5 ms·

>Isn't the original PageRank (...) defined by what Google engineers believe to be a relevant and informative webpage? No, quite the opposite. The original Pag

by dexen 7y ago

>Isn't the original PageRank (...) defined by what Google engineers believe to be a relevant and informative webpage? No, quite the opposite. The original PageRank was defined by what other internet publishers considered to be a relevant and informative webpage. Specifically, the original PageRank was weighting content by the links (and content around the links) it was getting. The trust was placed on the internet publishers linking to it. The trust was distributed and also transitive to a degree. Which, if you think about it, is pretty close to what journalists do - they prop up the good ones in various ways. And also reasonably close to the peer review process popular in sciences - equal peers judging relevant materials and marking / linking the ones they approve. All in all, the original algorithm was effectively a "wall" between Google engineering and the actual content, as ranked & searched. It was clear delineation that "we don't editorialize search results". Granted, the later incarnations of Google Search take into account much more than just the original PageRank, and thus stray away from the idealized original formulation - and also get much closer to editorial decisions.

michaelmior 7y ago
> And also reasonably close to the peer review process popular in sciences I don't see the connection between the peer review process and PageRank. Peer review consists of educated experts explicitly making a value determination on the quality of the work before it is published and publicly available. Citations can be useful for judging impact, but this happens after the peer review process has been completed.
danso 7y ago
Right, I understand your point, and maybe I'm being too reductive here, but the idea heuristic that what people link to is authoritative is an opinionated decision that ended up being satisfactorily accurate (and efficient to implement). And evaluating the heuristic's accuracy ultimately boils down to human judgment; from their early Google proposal (though not the original Pagerank paper): http://infolab.stanford.edu/~backrub/google.html http://infolab.stanford.edu/~backrub/google.html > The biggest problem facing users of web search engines today is the quality of the results they get back. While the results are often amusing and expand users' horizons, they are often frustrating and consume precious time. For example, the top result for a search for "Bill Clinton" on one of the most popular commercial search engines was the Bill Clinton Joke of the Day: April 14, 1997. Google is designed to provide higher quality search so as the Web continues to grow rapidly, information can be found easily. (Yes, the opinion that a "bill clinton" query should probably return official/biographical pages before "Joke of the Day" is a pretty obvious and uncontroversial one)
- dexen 7y ago
  >but the idea heuristic that what people link to is authoritative is an opinionated decision There are two decision spaces here. One is the selection of search ranking algorithm (whether to use PageRank or any other), and that decision was taken by Google[1]. The second decision space is the decision whether to link (or not) to any given document, and what content to put around it. That is a long lasting iterative process rather than any singular decision. Arguably it's geared towards approximating the distributed consensus - which is a global mutable state; expressed with edges the graph rather than any singular node. Can the process be gamed, perverted, or corrupted randomly or maliciously? Sure. Do we know any better one? Not yet, at least not in the public. -- [1] practically speaking we all selected PageRank, by preferring Google's over multiple competing engines.
duskwuff 7y ago
> The original PageRank was defined by what other internet publishers considered to be a relevant and informative webpage. Specifically, the original PageRank was weighting content by the links (and content around the links) it was getting. The trust was placed on the internet publishers linking to it. Which fell apart badly when: 1. Personal web sites (and, later, even personal blogs) fell out of favor, reducing the quantity of both independent web sites to surface in search, and the number of available signals of page value 2. The migration of users to social media silos reduced the visibility of genuine personal recommendations, especially as those sites often applied "nofollow" to links in user-generated content 3. Search engine optimization became a thing, leading to a glut of low-quality pages (even from previously reputable publishers!) and massive abuse of the remaining signals of page quality

by dexen 7y ago

michaelmior 7y ago

> And also reasonably close to the peer review process popular in sciences I don't see the connection between the peer review process and PageRank. Peer review consists of educated experts explicitly making a value determination on the quality of the work before it is published and publicly available. Citations can be useful for judging impact, but this happens after the peer review process has been completed.

danso 7y ago

Right, I understand your point, and maybe I'm being too reductive here, but the idea heuristic that what people link to is authoritative is an opinionated decision that ended up being satisfactorily accurate (and efficient to implement). And evaluating the heuristic's accuracy ultimately boils down to human judgment; from their early Google proposal (though not the original Pagerank paper): http://infolab.stanford.edu/~backrub/google.html http://infolab.stanford.edu/~backrub/google.html > The biggest problem facing users of web search engines today is the quality of the results they get back. While the results are often amusing and expand users' horizons, they are often frustrating and consume precious time. For example, the top result for a search for "Bill Clinton" on one of the most popular commercial search engines was the Bill Clinton Joke of the Day: April 14, 1997. Google is designed to provide higher quality search so as the Web continues to grow rapidly, information can be found easily. (Yes, the opinion that a "bill clinton" query should probably return official/biographical pages before "Joke of the Day" is a pretty obvious and uncontroversial one)

dexen 7y ago
>but the idea heuristic that what people link to is authoritative is an opinionated decision There are two decision spaces here. One is the selection of search ranking algorithm (whether to use PageRank or any other), and that decision was taken by Google[1]. The second decision space is the decision whether to link (or not) to any given document, and what content to put around it. That is a long lasting iterative process rather than any singular decision. Arguably it's geared towards approximating the distributed consensus - which is a global mutable state; expressed with edges the graph rather than any singular node. Can the process be gamed, perverted, or corrupted randomly or maliciously? Sure. Do we know any better one? Not yet, at least not in the public. -- [1] practically speaking we all selected PageRank, by preferring Google's over multiple competing engines.

duskwuff 7y ago

> The original PageRank was defined by what other internet publishers considered to be a relevant and informative webpage. Specifically, the original PageRank was weighting content by the links (and content around the links) it was getting. The trust was placed on the internet publishers linking to it. Which fell apart badly when: 1. Personal web sites (and, later, even personal blogs) fell out of favor, reducing the quantity of both independent web sites to surface in search, and the number of available signals of page value 2. The migration of users to social media silos reduced the visibility of genuine personal recommendations, especially as those sites often applied "nofollow" to links in user-generated content 3. Search engine optimization became a thing, leading to a glut of low-quality pages (even from previously reputable publishers!) and massive abuse of the remaining signals of page quality