In this version, the codebase has been refactored into a package-able state so it can be pushed to PyPi.
A port of the codebase from Python 2 to Python 3. The politeness classifier has been retrained and repickled using the modern versions of pickle, scikit-learn, scipy, numpy, and nltk.
The original training data from Stack Exchange and Wikipedia has been included in the /corpora/ folder at the root of this project. If you wish to retrain the model, you will either need to use /corpora/stack-exchange.annotated.csv and /corpora/wikipedia.annotated.csv in combination with the Stanford CoreNLP dependency parser to generate the documents in the expected format for /scripts/train_model.py. Because this is a royal pain, please feel free to use the preparsed files (linked to under Further Resources)! Simply download an extract these two files into /corpora/.
Python implementation of a politeness classifier for requests, based on the work described in:
A computational approach to politeness with application to social factors.
Cristian Danescu-Niculescu-Mizil, Moritz Sudhof, Dan Jurafsky, Jure Leskovec, Christopher Potts.
Proceedings of ACL, 2013.
We release this code hoping that others will use and improve on our work.
NOTE: If you use this API in your work please send an email to cristian@cs.cornell.edu so we can add you to our list of users. Thanks!
pip install politenessIn order to classify documents that have not been preprocessed, we rely on the Stanford CoreNLP for generating dependency parses.
However, this codebase does not come packaged with CoreNLP; you will need to download and run a CoreNLP server and tell the politeness API where it is. See the previous link for details on setting up a CoreNLP server. There are two ways to tell the politeness API where the server is running:
bash:
python3 main.py url -u some-url.org:1234python:
from politeness.helpers import set_corenlp_url
set_corenlp_url('some-url.org:1234')When you set the URL, it will persist until another call to politeness.helpers.set_corenlp_url() is made. To see what the current URL is run python3 main.py url -l.
$ python3 main.py --help
usage: main.py [-h] {train,predict,download,url} ...
Command line access to the Stanford Politeness API.
optional arguments:
-h, --help show this help message and exit
Commands:
{train,predict,download,url}
train Train politeness classifier.
predict Predict politeness of a sentence.
download Download training data.
url Set the URL for the Stanford CoreNLP Server.import politeness
from politeness.classifier import Classifier
cls = Classifier()
# String Input
cls.predict("This is a sample sentence for prediction.")
# File Input
cls.predict("sample_sentences.txt")
# Dictionary Input
data_dict = {'sentence': 'If yes would you please share it?',
'parses': ['ROOT(root-0, please-5)', 'dep(please-5, If-1)',
'dep(please-5, yes-2)', 'aux(please-5, would-3)',
'nsubj(please-5, you-4)', 'dobj(please-5, share-6)',
'dep(please-5, it-7)', 'punct(please-5, ?-8)']}
cls.predict(data_dict)- Info about Cristian & Mortiz' work: http://cs.cornell.edu/~cristian/Politeness.html
- A web interface to the politeness model: http://politeness.mpi-sws.org/
- The Stanford Politeness Corpus: http://cs.cornell.edu/~cristian/Politeness_files/Stanford_politeness_corpus.zip
- The Stanford Politeness Corpus as compressed JSON containing the tree and dependency parses used to train the model in version 2.00: Wikipedia (~2GB; ~8GB uncompressed). Stack Exchange (~4GB; ~16GB uncompressed).
For questions regarding versions 3.00 and 2.00, please email bsm9339@rit.edu (Benjamin Meyers) or nm6061@rit.edu (Nuthan Munaiah). Please direct questions regarding the port from Python2 to Python3 to Benjamin Meyers.
For questions regarding the implementation and the theory behind the politeness classifier, please email cristian@cs.cornell.edu (Cristian Danescu-Niculescu-Mizil) or sudhof@stanford.edu (Moritz Sudhof).