Collarity CompassT, launched Monday, is a search engine that automatically ranks search results based on an individual’s interests. By tracking the search terms, url choices, and selections of users and visitors, Collarity responds to a user’s ‘hot spot of interest’ (i.e. query) with sites visited by expert-users in that area.
Levy Cohen, CEO and founder of Collarity, was nice to answer some of my questions; but as to how Collarity determines which results are considered successful for a given expert-user’s search (versus sites they visit to evaluate, but decide are not useful/credible), he says that would dip into the secret sauce.
UPDATE: For some reason, some readers are having trouble posting comments to this post. Among them Mr. Cohen. So I’ve posted what he’d like to, below. From Levy Cohen, CEO of Collarity:
I certainly didn’t mean to be evasive or dismissive regarding your question above. The issue is that there is no short concise answer to the question, but I’m happy to attempt one.
I think your question was, in essence, how Collarity recognizes a “good or right” choice vs. a “bad or wrong” choice when developing a personal relevancy formula for a user.
Collarity is a learning process that eventually converges, after a number of samples, to what best fits a registered user’s contextual relationship with a given keyword. There are many dimensions of learning (single event, session level, personal history, community membership, and global) that are brought to bear in the process of distilling personal relevance. All of these dimensions audit the process in real-time. They determine what has “fit” in the past but, more importantly, they determine how a user’s tastes might be changing over time.
It is not so much a “what is right” thing. It is more of a “what is MORE right today” thing – there really is no wrong. Some things are just “more right” statistically over time. The interaction score is based on the confluence of those micro events (all supporting and testing each other), and the system basically “course corrects” with each event.
It seems pretty confusing and the results weren’t that good. I’ll stick with Google thanks.
I really don’t buy this. This is being said for two reasons: PR so everyone knows google is getting focused or to throw off competitors.
I certainly didn’t mean to be evasive or dismissive regarding your question above. The issue is that there is no short concise answer to the question, but I’m happy to attempt one.
I think your question was, in essence, how Collarity recognizes a “good or right” choice vs. a “bad or wrong” choice when developing a personal relevancy formula for a user.
Collarity is a learning process that eventually converges, after a number of samples, to what best fits a registered user’s contextual relationship with a given keyword. There are many dimensions of learning (single event, session level, personal history, community membership, and global) that are brought to bear in the process of distilling personal relevance. All of these dimensions audit the process in real-time. They determine what has “fit” in the past but, more importantly, they determine how a user’s tastes might be changing over time.
It is not so much a “what is right” thing. It is more of a “what is MORE right today” thing – there really is no wrong. Some things are just “more right” statistically over time. The interaction score is based on the confluence of those micro events (all supporting and testing each other), and the system basically “course corrects” with each event.
Hope that helps.
I checked Collarity. I did a few searches and found the search engine to be intuitive. The search box looks little more sophisticated than Google Suggest. But I am not impressed with the results. Most results had the search keyword in the title tag. May be it is because they still have to index more sites.
Collarity has a nice intuitive interface better than Google Suggest but the search results are not that impressive. May be they have not yet indexed many websites.
Why the heck I can not enter my website? Your captcha does not allow it. strage !!
http://www.xb6.org