If you've been following my blog for any time at all, then you know that I am fascinated with search engines, ranking, algorithms and the like. It's my dream one day to design a better search engine, and I'm always tinkering and working on ideas to that end.
As I run queries on the major search engines these days, I'm finding that link-based ranking of pages has a major drawback: the most relevant results often don't make it to the top.
Let's take, for instance, the query "acne home remedies". Run that phrase through Google and you'll get back the results that are the best optimized (that is, that have the most in-bound links related to the query).
As of right now, the #1 ranking result in Google for "acne home remedies" is rather mediocre. You have to click onto a bunch of other pages linked to on that page to get to any real information. It's time consuming and difficult.
For me, the number one result would ideally contain a general summary of information related to the query. That is, "acne home remedies" should show pages that list a number of home remedies for acne on the ranking page -- not just links to other pages that talk about those remedies. And the top ranking pages should talk about a number of remedies, not just one. Also, the remedies that are talked about should be well known and referenced on other web pages so that I, the searcher, can have a reasonable amount of trust in the information.
How well-linked a page is should play a part, because those links help establish some authority for the page, but they should not be so strong a factor that the links cause mediocre pages to rank the way they do for a lot queries in Google and Yahoo and Bing these days.
So how do we get search results that use linking to help judge authority, but that contain solid information that is reasonably trustworthy?
That's the goal of my latest search engine, Shablast. The way it works is pretty simple, but very effective (in my opinion):
- Get the top ranking pages from a major search engine (in this case, Bing).
- Analyze each result to see what topics are being discussed on the ranking pages.
- Resort the results, showing the pages that touch on the greatest number of popular topics first.
- Filter out the obvious spam.
The results I'm seeing from this four step process are pretty good so far, but I need a lot more people to test it out and let me know what they think about it and where it falls short (which, no doubt, this early in the game it does for some queries).
So what I'd like for you to do is hop over to Shablast.com and run some subjects through it that you are familiar with and let me know if the results are good or poor, and why the results are good or poor. Keep in mind that Shablast is designed primarily for informational queries, so don't expect grand results when doing product-based searches.
You can post a comment here, or (preferably) you can go to the Shablast Forum and post a message with the keywords you searched and what was good or bad about the results returned by Shablast. I can then ask you questions (if needed) and refine the algorithm to improve the process.
If this is something you're interested in experimenting with, why not take a moment to hop over to Shablast and give it a go?
Thanks in advance, and please feel free to post your thoughts and questions in a comment below.