SiloBreaker was one of my favorites at DEMO this year.  Based on statistical relationships, they are the first company I’ve come across that is able to perform sophisticated visualization and text analytic-like functionality on large unstructured data sets like news articles.  They perform standard entity extraction to identify people as people, places as places and things as things that are statistically relevant to the topic being searched – all in real time. I’d like to contrast SiloBreaker with other solutions that are based on natural language processing. The challenge with these solutions is that while the actor-action-object relationships are highly relevant to complex search queries (like who is doing what, where), the solutions cannot scale in the open internet across broad topic sets. Canned taxonomies have to be customized to fit event definitions written by the user, or the results are poor. If no customization is employed, then indexing of topics is painful and inconsistent across a diverse set of users.

But, what really gets me excited about SiloBreaker is the way a user can manipulate the relationships on the entity map. This visualization tool itself is not rocket science, but a user can “trash” entities by simply dragging and dropping them into the bin, and see the reconstruction of the relationships on the fly. Just wait until you can add your own entities to the entity relationship map

SiloBreaker is 100% on the MSFT stack. The product was publically released in the fall of 2007 and is estimated to be growing at 100,000 users per month.  It’s free today, and I’m hooked. Our London team is working with them, so hats off to them for identifying this rising star.