Paper Reading #25: Finding Your Way in a Multi-Dimensional Semantic Space with Luminoso

Comments
Jeremy Nelissen
Vince Kocks 

Reference Information
Title: Finding Your Way in a Multi-dimensional Semantic Space with Luminoso
Authors: Robert Speer, Catherine Havasi, Nichole Treadway, Henry Lieberman
When/Where: IUI 2010


Summary
This paper discussed Luminoso, a tool for helping researchers to visualize and understand a semantic space. The idea is to help researchers to easily see semantic patterns in the data. The user inputs a set of documents. One or more of the documents can be marked as canonical, which is used to test whether input documents agree with it semantically. The documents are analyzed using natural language patterns to draw general conclusions about the data, cluster the documents, and determine relations between documents.

Luminoso's depiction of a semantic network. Source: Paper.

Documents are represented with a sort of scatter plot, whereby each part of a document corresponds to a point in the space. The size of each point indicates the number of times that the item appears in the input and has a text label that describes a common feature or name of a document. The user can perform an action called grabbing that allows the user to select a point, change the projection of the data, and learn more about that point. The colors of the points indicate correlation to the grabbed point.


Discussion
I thought this application was really cool. It takes a concept that is sometimes difficult to understand (large amounts of data), and represents it in a visual way that is fairly intuitive to understand. I think using colors to determine correlation could be confusing if the user is not aware of the meaning or is color blind, and a user study would have been nice to see how easy to use this system actually is. However, I think that something like this could have many useful applications, such as allowing researchers to view correspondences between data that could have previously been missed.