text-algorithms-benchmarking

Python implementation of multiple text as data methods and benchmarking on publicly available datasets. Below a brief description of the data and the main methods demonstrated.

Data

Economic Policy Uncertainty

• Paper
• Labeled articles available here

Methods

1. Dictionary methods

For gigantic corpora and interesting alternative to the implementation from this repository is flashtext

2. Logistic regression on a bag-of-words

3. Small language models (i.e. BERT and friends)

4. Large language models

Zero-shot and no fine-tuning.

5. Finetune open-source large language models

Llama 3