Stanford OpenIE using customized NER model

Question

I am trying to use Stanford's OpenIE (version 3.6.0) to extract relation triples based on a NER model I trained in chemistry domain. However, I couldn't have OpenIE to extract relation triples based on my own NER model. It seems OpenIE extracts relation triples based only on the default NER models provided in the package.

Below are what I've done to train and deploy my NER model:

1. Train the NER model based on http://nlp.stanford.edu/software/crf-faq.html#a.

2. Deploy the NER model in CoreNLP server and then restart the server. I modified the props attribute in corenlpserver.sh. The props attribute now looks like below:

   props="-Dner.model=$scriptdir/my_own_chemistry.ser.gz,edu/stanford/nlp/models/ner/english.all.3class.distsim.crf.ser.gz"

Please take a look at an example NER + OpenIE results here. In this example, I expect that OpenIE builds the relation triples on the entities (such as Cl, Br, and Windjana) recoginized by my NER model, but it doesn't. Is it possible to have OpenIE extract relation triples based on a self-trained NER model? If so, would you please give me some breif instructions on how?

Thanks in advance!

Answer 1

Contacted the author of OpenIE, and the author confirmed that OpenIE more or less ignored NER altogether. Hope this can help others who have the same question.