I have been playing with a few old Power Macintosh computers I have aquired, and trying to install Linux on them (yes it is possible). Things were going ok, when I got an error citing "Kernel Panic."

So, being the good search researcher that I am, I copied the exact text of the error and went to Google to perform a search. I used quotes around the phrase, so that I got exact matching results, of which there were about 123,000.

In other words, there were 123,000 pages out there that have that exact error message. So I started browsing the listings, and they had a common theme - remove the memory cache and that will clear up your Kernel Panic error.

So I did that - I pulled out the motherboard, and removed the L2 cache memory module, just like the posts all stated. I then reinserted the board and restarted the install process and guess what? I got the same error.

I double checked the wording to make sure I had the write wording (which I did). So I went back to the search results to see if perhaps I had missed anything. It turns out that I didn't. The top 30 results all said roughly the same thing.

So, while on the surface, the results returned were relevant to the query entered, the results weren't relevant to my problem. I was not able to solve my problem given the set of results returned.

Therefore, the search returned non-relevant results.

And I think this is the key to the whole relevancy issue. Not whether the search engine can better match results because, lets face it, they are really good at matching results now and have been for some time. The engines need to develop an understanding of the query before they can determine relevancy. And that is the problem.

My query was 6 or 7 words long. And the search engine did match the phrase, but it wasn't able to understand what my problem was from that short query.

But for me to search better, I would have to explain my problem, yet I couldn't do that in the search box because it would be too long. This is where there tends to be a disconnect: When one turns to the web for answers, they have to figure out how to ask the right question.

I'm sure there is an answer out there to my problem, but I have to further refine my query into the proper question to ask. This means I have to find a way to supply the search engine with more words, so it can develop an understanding of what I want.

Microsoft's search query builder does start to do this, but it is extremely limited. In order for a search engine to properly provide an answer to a question, it needs to be able to refine within returned results.

Imagine that if I am having this much of a problem finding the information I want, and I am considered an "expert" on search engines, how is the average, or even new user finding information?

Which leads back to my original question - who (or what) defines relevancy, and how do they do this? Simply finding words near other words won't be a total solution, as you can see from my example, and even monitoring click throughs doesn't give enough information on whether returned search results are valid. Sure this could provide aggregate data, but I don't see it as a total solution. Because based on aggregate data, I am sure that most people found the listings I found about the kernel panic useful, yet I didn't.

While it may seem like a simple technological fix, I don't think it will be that easy. How can a search engine guess a person's mindset when they perform a query? How do they know what a person is searching for when they type in "MP3 players." Are they looking to buy? Do they want reviews? Do they merely want to know who makes them?

Maybe the search engines need to define a way to refine queries, just like they can correct spelling. By asking questions, they can help refine the search query, so that when a person does a search for "MP3 Players" the first response from the engine will be the start of something like 20 questions: "Are you looking to buy and MP3 player?" (yes/no) then logic can determine the next question based on the answer to the previous question.

This way, I could put in my kernel panic query and it could walk me through a refinement so that I could find the exact results I am looking for.

Rob Sullivan
Production Manager
Enquiro.com

Copyright 2004 - Searchengineposition Inc.