Political Tweet Classification

Set-up

Local

1. Set up the environment:

   • make venv
   • make install

2. Run the project

   • source ./.venv/bin/activate
   • python src/main.py

Docker

1. Build and run the Docker container
   • make build
   • make run

Idea

Data

The folder data contains datasets used to train and test our ML Text Classification Process.

• tweets.csv is our dataset containing tweets with the respective user who posted them.
• tweets_text.txt only contains the text of each tweet independently from the user.
• labeled_tweets.txt is a first attempt to classify to which candidate the tweet refers.

ML

1. Identify the candidate to which the tweet refers.
   • Naive approach: if "Trump" in tweet
   • ML approach: train a text classification model to label the tweet: __label__trump, __label__harris
     • First attempt:
       1. Classified each tweet with ChatGPT
       2. Trained fastText fasttext.train_supervised. Check it here.
2. Assess whether the message, referring to either one of the candidate, is positive or negative.
   • Check the file.
   • For the experiment it has been used the base model twitter-roberta-base-sentiment-latest,
   • Which is easily viable trough the Python3 library tweetnlp.
   • However a more fine-tuned version of the model towards politic tweets could be used. Check it here.
3. Increase the score of the candidate with a positive tweet. Decrease the score of a candidate with negative tweet.