Applied-NLP-Project-IN4325

Authors: Daniël van Gelder ([email protected]) and Thomas Bos ([email protected])

Requirements:

• Python
• simpletransformers
• transformers
• numpy
• pandas
• jupyter notebook
• scikit-learn
• nltk
• gensim
• tqdm

Getting Started

1. Dataset and Baseline Model

The dataset for the FNC-1 challenge can be retrieved from the dataset repository. Place all dataset files in the /data/fnc-1/ directory. Download the baseline from the baseline repository and run the baseline model on the dataset. The code needs to be adapted so that it can output the results as a stance file. Change and add the following lines of code to the fnc_kfold.py file at line 97:

OUT_DIR = "DIRECTORY_TO_REPOSITORY_DATASET" # Change this to the directory where you want the predictions to be stored
df = pd.read_csv("fnc-1/competition_test_stances.csv", names=['Headline', 'Body ID', 'Stance'], header=0)
df['Stance'] = predicted
df.to_csv(OUT_DIR + "/baseline_output.csv")

2. Running ALBERT

The notebook albert_fnc1.ipynb containing further instructions can be opened in Google Colab, which was used to generate all our results regarding the use of ALBERT on the FNC-1 data set.

Contact:

If information is missing from this repository, please reach out to either of us so that we can clarify.