Interesting article on search engine usage

Check out this study by Enquiro and Eyetools

[Irie]

Enquiro and Eyetools came up with this awsome study that shows just how much emphasis a search user places on the first three results and conversly how little he/she places on results further down the page (not to mention on pages 2+) [prweb.com...] .

Does anyone know of further user or mathematical studies related to search?

Also, are there fora for discussing the architecture of internet search rather than SEM?

-Irie

[digitalghost]

The general objective of observation is to record the activity in its natural state. Not in a lab, given directives.

[Irie]

I understand, but we don't know the parameters or methodologies of this study. I think that take-away is that there is no real need for tens of pages of useless results. My big kick is that we need to refine the search algorithms and the whole paradigm of internet search. I think that we're doing a terribly crappy job thus far. Take for instance a search for "lie algebras". There are 437,000 hits for this obscure term. If users only look at the first 10, are the search engines really doing their job? Take another case, "ROI SEO" (I wanted to find out what is a typical ROI on SEO). This query is completely populated by useless business spam entries. If 80% of queries are for information, why do 99% of the results for this comercial in nature?

What are your thoughts?

[digitalghost]

Check out Yahoo Mindset [mindset.research.yahoo.com]. The issue of commercial results and information results is being addressed.

As for the protocols in the study, well, they didn't publish those did they? ;)

[Irie]

Admittedly, I did not know about this one. It's a good start. However, my search for "ROI SEO" at the maximum research level still produced exclusively comercial results. I know, I know. Nothing is perfect. But I think that the larger question is how much emphasis should we be putting on citation-based scoring (Google) vs. content. Do you think that we're crowding out good information from small contributors by poor information from larger contributors. So say that I publish a pretty effective methodology on calculating the ROI on search engine optimization. I won't show anywhere near the top if these other sites are publishing sites geared specifically to attain high rankings. If someone publishes a site on SEO that has 100 pages, thousands of occurances of SEO and dozens of occurances of ROI, and has his buddies link to him, my site might as well not exist based on current search algorithms. Is this how we should be treating the web's content? I don't see the argument that reference-based ranking works in the internet as it does in professional journals. For example, I wouldn't write a thousand articles in scientific journals citing my prize article just so that one would be more popular when searching Nexis-Lexis. In fact, one couldn't do it. There are certain parameters in that landscape that allow for the effectiveness of this method. I don't think that that maps to the internet.

[digitalghost]

>>I don't think that that maps to the internet.

Nor do I. Pagerank is flawed methodology for ranking pages. Yes, links were votes, and everyone soon started stuffing the ballot box with multiple votes. Enter the waning of the efficacy of Pagerank as a citation based ranking system. Enter new technologies, many still in their infancy, with regard to ranking a large, dynamic corpus of documents. There are many that will tell you that Pagerank died long ago.

Search technology is relatively new. Now that large dollars are involved, expect more innovations.

As for the eye-tracking study, their numbers don't match up to log stats tracked by position. So be it.

[edited by: digitalghost at 10:23 pm (utc) on May 27, 2005]

[Irie]

Regarding your last point, what log studies do you speak of. I'd be interested to know. Can you point me to the info?

Also, I'm trying my best to stay on top of evolving search engine technologies; can you point me to your favorite haunts?

[digitalghost]

The log studies are personal, #10 spot outperfoms 4-9, #3 spot converts best, 1-3 get the most traffic, better to be #11 than #10 though. Traffic isn't the sole aim. It's about traffic that converts.

As for search technologies, look for LSI and LSA. Latent semantic indexing and latent semantic analysis. Markov chains, singular value decomposition or SVD, eigen values and vectors, etc. That should get you started. IR is a large and growing field. Expect to see linguistics become more prominent. Multi-language concurrent search, etc. It's a world wide web. Search engines need to be polyglots.