Removing Objectionable Content From Social Networks - CS Problem?

Replies

• 

  Prasad Ajinkya

  Create an expletives database! Whenever the keyword occurs, target that content. Obviously there would be false positives, but the algo can be made to learn from its mistakes.
  
  Second option is to do what they are currently doing, crowdsource it and ask community members to flag inappropriate content.
• 

  Kaustubh Katdare

  kidakaka

  Create an expletives database! Whenever the keyword occurs, target that content. Obviously there would be false positives, but the algo can be made to learn from its mistakes.
  
  Second option is to do what they are currently doing, crowdsource it and ask community members to flag inappropriate content.

  Well, that's one of the most common ways of preventing abusive words. We do that on CE as well. But then, it's an old mechanism. Abusive material can be uploaded in various forms: morphed photographs and double-meaning jokes; which are difficult to scan for objectionable content.
  
  Can there be an algorithm that matches ip, geolocation, history, past content, age, friendlists etc. to figure out who's posting objectionable content?
• 

  Prasad Ajinkya

  I think so.
  
  You can create a predictive model even.
  
  Statistics (more specifically correlation) tells us how to correlate two seemingly independent variables and predict one on the basis of the other.
  
  For. e.g If a person is continuously clicking only on certain types of links, then perhaps that person is more vulnerable to phishing scams. As long as you have different factors (some of which you have already listed down), you can take this historical data and find the exact significance of them with that person putting up abusive content.
  
  Biggie, if you can share this data (for CE) on a spreadsheet, you can create one predictive model yourself as well. Softwares such as SPSS/SAS can be used to create such models.
• 

  Kaustubh Katdare

  Well, the only data we collect is IP address and e-Mail address. Not at all useful to create a predictive model, I think. But we can surely generate some test-data, right?
• 

  PraveenKumar Purushothaman

  Moreover, there are profiles, which do not allow public sharing of the content, which is restricted inside their friends' circle. Another thing is language. You can say a vulgar word, of one language in a different language, right? "Loose" can be said as "Loosu" and this can be understood by a human but not a computer. If people crowdsourcingly do this job, then what is the meaning of the privacy policies been held in their ... commitments???
• 

  Anoop Kumar

  There is option to mark offensive and if certain post/profile marked as offensive N number of times then it should be removed.But to collect N numbers is big task so one option is marking as hirarchy of verified profiles. On facebook every use is on 4.65 degree connetion. so marking a verified profile with degree of reliability is not a big task.
  
  suppose 100 points needed to remove a containt then:
  (marked offensive by unverified) + (marked offensive by verfied * reliability) = x points.
  by this way making x=100 is a easy and quick fast.
  Also verified profile should be liable to their action. and user should be on choice that their profile can be marked as verified or not.also user can claim that his post is not offensive.
  Ex.: mark zukerberg profile is verified and degree of reliability is 100 then any post which is marked by zukerberg will be removed instantly.
  Most of the containts only shared by friends then user's friends can mark his post as offensive and can be removed.
• 

  PraveenKumar Purushothaman

  ianoop

  There is option to mark offensive and if certain post/profile marked as offensive N number of times then it should be removed.But to collect N numbers is big task so one option is marking as hirarchy of verified profiles. On facebook every use is on 4.65 degree connetion. so marking a verified profile with degree of reliability is not a big task.
  
  suppose 100 points needed to remove a containt then:
  (marked offensive by unverified) + (marked offensive by verfied * reliability) = x points.
  by this way making x=100 is a easy and quick fast.
  Also verified profile should be liable to their action. and user should be on choice that their profile can be marked as verified or not.also user can claim that his post is not offensive.
  Ex.: mark zukerberg profile is verified and degree of reliability is 100 then any post which is marked by zukerberg will be removed instantly.
  Most of the containts only shared by friends then user's friends can mark his post as offensive and can be removed.

  I guess you should be heard of fake votes!!! 😛
• 

  Anoop Kumar

  Thant's why I added verified profiles+facebook's 4.65 degree+level of reliability paradigm to avoid fake votes.
  My idea could be wrong. But in my views its a good way 👀
• 

  PraveenKumar Purushothaman

  ianoop

  Thant's why I added verified profiles+facebook's 4.65 degree+level of reliability paradigm to avoid fake votes.
  My idea could be wrong. But in my views its a good way 👀

  Dude, how long does it take to make a verified profile and do all the fake stuffs?
• 

  Anoop Kumar

  google map maker works that way... most of the streets and roads and places added by users and then moderated by google moderators. 99% of google maps is accurate.
  if you raise the question about fake profile that why I suggested hierarchy with level of moderation.
  Most powerful is facebook degree of connectivity (4.65) that means a user is at most 4.65 connection away from any other facebook's user.
• 

  Prasad Ajinkya

  Praveen, I think ian has a good valid point.
  
  By verified person, he means a trusted source (such as our CE Mods). If a Mod says this is spam, then its spam. Hey ... Mod is God ;-)
• 

  PraveenKumar Purushothaman

  kidakaka

  Praveen, I think ian has a good valid point.
  
  By verified person, he means a trusted source (such as our CE Mods). If a Mod says this is spam, then its spam. Hey ... Mod is God ;-)

  GaaaD!!! 😛 Lolz, yeah... I agree, but yeah, how do people become MaaD and GaaD in these cases? It will be automated. In our case, Biggie knows the Mods personally, but not in the case of public workers like G Maps... 😲
• 

  Prasad Ajinkya

  hmm, take the case of Wikipedia, even in those case the becoming a mod process is still part-manual, part-automated.
• 

  PraveenKumar Purushothaman

  kidakaka

  hmm, take the case of Wikipedia, even in those case the becoming a mod process is still part-manual, part-automated.

  In Wikipedia, since there are millions of users and people find some changes, they act upon it... Moreover, there are bots, which go through the content, full protected, semi protected, etc., and moreover, same as here, the mods in Wikipedia are well known to a single person... Agree?
• 

  Prasad Ajinkya

  Yep. Certain processes cannot be automated. Certain can.
• 

  sookie

  Human Moderators ?
• 

  Kaustubh Katdare

  sookie

  Human Moderators ?

  Not a very workable option.
• 

  PraveenKumar Purushothaman

  The_Big_K

  Not a very workable option.

  But right now all the mods are humans right?
• 

  Kaustubh Katdare

  Praveen-Kumar

  But right now all the mods are humans right?

  Mods may answer this 😉
  
  The content generated on CE is a fraction of what's generated on FB or Google.
• 

  PraveenKumar Purushothaman

  The_Big_K

  Mods may answer this 😉
  
  The content generated on CE is a fraction of what's generated on FB or Google.

  Consider Wikipedia... Or even ODP... They all are humans... 😛