Institute For The Future

A new kind of information model

 

I swung by the Metaweb offices in San Francisco a few weeks ago to hear them describe the technologies behind their project Freebase, released in March 2006. They also shared some ideas about where they see the project going and how they hope to further leverage external developers into making their killer apps.

 

Freebase is like Wikipedia for data. Here's an example use case: if you have the data about all the Nobel Prize winners and wanted to share this, you could try to put it into tables at Wikipedia. This would take a lot of manual effort and you would end up dismantling the data structure, putting the data into a page-link format that is not friendly for machines to read.

 

With Freebase, you can use a little script or an applet to upload the raw data, which can then be read by humans on their site or by computers through their API. It also eliminates redundancy - adding a person to a list of Nobel winners adds information on that person's page as well.

 

The intricacies of how they were able to make a giant reconfigurable data structure are also interesting. Unlike most databases - which consist of tables with rows and columns - Freebase is a graph database. This means the data is stored as relationships between nodes. The node for Albert Einstein is connected to the node for Nobel Prize winner, and the connection contains the year 1921 and the fact that he won in physics. Trying to attach a structure to all the world's knowledge in a table format would generate numerous problems - especially if you decide later to change the structure.

 

For more, you can read up about the technology at the Freebase blog.