Paper Categories Prediction

# Paper-Categories-Prediction

In this repository, I fine-tuned the BERT pre-trained to perform the multi-label classification task on the dataset that contains the abstract, titles, and subject areas for papers in arXiv.

Different experiments were tried:

- Using only "Abstracts" as single feature:

- with stemming.

- without stemming.

The training result shows that the loss function value without stemming is better than when doing stemming.

- Using two inputs "Abstracts" and "Titles" to the BERT.

The result shows that the two inputs give better accuracy than only a single feature.