RSS Feed

The Joel on Software Discussion Group (CLOSED)

Treat each click as a vote.

They could already be doing this, but from the number of garbage in the results, I doubt it.

The problem of people clicking up their sites in the search results is probably not applicable to image search, so no reason not to do it.

Nah, everyone knows that people will click on the porn just to get a bigger picture, regardless of what they are actually searching for!

One person's garbage is another person's treasure.

You'd be tied to the tyranny of the masses of what should show up when you look for a picture of "Paris"

They already have a solution:

http://images.google.com/imagelabeler/

Check out wired for a recent article.

> http://images.google.com/imagelabeler/

So inviting 1000s of editors to manually rate images is better than automatically deducing quality based on which images people actually click?

> You'd be tied to the tyranny of the masses of what should show up when you look for a picture of "Paris"

How is that different from regular search, where you get the sites that "the masses" link to for the term "Paris".

"How is that different from regular search, where you get the sites that "the masses" link to for the term "Paris"."

It depends on how you weight the user's votes for an image.  Voting for a regular search page by linking to it is only part of a search result ranking. 

If you had image voting and the ability to enable result weighting with that value (or hell, it'd be nice to be able to configure how a search engine weights its metrics for your own preferences) then I'd be more interested.


As an aside, it is interesting to search for "Paris" in Google Search, then image search, then groups:

Main web search shows only results for the city on page 1 for me, although a Wikipedia result for the city does show the Paris Hilton Wikipedia article as a sub-result.  It would seem that Google 'thinks' that "Paris" by itself refers, in general, to the city.

Image search is mostly Paris maps, and photographs of Paris locations, but has 3 Paris Hilton pictures.  I would fear a user-based ranking system that did not correct over large stretch of time to drown out spikes would pollute these results more.

Groups search shows 4/10 results about Paris Hilton, the rest about the city.  Obviously not the same degree of semantic or other filtering applied to Usenet searching, or simply people on Usenet don't talk a lot about Paris the city.

Interesting differences, but the question is: If most people searching images for "Paris" are actually interested in Paris Hilton (God only fathoms why) isn't that the most relevant pictures? That's not really pollution I'd say, but rather finding what you (or most) are looking for.

You are arguably correct; it is the difference between a search engine that reacts to popularity/current interest spikes versus one that tries to put a longer term view on things.

A compromise would be if you could flip a switch that could enable or disable application of "current interest" zeitgeist weightings on search results.

Assuming you had a way of measuring what people actually mean when they type in a search keyphrase, then you could either apply a decision of what that phrase has meant historically versus what it has meant over the last month.

Assuming, of course, you just don't "-hilton" or something :)