Google Letting Users Rank Search Results

Myriad writes "C|Net News is running an article about Google testing out a new system which would let users rank pages. From the article, 'Two weeks ago, Google began quietly testing a Web page voting system that, for the first time on a large scale, could eventually let Web surfers help determine the popularity of sites ranked by the company's search engine.'" As someone who has a lot of experience with systems where users self rate content, let me just wish Google the best of luck. Especially since for many unscrupulous businesses, ratings in search engines directly translate to dollars.

Comments

How this new system might *reduce* abuse

[dmoen]

A few weeks ago, I encountered "spam" on google. 8 of the top 10 links had been captured by a spammer using "cloaking" technology:

One method, called "cloaking," sets up a dummy page including lots of relevant information for keywords hidden through a special link. The cloaked page is fed to the search engine to boost a site's search ranking for specific terms such as "games," "sports" or "books." When surfers go to that link, however, they see a page that is different from the one indexed by the crawler.

I can't show you what it looks like, since Google has already fixed the problem.

What I wanted then was a "moderate" button I could click beside the link to indicate that it was spam. With a voting system like this, Google could locate and remove spam a lot quicker. Maybe that's what this is all about.

Doug Moen.

Re:How this new system might *reduce* abuse

[90XDoubleSide]

The new system is more geared toward lowereing the relevance of just plain low-quality sites; you can already report outright spamming, cloaking, and other abuse to search-quality@google.com.

Might work if...

[BluePenguin]

You know, this might work if Google implemented it the right way. I'm just thinking there are a few simple things they could do right off...

1. Don't put "rate this sight" next to every hit. Instead, use a system of random assignment. Every x(where x is a random number) hits, give the user a "rate this site" dialogue. This cuts down on the potential for direct abuse.
2. Add an option to sort by user rating, or sort by the current standard. This way, if people don't want to see user rated results, they don't have to.

I love google and all, but some of the things that make it to the top of the list from time to time are as useful to me as a 16 bit dos driver (for my RS/6000). It'd be good to see something resembling peer review on the web after all. Who knows, even if it fails, it might spark something that works! Best of luck google!

Re:Why not just monitor clickthroughs?

[realdpk]

They already monitor clickthroughs for 1% of the users hitting their site each day. Sometimes you'll notice it because the URLs you see when you hover over the links are different.

How to make automated votes expensive

[Tom7]

It's not that hard to make it really expensive to forge votes. For instance, check out the captcha project at CMU. (Basically, it generates images that are difficult for a computer to recognize, but easy for a human, and challenges the user to respond to them in some way to prove that they are human.) If they could find the right balance of convenience for humans and difficulty for perl scripts, I think they'd have a great thing going. I have always wanted this feature in a search engine ... I'm glad to see it happen.

Epinions Web of Trust

[telebear]

As someone who used to use Epinions all the time (making over $1000 from them), I have to say that the epinions "Web of Trust" system seems to work rather well, at least on a small scale (100,000 users).

Basically, you can see what users rated the article as useful. If you think that certain people have similar tastes to you, you put them in your Web of Trust. You'll get articles posted in a different order depending on who you trusted.

It is actually more complicated than that, as there are epinions "Experts" who are judged by epinions to have good ratings. I think Amazon has a similar system (and has way more users, but the system still seems to work ok).

The big problem is that the internet at large has so many bloody users and so many bloody pages... I think introducing groups of users or groups of groups that you trust might be a better way for the Web of Trust idea to work with the internet at large.

Re:Why not just monitor clickthroughs?

[Alomex]

This doesn't make any sense. Higher rated results will automatically get more clickthroughs.

Duh, you can adjust for that.

Compute the standard distribution of click-throughs according to position in the result set and any page outdoing this number gets "moded" up, any one underperforming this number gets moded down...

Another (Different) Rating Method

[dasunt]

If Google (or another search engine) set up all links to visit an internal google page that quickly redirected the user to the target site, it could rate on how many people visited the site, instead of a potentially biased rating of users.

Of course, shady websites could still influence it, either by hitting the pages themselves, or by crafting their page so that the google-selected text is tempting to search engine users, but the system still has the advantage of not requiring active participation of users.

Just my $.02

Disabilities interfere with these tests

[yerricde]

For instance, check out the captcha project [captcha.net] at CMU.

I looked at captcha and found that it may generate problems with disability legislation in some jurisdictions. For instance:

• Blind people and people behind text terminals can't pass bongo because it requires GIF images.
• Deaf people, people behind text terminals, people too poor to afford a sound card whose hardware interface is documented, and people not highly fluent in one of the six chosen languages can't pass byan.
• The fbw test generates sentences that still make perfect sense. For instance, it often chooses a proper name as the word to substitute, and users who do not have knowledge of the geography ("Evansville, CA" instead of "Los Angeles, CA") or the personal names of a particular region will often fail. The long sentences common to pre-1923 English literature produce a "needle in the haystack" effect. (The developers acknowledge that the fbw test is still under development.)
• Gimpy delivers broken images.

The only accessible test (fbw) doesn't always work, and the other three are not accessible to those with disabilities. Watch somebody get sued under the ADA.