udacityCapstone-FakeNewsDetector
A repository made for showing my Udacity Capstone Project, entirely realized on Amazon SageMaker. This was my very first time using SageMaker from end to end.
Structure
The folder "data" contains data downloaded from Kaggle: https://www.kaggle.com/clmentbisaillon/fake-and-real-news-dataset
In the Notebook "Data Exploration and S3 Integration" I am exploring the data and saving the cleaned data to S3.
Then in "Model Building", I tried a Naive Bayes approach which did not go well.
In "Text shaping for RNN and training", I used the tokenized articles to train a Bidirectional LSTM with single sigmoid output, which led me to a test accuracy of ~ 0.985.
In the source_train folder, training scripts for Naive Bayes and LSTM are provided.
The helper.py file contains a couple of useful (and non useful) functions.
The tokenizer.pkl file contains a pickled version of the fitted Keras tokenizer I used to preprocess data for the LSTM.