having any type of reach if you want to test against domains names, something that arguable may go away as far as the user is concerned in the next decade.

I'm unconvinced there is anything but content that you can really go off of. Keywords are not even used in search results, why would they determine news legitimacy? Even references are not always useful.

> Why should someone be punished and labeled as fake if they're on a wordpress.com subdomain

i don't know about "labeled as fake", exactly, but it's possible that quickly and easily acquired urls (such as a wordpress subdomain) may display a different amount of reliability (per whatever heuristic) in comparison with various established urls.

> Why should someone be punished and labeled as fake if they're on a wordpress.com subdomain?

Because metadata can be super useful in making inferences about page quality. Pagerank exploited this idea a ton. If wordpress subdomains are more likely fake in a training corpus, then that piece of evidence should be counted.

This problem has been going on for years and only now is it receiving wide recognition due to the recent election. I wrote about fake Google News in mid-2014:

"Google News Growth Hack Exposed" http://jmarbach.com/google-news-growth-hack-exposed

Good stuff! The black hat thread you link[1] is interesting reading: https://www.blackhatworld.com/seo/candybars-how-to-make-1-50...

I haven't worked on a news app, but why doesn't something like PageRank plus keeping track of a site's longevity handle this. It seems like fake sites would come and go quickly -- I doubt "70news.wordpress.com" has been around long, or gets many (non-gamed) links.

I'm surprised Google messed it up so badly. Was 70news manipulated to the top of the results page, or did their ranking algorithm totally screw up a simple query? Neither makes much sense.

Whitelist or blacklist or assigning PageRank to websites are tempting but also controversial because it will fall under the accusation of censorship. Take the below link as an example:

http://www.breitbart.com/2016-presidential-race/2016/10/16/m...

This is not 'fake' news, in a literal sense, because it just says someone is about to release something about Clinton. But for some people, this news could simply be seen as an indictment. Can we claim it is fake in that sense? As I have said above, it is equally controversial to penalize such websites for editorialization, because find story out of non-story is how most news agencies makes money and they have incentive to do this. Blame modern media and journalism for this.

In the end, using one simple algorithm to battle this is kinda of naive, not because it won't work for simple cases, but it fails to cover corner cases. In such scenario, human judgement is still crucial.

> In the end, using one simple algorithm to battle this is kinda of naive, not because it won't work for simple cases, but it fails to cover corner cases. In such scenario, human judgement is still crucial.

I agree that it doesn't cover corner cases, but why should it? If an algorithm knocks out 90% of the crap (like 70news.wordpress.com), there's much less work for the humans.

Also, wow, I didn't know Breitbart was quite that nutty. It reminds me of that LBJ quip along the lines of "it doesn't matter if he porks pigs or not, but now I made him deny it." This seems like a case where a human could do a bit of actual investigation, decide that the article's "Hillary is a lesbian" gist was BS, and permanently penalize Breitbart for publishing it.

Reply on news.ycombinator.com