wehub-resource-sync
593b94c120
Twitter Bots Example
We'll be using the twitter human-bots dataset which is composed of 37438 rows each corresponding to a Twitter user account. Each row contains 20 feature columns collected via the Twitter API. These features contain multiple data modalities, including the account description and the profile image.
The target column account_type has two unique values: bot or human. 25013 user accounts were annotated as human accounts, the remaining 12425 are bots.
Preparatory Steps
Create and download your Kaggle API Credentials.
The Twitter Bots dataset is hosted by Kaggle, Ludwig will need to authenticate you through the Kaggle API to download the dataset.
Examples
Run python train_twitter_bots.py to train a single model.
For a faster, more lightweight model run python train_twitter_bots_text_only.py, which does not use image features.
This will download the Twitter Bots dataset into the current directory, train a model, and write results into the following directories:
./outputs/results/
api_experiment_run/
./outputs/visualizations/
confusion_matrix__account_type_top2.png
confusion_matrix_entropy__account_type_top2.png
learning_curves_account_type_accuracy.png
learning_curves_account_type_loss.png
After training, the script will generate the following plots:
