Chris Bunk.com

I've had the opprotunity to evaluate Starlight for the NASA Assurance Technology Center and so far I have to say overall I'm impressed. Starlight combines features that work on structured fields and unstructured text. It has features including text clustering, catagorical breakdowns, relationship mapping, hierarchical breakdowns, and spatial mapping. Now I know what your saying, "none of these graphical views are anything new." Well your right but these features aren't what makes Starlight stand out from the crowd, its the software's user interface and integration among these different models.

Let me use an example use of the software to best illustrate why this software is so cool. Lets take a set of safety incicent reports and their corrective actions and want to determine what I should focus my resources on next year to reduce safety incidents. I am given a dirty data set that has some structured fields, some large unstructured fields, and some documents (word, pdfs, etc.) associated with individual records. Once I get the data into Starlight (more on this later) I proceed to see a set of spaced out dots on a three dimensional grid. First I decide to change the color of these dots(records) based on a structure field of severity(determined by injury and cost). I then change the shape of dots based on another structured field of in (vehicle incident, workplace violence, etc.) I then text cluster the records to point out areas where words and phrases commonly appear together. I notice a cluster of words with terms like ladder, scafolding and climbing. I then select and look at these records. Quickly I switch to a view where I view how the subset of the data's structured fields distribute. I can setup a timeline and watch as time moves on how this distribution changes. I notice that there has been an increasing amount of incidents in this subset in the last three months at buildings 6 and 8 at the green river facility. I proceed to further filter my data down to these facilities and switch back to my text clustering view. I perform another clustering. I then can look into these individual records and gain some information on the nature of these incidents. I now have an area of focus for my funding next year.

Starlight has some cons however. The data needs to be in XML format which means some preproccessing. Starlight gives you tools to help facilitate this but it still takes time. Another neggative is the limit of data records it can handle at one time. I seem to remember it being 100,000. They mentioned their next version would be able to support more.

Overall the best tool I have scene for combining together these preexisting visualization methods in a three dimensional very efficient user interface.