DUALIST: a user-friendly active learning + annotation tool

[Burr Settles]

Greetings, all.

I have released an open-source active learning tool called DUALIST
based on research from an upcoming EMNLP paper, "Closing the Loop:
Fast, Interactive Semi-Supervised Annotation With Queries on Features
and Instances." You can watch a demo video and/or download it here:

http://code.google.com/p/dualist/

In user experiments, we often saw DUALIST achieve 90% of state-of-the-
art accuracy with only a few minutes of human effort for several
natural language tasks (mostly benchmark text classification corpora).
I'm hoping the tool works equally as well "in the wild," so let me
know your experiences if you try and adopt it in practice. If you are
doing research in a similar vein, the user interface could be a nice
framework for you to "plug in" your learning algorithms, as well.

DUALIST combines active learning, semi-supervised learning, and
learning with domain knowledge (in the form of feature labels/
queries). I'm also co-organizing an ICML 2011 workshop on the topic of
combining learning strategies like this, so please consider joining
the discussion if you will be in Bellevue next month:
https://sites.google.com/site/comblearn/

Best,
Burr Settles
http://www.cs.cmu.edu/~bsettles/