Some thoughts on search
Jan. 9th, 2007 09:54 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
dsrtaoThe problem with Google's PageRank is that the pigeons don't have much incentive to get the best results, only the most popular results, as determined by looking at how many sites link to them. The first layer to solve that is to look at the pages people actually choose to click on that aren't already on the first page of returned results. That lets you improve your relative rankings.
But there's still a gap between popular and high-quality. At the extreme end of that are individual users' bookmarks, which presumably indicate
which sites are actually useful. You can rifle through those with Delicious and the like. But bookmarks are relatively unfocused, and the easier that they are to store, the lower quality any given site is likely to be. And people are notoriously uneven about weeding their bookmarks.
There's an earlier model which had some promise, but didn't scale well: the individual or small-group managed directory service. DMOZ/Open Directory Project is the leading exemplar there. An editor is in charge
of a particular subhierarchy, and some degree of reputation is on the line. But this is limiting in three respects: first, significant human attention is needed in selecting editors; second, editorial attention time is limited; third, suggestions are made manually and on a retail, not wholesale, basis.
Perhaps there's a way to distill the best parts of these approaches into something superior. The desired characteristics (I think) are:
But there's still a gap between popular and high-quality. At the extreme end of that are individual users' bookmarks, which presumably indicate
which sites are actually useful. You can rifle through those with Delicious and the like. But bookmarks are relatively unfocused, and the easier that they are to store, the lower quality any given site is likely to be. And people are notoriously uneven about weeding their bookmarks.
There's an earlier model which had some promise, but didn't scale well: the individual or small-group managed directory service. DMOZ/Open Directory Project is the leading exemplar there. An editor is in charge
of a particular subhierarchy, and some degree of reputation is on the line. But this is limiting in three respects: first, significant human attention is needed in selecting editors; second, editorial attention time is limited; third, suggestions are made manually and on a retail, not wholesale, basis.
Perhaps there's a way to distill the best parts of these approaches into something superior. The desired characteristics (I think) are:
- superior quality in search results
- uses human filtering in a way...
- which does not demand large time investments
- which rewards the filterer with at least reputation and, better yet, some tangible reward
- rewards the filterer proportionately to the quality of the filtering
- which streamlines the process of moving information both in and out
partly-baked thoughts
Date: 2007-01-10 04:12 am (UTC)I wonder how much value there would be in offering, for each item on the results page, a simple thumbs-up/thumbs-down option. Would that provide better data than just the click-throughs? It seems like thumbs-down is valuable as a way of distinguishing between "irrelevant" and "relevant but functionally a duplicate" (or "relevant but I've already seen that page", for that matter, or "maybe relevant but I'm not interested in PDF").