Condolence Models

Intro

condolence-models is a package used to detect condolence and distress expressions, as well as empathetic comments. It is released with the EMNLP 2020 paper Condolence and Empathy in Online Commmunities.

Install

Use pip

If pip is installed, question-intimacy could be installed directly from it:

pip3 install condolence-models

Dependencies

python>=3.6.0
torch>=1.6.0
pytorch-transformers
markdown
beautifulsoup4
numpy
tqdm
simpletransformers
pandas
numpy

Usage and Example

See example.py for an example of how to use the classifiers.

Note: The first time you run the code, the model parameters will need to be downloaded, which could take up significant space. The condolence and distress classifiers are about 500MB each, and the empathy classifier is about 1GB.

The interface for condolence and distress are the same. The interface for empathy is slightly different, to align with the simpletransformers interface more closely.

Classifying condolence or distress.

from condolence_models.condolence_classifier import CondolenceClassifier

cc = CondolenceClassifier()

# single string gets turned into a length-1 list
# outputs probabilities
print("I like ice cream")
print(cc.predict("I like ice cream"))
# [0.11919236]

# multiple strings
print(["I'm so sorry for your loss.", "F", "Tuesday is a good day of the week."])
print(cc.predict(["I'm so sorry for your loss.", "F", "Tuesday is a good day of the week."]))
# [0.9999901  0.8716224  0.20647633]

Classifying empathy.

from condolence_models.empathy_classifier import EmpathyClassifier
ec = EmpathyClassifier(use_cuda=True, cuda_device=2)

# list of lists
# first item is target, second is observer
# regression output on scale of 1 to 5
print([["", "Yes, but wouldn't that block the screen?"]])
print(ec.predict([["", "Yes, but wouldn't that block the screen?"]]))
# [1.098]

Contact

Naitian Zhou (naitian@umich.edu)