4 ms·

Actually, shortly after I learned that Google switched to neural networks for search results, I noticed an old style of spam making a huge comeback: Either mark

by contem 8y ago

Actually, shortly after I learned that Google switched to neural networks for search results, I noticed an old style of spam making a huge comeback: Either markov chain or neural network generated text, taking from related websites (I suspect they look at the top 10 websites for certain valuable terms). This gibberish actually outranks legit content which refers to my content, sometimes even my own articles, especially when it is turned into a PDF. Seems like it is easy to block ~250k webpages like: https://www.google.com/search?q=inurl:?yhjhuyfib https://www.google.com/search?q=inurl:?yhjhuyfib but I guess Google keeps them there to keep the spammers in the dark? I hope so, else their new ranking signals allow for easier spam.

scrollaway 8y ago
> Seems like it is easy to block ~250k webpages like I think it's unfair to look at these results and say "but it's so easy to block these". Google's time is best spent on solutions which will reliably and automatically block these, without going through fairly manual steps. I did notice a decrease in quality in gmail's spam filter though. Increase in false positives and false negatives lately. I guess it's unlikely to be related...
- contem 8y ago
  To me it seems all the work of the same spammer(s). In such a case, do some manual intelligence and wrap it up. It won't scale to all forms of spam, but if a simple regex can uncover 250k+ results in 10 minutes, a manual spam fighter can still block millions of pages a day (and warn the webhost, remove these flakey ads from their networks, etc.). No doubt the recent machine learning hype has given spammers more advanced tools to avoid detection.
  - scrollaway 8y ago
    False positives are far more problematic than false negatives...
    - contem 8y ago
      If you remove from index... sure. But for that URL that I posted, do you think there is even a single false positive in there?

by contem 8y ago

scrollaway 8y ago

> Seems like it is easy to block ~250k webpages like I think it's unfair to look at these results and say "but it's so easy to block these". Google's time is best spent on solutions which will reliably and automatically block these, without going through fairly manual steps. I did notice a decrease in quality in gmail's spam filter though. Increase in false positives and false negatives lately. I guess it's unlikely to be related...

contem 8y ago
To me it seems all the work of the same spammer(s). In such a case, do some manual intelligence and wrap it up. It won't scale to all forms of spam, but if a simple regex can uncover 250k+ results in 10 minutes, a manual spam fighter can still block millions of pages a day (and warn the webhost, remove these flakey ads from their networks, etc.). No doubt the recent machine learning hype has given spammers more advanced tools to avoid detection.
- scrollaway 8y ago
  False positives are far more problematic than false negatives...
  - contem 8y ago
    If you remove from index... sure. But for that URL that I posted, do you think there is even a single false positive in there?