News generation

Fake news generation using Markov chains (n-gram language modeling) and long-short term memory network (LSTM).

Usage

Install dependencies (keras, tensorflow, numpy, pandas):

# Upgrade your pip3
python3 -m pip install --upgrade pip

# Install deps
pip3 install -U -r src/requirements.txt

Then, clean up your CSV containing tweets (e.g., remove stopwords):

python3 src/data_processing <input.csv> <output_clean.csv>

If you want to run n-gram model, run the following command with the output csv file from the previous one:

python3 src/model_ngrams.py <output_clean.csv>

If you would like to run LSTM model based on characters, run the following the same way:

python3 src/model_lstm_chars.py <output_clean.csv>

These will either output results to command line, or generate a file called generated_text_rnn_chars.txt. Feel free to modify, play with the code, and contribute if you find a bug.

Project report

To see the report, click here.

Data clean-up

Code here

Examples of generated text

Using n-grams (markov chain)

Using LSTM