Kumar Ramanathan is the CTO of Relona, a developer of Calculus Intent-Based Search Algorithm. Kumar took a few minutes to discuss Relona’s purpose and capabilities:

SEL: How did Relona come about?

KR: By late 2001, during the middle of the dot-com bust, I was convinced that search would continue to be the most important application on the web. So I spent 3 months trying to create a new search-algorithm that would be a significant improvement over existing technologies. The result of that effort is the “Perfect Search” algorithm. I distributed a paper about this algorithm to about 30 leading researchers in information retrieval around the world, and the response was very encouraging. About 20% of them wanted to join our advisory board. Perfect Search is a learning algorithm and eventually, over time, it learns from user feedback and converges towards near-perfect results. But it can work only if installed in a site with the kind of traffic that Yahoo or Google get. It couldn’t be used by a startup.

In 2005, I started work on an algorithm that would be an immediate improvement over existing search-engines, without relying on user-feedback or other social search techniques. This would have to be an algorithmic search which outperformed existing engines. The result of this effort is Relona Calculus.

So Relona now has two search algorithms - one a social learning system, and the other a pure algorithmic refinement system.

SEL: What would you consider to be Relona's primary purpose?

To help users find the information they need with the least effort.

SEL: Relona currently appears to be a "filter" on Yahoo results for the purpose of refinement. Are there applications for the software if MSN, Ask or Yahoo chooses not to integrate?

The Relona Calculus algorithm has wider applications than something like PageRank which really only works on the web. Our algorithm can be used for enterprise search, job-search, and a number of other applications where link-analysis cannot be used. However, the market for which we bring the most value is web-search. An improvement in quality of Yahoo's search-engine can produce a vast increase in its market-cap, in the range of billions of dollars. So this is where we are focused for now.

SEL: You had mentioned that you felt Relona could help MSN, Ask or Yahoo overtake Google. Do you feel that improvements in these long-tail-query SERPs are enough to increase any of these engines' market shares?

The difference in performance between Google/Yahoo/Ask/Microsoft is very small for the most popular queries. Most users will find it difficult to say that any engine does better than the other. But on long-tail queries, the difference is obvious.

A search-engine that performs well on long-tail queries will be perceived in the user-community as the "best". Since there are no switching costs, users tend to migrate to the engine that they consider "best". It is a matter of perception. Winning in the long-tail will help Yahoo/Ask/Microsoft become known as the technical-leader and the change in perception will automatically lead to gains in market-share.

In addition, Yahoo reported that queries are getting longer. So what is “long-tail” now may become mainstream soon. In 1998, the average search-query was 1.2 words long. By 2006, it had grown to 3.3 words.

SEL: How does Google's Universal Search affect Relona's offering?

Google's Universal Search is a way to integrate results from multiple narrow verticals (maps, books ...) into the main SERP. The difficult problem here is figuring out if the results from the vertical are really pertinent to the user's query. I am very excited by this development because Relona's Calculus algorithm does really well in this aspect. Given a query, we can very accurately figure out the relevance of any vertical database. Once we have a score about the relevance of the vertical, figuring out where to insert it within the main SERP is straightforward. So I see Google Universal Search as the validation of an additional opportunity for Relona. We can either help Google do this better, or we can help Google's competitors respond with a better universal search. [Here’s Kumar’s blog post concerning this topic]

SEL: Are there plans to increase Relona's capability to include images, audio or video?

Relona performs well on Audio and Video, but the market is still in its infancy. If you look at the success of YouTube and compare it to the failure of Google-Video, the difference was not in the search-algorithm. YouTube did better simply because it had a larger database of videos to select from - not because its search algorithm was better. We don't plan to enter such markets until it matures to the point where the algorithm can really make a large difference. But there are other markets such as job-search where we might look for partnerships.

SEL: Could you tell me how Relona differs from Ask3D?

Ask3D is somewhat similar to Google’s universal search. Instead of embedding the results from other vertical databases (images, video, encyclopedia) Ask places them on the right-hand-side. But since there are more verticals than space to fit them in on the side-bar, Ask has to decide which ones are more relevant for the query. Relona can help Ask make a more accurate choice of verticals.

Interestingly, Ask3D doesn’t enhance the results in any meaningful way if you enter long-tail (complex) queries. But for shorter queries, the snippets from Wikipedia are really useful.

SEL is proud to offer interviews from Search Leaders to our readers. If your company has something to say, please shoot me a note at jeremy.swiller at thinkpartnerhip dot com.