4 ms·

I am not sure. My understanding is that Panda 4.0 is an "anti thin content" algorithm update. If I were to define "good thick content" in a way that a robot co

I am not sure. My understanding is that Panda 4.0 is an "anti thin content" algorithm update. If I were to define "good thick content" in a way that a robot could understand, I might look at the following variables: - High wordcount, let's say maybe 1,000 words or more - At least 1 relevant image on the page - Bonus points if it's an original image (i.e. not a stock photo) - High ratio of words to links out - Lots of social shares Going with that profile, it's possible that ask.com etc. were the intended targets and Metafilter was collateral damage. This is all speculation so take it with a grain of salt.

protomyth 12y ago
Thanks, but I hope you are a bit wrong. The "lots of social shares" generally indicate to me IMHO that the site is not high quality, and the image thing just bugs me. I hate images for no good reason.
- hagbardgroup 12y ago
  For a dollar, you can get a ton of chaff social shares. All of these systems are thoroughly pwned.
fragmede 12y ago
> At least 1 relevant image on the page How would you figure out relevance algorithmically?
- protomyth 12y ago
  really far out there: Google has a search by image, maybe reverse that on an image and see if it is like one on a page with relevant keywords?
  - ehsanu1 12y ago
    Interesting idea, but sounds too expensive. I do believe the the image filename and alt/title attributes come into play, though those are easily spoofed so can't be taken too seriously by Google.
    - blauwbilgorgel 12y ago
      For example: Surrounding text, links to the page or image, topic analysis, color analysis, filename, alt attribute, OCR, advanced image recognition, Google Image Labeler (manual labeling). Reverse image search is probably too expensive to do this for every image found on the web, but Google could surely do this for a subset. Then they could match alt attributes: If two sites use the same image, and one has an alt attribute of "automobile" and the other "car in street" you can ask Bayes how likely it is that these two sites both made fake alt text that does not describe the image. Basically if Google Images can show you images relevant to your search term, then Google Search can find out if an image is relevant to a topic in much the same way.

protomyth 12y ago

Thanks, but I hope you are a bit wrong. The "lots of social shares" generally indicate to me IMHO that the site is not high quality, and the image thing just bugs me. I hate images for no good reason.

hagbardgroup 12y ago
For a dollar, you can get a ton of chaff social shares. All of these systems are thoroughly pwned.

fragmede 12y ago

> At least 1 relevant image on the page How would you figure out relevance algorithmically?

protomyth 12y ago
really far out there: Google has a search by image, maybe reverse that on an image and see if it is like one on a page with relevant keywords?
- ehsanu1 12y ago
  Interesting idea, but sounds too expensive. I do believe the the image filename and alt/title attributes come into play, though those are easily spoofed so can't be taken too seriously by Google.
  - blauwbilgorgel 12y ago
    For example: Surrounding text, links to the page or image, topic analysis, color analysis, filename, alt attribute, OCR, advanced image recognition, Google Image Labeler (manual labeling). Reverse image search is probably too expensive to do this for every image found on the web, but Google could surely do this for a subset. Then they could match alt attributes: If two sites use the same image, and one has an alt attribute of "automobile" and the other "car in street" you can ask Bayes how likely it is that these two sites both made fake alt text that does not describe the image. Basically if Google Images can show you images relevant to your search term, then Google Search can find out if an image is relevant to a topic in much the same way.