Welcome to the Tweet Classifier repository!

This repository contains the source code that extracts the tweets of a specified user and uses a LSTM model to predict if said tweets contain manipulated information or if they are reliable.

Deep Learning Model used= LSTM

Twitter Authorisation= OAuth 1.0

pre_proc: Preprocessing/Cleaning the text.

pre_model: Loading the pre_trained GloVe(in this code 6B tokens, 200d vectors) and converting text into tokens.

model: Model construction and execution. (Computationally expensive. Code executed on VM of 16vCPUs and 60GB RAM)

CHIRP: OAuth Authorization and methods to extract tweets based on name or keywords and trends based on location.

Integrator: Loading saved model and predicting the probability of credibility of tweets extracted.

The trained model has the following metrics: Total Size of Dataset: 65698 records

Training Data: 45,988 records

           Accuracy Score  = 0.9911

           F1 Score        = 0.9908

           Precision Score = 0.9923

           Recall Score    = 0.9894

Testing Data: 19710 records

           Accuracy Score  = 0.9702

           F1 Score        = 0.9689

           Precision Score = 0.9763

           Recall Score    = 0.9616

In order to run this source code and create your own model from your dataset, please read the pre_proc.py file and perform modifications to suit your data sets.

For News articles:

Kaggle

• https://www.kaggle.com/c/fake-news/data
• https://www.kaggle.com/clmentbisaillon/fake-and-real-news-dataset

WOE Id

https://codebeautify.org/jsonviewer/f83352