A monster Einstein to power your 'search'!

Type in a keyword in a search engine and your frustration will probably have you tearing your hair out because, of course, when you're searching for Java, the Indonesian island, you'll probably end up with Java, the computing language.

It is not funny, but it happens all the time. Search terminologies "are basically keyword search engines, not context-based search engines", says Syed Yasin, R&D project head of Bangalore-based Sobha Renaissance Information Technology (SRIT).

Searchers say: "This happens because the algorithm embedded in your search engine is unable to correlate the word with the context." An algorithm is a set of rules a search engine uses to rank the responses to a particular query.

Most search engines today face two problems - those of synonymy and those of polysemy. The first occurs when people use different words to search for the same object (say cell phone and mobile phone). The second occurs when the search engine is not able to differentiate between words with more than one meaning.

But these problems could soon be history. Artificial intelligence developed by SRIT will differentiate search results on the basis of the context rather than the exact keyword, thereby providing users with the flexibility to zero in on search information.

Latent Metonymical Analysis and Indexing (LMai), the new algorithm, uses mathematical techniques from the data and phrases throughout the documents to analyse and identify the contextual relationship between words automatically. It will pick up the exact matches the user is looking for as well as throw up search results which have a relationship with the keyword.

So, if you search, for instance, for "heart surgery", the system will throw up information on "open heart surgery", "minimal invasive heart surgery", "heart attack", "heart bypass surgery", "vascular surgery", "angioplasty" and "cardiac catheterization". Under the single decomposition (SDV) technique which deconstructs single terms, LMai can identify words that go together.

So, "mental health" - two words - will be identified by LMai as a single word and the search results will relate the keyword with other related words to throw up results like "homeless health concern", "teen health", "depression", "schizophrenia", "panic disorder", "anxiety" and "alcoholism", among others. "This is a powerful feature wherein a machine behaves like an expert, although not to an accuracy of 100 per cent," says Yasin.

Microsoft, Google and Yahoo! have been investing heavily on search-engine optimisation. Google, considered the best in this category, has huge spends for R&D, and even Microsoft has a budget of $1.1 billion for year-ending June 2007.

Will SRIT be wooing these competitors to sell the product, or market it on a licencing basis? "We have various options in place," says Madhu Nambiar, CEO of SRIT.

"A renowned US-based media house has offered to jointly market the product by setting up a special purpose vehicle. We may also consider partnering a third-party search engine, or sell them the algorithm since to start a search engine requires huge investment."

Meanwhile, it's filed a patent in the US for the technology. Google that!