On 12/13/2012 01:56 PM, Miller, Timothy wrote:
Any interest in generifying the relation extractor to make it easier to
instantiate all relation extractors as clearTK style annotators? This would
make it easier to share feature extractors, allow me to rewrite the coreference
code base, and make it easier to contribute to multiple projects without
learning a bunch of different implementations.
What do you think about starting a relation extractor component over at
OpenNLP? It should be generic in the style
of the other OpenNLP components and be customized during training time
via training data, configuration and extensions.
I already started to work on a relation extractor I would like to use in
one of my projects where we need to extract relations
in news articles.
We have a very basic relation extractor in the OpenNLP component style.
The training format encodes the relations in a
Penn Treebank style parse tree with additional relation nodes. The
current implementation can already produce a simple
relation extractor but is still very limited on the feature generation.
Anyway feature generation is usually easy to
implement when you know which features you want.
The code is currently not Open Source but if there is an interest to
work together on it we will publish it.
Jörn
