Jump to content
toonkee

The Best Use of AI So Far

Recommended Posts

https://www.newscientist.com/article/2127107-sock-puppet-accounts-unmasked-by-the-way-they-write-and-post/

 

By Edd Gent

“Sock puppets” are the scourge of online discussion . Multiple accounts controlled by the same user can dominate comment forums and spread fake news. But now there’s a way to unmask the puppeteers.

A study of nine websites that use comment service Disqus to let readers post responses to articles found that sock puppets can be identified based on their writing style, posting activity and relationship with other users.

In the era of fake news, detecting sock puppets is important, says Srijan Kumar at the University of Maryland. “Whenever multiple accounts are used by the same party it is harmful and it skews the discussion and fake news can be propagated very confidently,” he says.

Kumar and his colleagues at the University of Maryland and Stanford University in California analysed commenter accounts on news websites including CNN, NPR, Breitbart and Fox News. They identified the sock puppets by finding accounts that posted from the same IP address in the same discussion at similar times. This approach isn’t always possible, so they wanted to develop a tool that automatically detects sock puppets based only on publicly accessible posting data.

They found that sock puppets contribute poorer quality content, writing shorter posts that are often downvoted or reported by other users. They post on more controversial topics, spend more time replying to other users and are more abusive. Worryingly, their posts are also more likely to be read and they are often central to their communities, generating a lot of activity.

Based on their findings, the researchers created a machine learning tool that can detect if two accounts belong to the same person 91 per cent of the time. Another tool can distinguish between a regular account and a sock puppet with 68 per cent accuracy. The research will be presented this week at the World Wide Web Conference in Perth, Australia.

This is the most comprehensive investigation of sock puppets in discussion forums, says Meng Jiang, who studies suspicious online behaviour at the University of Illinois at Urbana-Champaign.

But given that the group used sock puppets already identified by their IPs, it’s impossible to know if the tool could detect sock puppets the IP approach missed, he says. A real-world system would likely incorporate both approaches.

Kumar is confident the new tool could detect other sock puppets, and points out that IP addresses are not always available and can easily be spoofed.

The system could be useful to detect sock puppets on any forum that makes an account’s posting history available, such as social media site Reddit and most websites’ comment sections, he says. A person could then verify if an account breaches the site’s rules. “These tools always have a human in the loop,” he says. “It would flag suspicious accounts and a moderator would decide.”

Link to comment
Share on other sites

4 minutes ago, toonkee said:

https://www.newscientist.com/article/2127107-sock-puppet-accounts-unmasked-by-the-way-they-write-and-post/

 

By Edd Gent

“Sock puppets” are the scourge of online discussion . Multiple accounts controlled by the same user can dominate comment forums and spread fake news. But now there’s a way to unmask the puppeteers.

A study of nine websites that use comment service Disqus to let readers post responses to articles found that sock puppets can be identified based on their writing style, posting activity and relationship with other users.

In the era of fake news, detecting sock puppets is important, says Srijan Kumar at the University of Maryland. “Whenever multiple accounts are used by the same party it is harmful and it skews the discussion and fake news can be propagated very confidently,” he says.

Kumar and his colleagues at the University of Maryland and Stanford University in California analysed commenter accounts on news websites including CNN, NPR, Breitbart and Fox News. They identified the sock puppets by finding accounts that posted from the same IP address in the same discussion at similar times. This approach isn’t always possible, so they wanted to develop a tool that automatically detects sock puppets based only on publicly accessible posting data.

They found that sock puppets contribute poorer quality content, writing shorter posts that are often downvoted or reported by other users. They post on more controversial topics, spend more time replying to other users and are more abusive. Worryingly, their posts are also more likely to be read and they are often central to their communities, generating a lot of activity.

Based on their findings, the researchers created a machine learning tool that can detect if two accounts belong to the same person 91 per cent of the time. Another tool can distinguish between a regular account and a sock puppet with 68 per cent accuracy. The research will be presented this week at the World Wide Web Conference in Perth, Australia.

This is the most comprehensive investigation of sock puppets in discussion forums, says Meng Jiang, who studies suspicious online behaviour at the University of Illinois at Urbana-Champaign.

But given that the group used sock puppets already identified by their IPs, it’s impossible to know if the tool could detect sock puppets the IP approach missed, he says. A real-world system would likely incorporate both approaches.

Kumar is confident the new tool could detect other sock puppets, and points out that IP addresses are not always available and can easily be spoofed.

The system could be useful to detect sock puppets on any forum that makes an account’s posting history available, such as social media site Reddit and most websites’ comment sections, he says. A person could then verify if an account breaches the site’s rules. “These tools always have a human in the loop,” he says. “It would flag suspicious accounts and a moderator would decide.”

Why are people spending so much effort into identifying sock puppets? 

Most of the time they're lame and obvious. 

  • Like 1
Link to comment
Share on other sites

I don't know how well they'd do without the IPs, I think they're underestimating internet trolls. When I had multiple socks to troll the BCS facebook page and later the Coloradoan comments section for stadium articles, I made sure every sock had their own personality and writing style. I'd love to see how well I pulled it off and if their software would flag my accounts. 

  • Like 1
Link to comment
Share on other sites

14 minutes ago, DestinFlPackfan said:

Why are people spending so much effort into identifying sock puppets? 

Most of the time they're lame and obvious. 

Because people who use sock puppets are complete morons who deserve to be unmasked and mocked.

  • Like 1
Link to comment
Share on other sites

38 minutes ago, DestinFlPackfan said:

We have Mug for that role. Unless he's a sock...?

I'm not supposed to tell, but I have to spill the beans....Mug is actually @4UNLV's sock.  There, I've done it.

  • Like 1
Link to comment
Share on other sites

13 hours ago, Fort Fun said:

I don't know how well they'd do without the IPs, I think they're underestimating internet trolls. When I had multiple socks to troll the BCS facebook page and later the Coloradoan comments section for stadium articles, I made sure every sock had their own personality and writing style. I'd love to see how well I pulled it off and if their software would flag my accounts. 

sounds like alot time. 

  • Like 1
Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.



  • Recently Browsing   0 members

    • No registered users viewing this page.


×
×
  • Create New...