By Carsten Kraus

With reports across the media that Google is about to transform its search engine by building in semantics, it seems that a technology that has been under the radar for some time will finally hit the mainstream. So this is the ideal time to consider what approach Google will be taking and compare it to other search solutions around.

Google’s aim is to endeavour to answer users’ questions rather than simply hunt down individual keywords, which it does at the moment. This will mean that Google searches will no longer simply throw up a list of blue links, instead, the top of the results page will be dotted with information that 'answers questions’ posed in searches.

From the small amount of information currently available, it would appear that Google is not building ontologies that are classic semantics, but more aligned with ‘Bayesian inference’. This refers to a search function that analyses how keywords appear in relation to each other within search phrases. This would facilitate faster searches than classic semantics allow, which would make sense considering the vast number of queries Google receives.

For this approach to work effectively, it requires a large ‘universe’ of information to draw upon for each search topic. Google has amassed a huge body of data (around 200 million entities at the last count), which will help it deliver more accurate search results. However, the system’s performance level is likely to fall when searching within niche topics for which Google’s body of information is not so developed.

If Google’s new search technology is taking this approach, like many basic semantic systems it is likely to also find it difficult to interpret the vagaries and idiosyncrasies of language and phrasing that users employ. True semantic search uses an inference engine that analyses its own knowledge repository or ontology to try to understand the words and draw its own conclusions, rather than simply trying to match the words that have been typed in.

If someone was searching for ‘Beach Holiday around Christmas’, for example, the inference here would be that the destination they are looking for should be warm. This conclusion would be drawn from analysis of its ontology. However, you might want take a break in Scotland over New Year, and be able to take long walks on a nearby beach. In which case you might search: ‘Scotland New Year with beach’. A standard inference engine is likely to reject such a request, because it would associate the term ‘beach’ with someone looking for a warm holiday – and it is never going to be ‘warm’ in Scotland over New Year.

So although Google’s new search system should be a major improvement over its current keyword approach, users could face problems searching for more specialist topics that don’t feature prominently within its data universe, while searches featuring real-life ambiguities are also unlikely to generate good results. <>The first problem can be solved by using a specialist search engine, such as one that could be present on a travel or other niche website. This is likely to have a larger universe of information on the specific topic than Google delivering more accurate search results.

Handling the second issue, search ambiguities, requires a more intelligent system, such as the “probabilistic” inference engine we have developed, and which is used in the travel sector to answer the kind of search queries that I previously mentioned. This scores rather than judges information, combining semantics with the fuzzy logic approach. Essentially, it allows users to type in a query in their own words, which the engine then analyses and acts on due to its higher level of understanding.

Research has shown that on travel websites, as soon as consumers understand how the technology works and that they can really type in what they want, it takes them one third of the time to find the right product compared to conventional search. Understandably this has big implications for e-commerce both in the travel sector and beyond. Unfortunately, it’s unlikely Google will be able to deliver such a function across its platform just yet.

About the author

Carsten Kraus is chief executive of Europe’s leading online search and navigation specialist FACT-Finder. Based in Pforzheim, Paris and London, FACT-Finder’s clients include Harvey Nichols, Kurt Geiger and Swarovski.

FACT-Finder