hnatt

HNATT

Our HNATT model is built to predict the user political leaning given a set of comments they have posted. Our code was modified from the inspiring hnatt code base.

Overview

HNATT is a deep neural network for document classification. It learns hierarchical hidden representations of documents at word, sentence, and document levels. At both the word and sentence levels, HNATT makes use of an attention mechanism, in which it learns a context vector that determines a relevance weighting for its learned encoding of words and sentences. This model has been shown to outperform hierarchical models without attention, indicating that learning an input-dependent weighting across the various substructures of the document leads to improved performance in classification tasks.

Content

filename	description
hnatt.py	Main HNATT implementation with custom Attention layer
util.data.py	Data loader of training and unseen data
util.text_util.py	Utility function for normalizing texts
util.glove.py	Utility function for loading GloVe embedding weights
train_hnatt_model.py	Train HNATT model with 5-fold cross-validation on 162,102 seed users
predict_unseen.py	Predict user leaning for 9M unseen users
predict_new_user.py	Predict user leaning for a new user
app/	A simple Flask app for exploring a trained HNATT, allowing you to easily make predictions based on a text input and visualize the resulting attention activations at both word and sentence levels

Get started

The HNATT code was developed and tested on Python 2.7. Follow the instructions below to create a new Python 2.7 environment:

conda create --name py2 python=2.7
conda activate py2
python -V
pip -V
pip install virtualenv
pip install -r requirements.txt

Next time, just run

conda activate py2

to activate the Python 2.7 environment.

Predict the political leaning of a new user

First, download our trained models and extract the files by

tar xvfj trained_HAN_models.tar.bz2
mv trained_HAN_models/* saved_models/

Next, put all extracted files into the directory saved_models, and load our trained models. We have five trained model cv1_hnatt_model to cv5_hnatt_model, you can design your own rule for aggregating the prediction results.

from hnatt import HNATT

SAVED_MODEL_DIR = 'saved_models'
saved_model_filename = 'cv1_hnatt_model.h5'
saved_tokenizer_filename = 'cv1_hnatt_model.h5.tokenizer'

# initialize HNATT
h = HNATT()
h.load_weights(SAVED_MODEL_DIR, saved_model_filename, saved_tokenizer_filename)

The input is a set of comments, we need do some text processing first. Basically, we remove all punctuations within a comment, and then concatenate all comments by dot.

import re
# a user posts several comments, each comment consists of several words
comments = ["It is what it is or they were nevertrumpists so they didn't like me and I don't know them",
            'vote biden to save American',
            'We love #joebiden2020 for President.',
            '#bluewave2020 BUT PREP FOR TRUMP PUTIN CIVIL WAR',
            'trumpvirus came from a sick fatass loser',
            'trump has changed his tune about the corona trumpvirus. because trumpist are kicking the bucket']

# remove punctuations from each comment, and then join them by dot
comments = [re.sub(r'[^\w\s]', ' ', x).strip() for x in comments]
comments = '. '.join(comments)

Now you're ready to classify the new set of comments:

prob_lib, prob_con = h.predict_texts(comments)[0]
if prob_con <= 0.05:
    print('Predicted leaning: Liberal, prob_lib: {0:.4f}, prob_con: {1:.4f}'.format(prob_lib, prob_con))
elif prob_con >= 0.95:
    print('Predicted leaning: Conservative, prob_lib: {0:.4f}, prob_con: {1:.4f}'.format(prob_lib, prob_con))
else:
    print('Predicted leaning: Unknown, prob_lib: {0:.4f}, prob_con: {1:.4f}'.format(prob_lib, prob_con))

The above example input will print

Predicted leaning: Liberal, prob_lib: 1.0000, prob_con: 0.0000

Visualizing attention

Once you train an HNATT model and save it locally using the saved_model_dir and saved_model_filename arguments to train, you can easily play with the saved model in an interactive web app by running the following:

python run_hnatt_viewer.py

You can then visit localhost:5000 to interact with your HNATT.