Hacker News "Submit" functionality needs a Digg-like duplicate alert

https://news.ycombinator.com/item?id=144765

by jharper • 18 years ago

16 16 18 years ago

news.ycombinator.com

Do you guys agree? I hate posting dupes ... but this site doesn't exactly discourage it.

jakewolf• 18 years ago

Hah, I beat you by 1/2 an hour http://news.ycombinator.com/item?id=144730

manvsmachine• 18 years ago

If I notice it early enough, I usually do what you just did and post the link to the original submission as a comment.

brk2• 18 years ago

Usually when I've submitted a dupe story, if it already exists it just adds a vote to the pre-existing story.

Where the logic either breaks (or allows people to subvert it) is where different URL's can get you to the same story. Often times URL's contain some superfluous flags that don't change the content served, but just serve to log some referrer or layout type data (I'm sure everyone reading this site gets how this works). Adding, removing, or changing any of this data seems to pretty much break or confuse whatever dupe-detector logic exists.

jharper• 18 years ago

Human dupe-detection would be an excellent extension to this process.

brk2• 18 years ago

Are you suggesting that new submissions route through Mechanical Turk?

LOL.

Once approach might be that when humans detect dupes, they could be reported. Click the "dupe" link, specify the URL(s) of the dupe(s), and submit. The oldest submission "wins", and the data could be used to train a bayesian dupe detector. I imaging that you could start with a URL text match (it's the ends of the string that tend to be different), along with a check of the <title> for the supposedly dupe page, and maybe the first 128 characters of the story text or something.

It actually sounds like a fun project.

Xichekolas2• 18 years ago

Why not just compare the content at the other end of the link with the contents of existing links.

It wouldn't be that hard. Whenever a link is submitted, YC's server would visit the link, get the response, strip all html tags and white space from it, then hash whatever is left. It would then store this hash value with the link. Whenever a new story is submitted, it is likewise hashed and then a check is made for an existing link with the same hash value. If it exists, it's a dupe, if not, allow it.

This would be an extra check to the existing dupe URL string of course. It still wouldn't catch every single thing, but it should eliminate quite a few easy dupes.

If that turns to have a low success rate, try hashing the page title or maybe the http headers.

brlewis• 18 years ago

Sounds like a nice-to-have feature. I don't think it's that bad to delete your own post if you notice the dup soon enough. No skin off your nose if the moderators kill it, either. That's happened to me.

tim2• 18 years ago

The way they have it executed is incredibly annoying.

chaostheory• 18 years ago

the only thing I noticed before (not sure if it's still valid) is that www.website.com is a dif submission than website.com

jey• 18 years ago

Those aren't the same URLs; the dupe detector currently only rejects exactly duplicated URLs.

pg• 18 years ago

It catches most dupes, just not on sites like the NYT that have so many different urls.

far33d• 18 years ago

Maybe you could check if the title of the page is the same as well, not as an automatic detection, but instead as a reason to ask "are you sure this isn't the same as foo". This might prevent most of the NYTimes dupes.

derefr• 18 years ago

Would it be that hard to just take a fulltext index of each page that hits the hot page? From there, just show anything with some >N% similarity (probably 98 or so, as text ads can affect the site a little bit.)

Hacker News "Submit" functionality needs a Digg-like duplicate alert

Related Stories