Aim higher

Super search

Dutch start-up Semlab set out to make Internet searches more sophisticated with the help of its Swedish, Turkish and Lithuanian partners. Traders are already snapping up products resulting from Eurostars project SOLIM, boosting revenues for the companies involved.

Bram Stalknecht, chief executive of Dutch start-up Semlab, is one of the few people who believe we aren’t being bombarded with news in the internet age. “There is not really a data overload because there is a lot of repetition and the same news on all different outlets,” he says. “But what we need is more reliable and what I call ‘actionable’ data.” By actionable, Stalknecht means data that can serve the needs of his clients who range from traders at large banks to civil servants processing large volumes of documents.

Stalknecht realised that if you typed “Paris” into an ordinary search engine and wanted pictures of the French city some automated searches would accurately show Eiffel Tower images in the French capital but would also throw up a fair few “false” results like the Paris Las Vegas Hotel and desert images from the Wim Wenders film “Paris, Texas”. He wanted to produce a more sophisticated search system, adding key domain tags to the automated searches to “train” systems to get more right results and to not miss useful material. His quest became a Eurostar project, which the consortium estimates will make the companies €5 million to €10 million annually over the next 5 years and €10 million annually from year 5 to 8.

The research was a new departure for Amsterdam-based Semlab, which until then had focused on text searches since it specialises in internet language. Through SOLIM it explored picture searches, using the text names given to photographs, to assign better tagging to linguistic themes. It linked up with Swedish company Picsearch, which runs a database of 3 billion images. Picsearch offered Semlab its database, knowing a successful project would help it providing clients with even better search results.

With information published on the internet in all different languages, many internet users want to see other language results in their searches. Semlab asked Picsearch for Swedish text and also involved Latvian company Tilde Informacines Technologijos, based in Lithuania, which runs computer translation services. Turkish company AGMLab Information Technologies helped test the language models developed by Semlab.

The four SMEs working in four different countries, with pictures and texts in numerous different languages, used English as their common language. After a kick off meeting in Amsterdam to get to know each other face to face, the partners divided into working groups to tackle specific objectives and talked to each other via email and conference calls via Skype.

Stalknecht’s Semlab has begun incorporating the more sophisticated picture searches into services it offers to clients like U.S. bank JPMorgan. The new search system saves precious seconds. It could offer traders a faster link to a video on Bloomberg TV featuring European Central Bank head Mario Draghi, for instance, says Stalknecht: “A couple of seconds in financial services is an age and they could make millions out of it!”