pymetamap

Python wrapper around MetaMap. This will take a list of sentences and extract concepts using MetaMap then return them in the form of a list of Concept objects.

This is originally written by Anthony Rios.

I added some minor features and modifications that were necessary while I was working on my thesis using this library. I thought they might be useful for others too, so I committed my changes here.

How to Install

>>> python setup.py install

Example Usage

To use the package, you must have installed a MetaMap instance on your local environment. For instructions, please go to MetaMap. When using this wrapper, please start the SKR/Medpost Part-of-Speech Tagger Server in the terminal:

# <public_mm_dir>/bin/skrmedpostctl start

And then start the Word Sense Disambiguation Server:

# <public_mm_dir>/bin/wsdserverctl start

To start you must create a MetaMap instance from the pymetamap package.

>>> from pymetamap import MetaMap
>>> mm = MetaMap.get_instance('/opt/public_mm/bin/metamap12')

You must supply the metamap binary to get_instance() in order to extract concepts.

To extract concepts from a sentence use the extract_concepts() method. This method taks a list of sentences as input and will return a list of Concept objects.

>>> sents = ['Heart Attack', 'John had a huge heart attack']
>>> concepts,error = mm.extract_concepts(sents,[1,2])
>>> for concept in concepts:
...     print concept
Concept(index='1', mm='MM', score='14.64', preferred_name='Myocardial Infarction', cui='C0027051', semtypes='[dsyn]', trigger='["Heart attack"-tx-1-"Heart Attack"]', location='TX', pos_info='1:12', tree_codes='C14.280.647.500;C14.907.585.500')
Concept(index='2', mm='MM', score='13.22', preferred_name='Myocardial Infarction', cui='C0027051', semtypes='[dsyn]', trigger='["Heart attack"-tx-1-"heart attack"]', location='TX', pos_info='17:12', tree_codes='C14.280.647.500;C14.907.585.500')

This example shows two seperate concepts extracted via MetaMap from two different sentences (sentence 1 and sentence 2).

My Minor Modifications

As I was using this great python library, I realized that I needed a way to get the trigger terms as well as the indices of a trigger. So now you can get them by call these two methods in a Concept class get_pos_info() and get_trigger():

>>> for concept in concepts:
...     print concept.get_trigger()
...     print concept.get_pos_info()
Heart Attack
set([(0, 12)])
heart attack
set([(16, 28)])

Another modification is to automatically assign an id to each sentence. In the original implementation, sentences only get an id if you specify it by passing a list of ids, as shown above. Now you can simply pass a list of sentences as an argument. In this case, the id starts at 0.

>>> concepts,error = mm.extract_concepts(sents)
>>> for concept in concepts:
...     print(concept)
ConceptMMI(index='0', mm='MMI', score='14.64', preferred_name='Myocardial Infarction', cui='C0027051', semtypes='[dsyn]', trigger='["-- Heart Attack"-tx-1-"Heart Attack"-noun-0]', location='TX', pos_info='1/12', tree_codes='C14.280.647.500;C14.907.585.500')
ConceptMMI(index='1', mm='MMI', score='13.22', preferred_name='Myocardial Infarction', cui='C0027051', semtypes='[dsyn]', trigger='["-- Heart Attack"-tx-1-"heart attack"-noun-0]', location='TX', pos_info='17/12', tree_codes='C14.280.647.500;C14.907.585.500')
# the index here is the index of the sentence where the sentence is found

More Information

Licensed under Apache 2.0.

Written by Anthony Rios

Modified by Justin So