7Split - 10:55 pm on Aug 18, 2006 (gmt 0)

In search, relevance is everything – Duh! Given that Google’s market share has steadily increased over the past couple of years, (In July 60% of all searches were done through Google) it’s easy to assume that for now they offer the most relevant search solution.

Currently, Google’s ultimate goal is to find the most relevant answer for every possible search query. You can imagine the vast amounts of information they cull through to accomplish this task. We’re talking about trillions of possible queries, each of which must produce relevant results within seconds.

In fact, there are probably hundreds of factors that go into how Google determines relevancy for any given keyword. However, even though their search algorithm is very complex, it still only provides a one-size-fits-all level of relevancy.

Think about it, whether you are an 85 year old conservative grandmother or an 11 year punk rock skater kid - when you type a keyword into Google the same set of results will appear (barring which data center you reach – but let’s assume that the exact same data center is used). The Apparent flaw in this method of structuring results is that what’s relevant to an 85 year old grandmother might not be relevant to an 11 year old kid.

To correct this shortcoming we must step back from the current model of generating search results with computer algorithms, and focus more on human elements. Humans can be categorized in varying levels of complexity. From the broad, easily identifiable variables like sex, religion and income bracket to the less identifiable personality variables like whether someone is extroverted or introverted, whether they like cats, or whether they are afraid of scary movies.

Although there are millions of potential variables, identifying core variables and grouping them together allows us to model probable search results for any given persona - hence the term Persona Rank (I know… Yet another potential “PR” search related acronym – sorry, I couldn’t help it).

One foreseeable downside was that users simply want to be able to go to a site, do a search and get results. They don’t necessarily want to have to waste time filling out a personality profile, before a search utility works for them. The solution we found was to monitor personal search patterns in an unobtrusive manor so that we can gather the data from each user without having to ask each user to divulge that in-depth, personal information.

While this sounds simple in theory, it actually involves a rather complex set of data. Each persona contains a number of predefined characteristics, and each characteristic contains a varying level of intensity depending on the individual user’s preferences. In addition, any user may have multiple personas.

This is only the beginning of the complexity however. Once we attach personas to each user, we then must address the issue of ranking search results for each persona. For this we use what I call Reputation Scoring.

Instead of relying heavily on PageRank, as Google does, we rely on the Reputation Score of our users to determine which pages should rank well for each persona. For example, if an 18 year football jock persona types in the keyword “movie review” he would see a completely different set of search results than a 45 year old mother of 6.

By moving search relevance away from the current page based scoring system, we can present constantly evolving personalized search results for each individual user.

So, what do you think? Do I have a chance in the world of beating Google?

In any case, wish me luck. I’m going to need it.