La magia NER nascosta

Progetto di TLN

Luca Demma

---

Instructions to run the code

1. Clone the repository
2. Install the following python packages:
   • pandas
   • tqdm
   • conllu
3. In the root folder of the project create the following directory tree structure (case sensitive), putting the train and test file from https://github.com/Babelscape/wikineural/tree/master/data/wikineural/en and https://github.com/Babelscape/wikineural/tree/master/data/wikineural/en in their respective folders:

data
├── outputs
│   ├── DECODING
│   │   ├── EN
│   │   └── IT
│   └── LEARNING
│       ├── EN
│       └── IT
├── wikineural_en
│   ├── test.conllu
│   └── train.conllu
└── wikineural_it
    ├── test.conllu
    └── train.conllu

4. Start the Training launching: learning.py LANG replacing LANG with IT or EN (if no argument is passed it defaults to IT). The results of the training will be saved in data/outputs/DECODING/LANG/ in CSV format
5. Launch the Decoding with: main.py LANG BASELINE replacing LANG with IT or EN and BASELINE with baseline to calculate the NER tags with the baseline method. To don't calculate the baseline don't pass any argument as BASELINE. The results will be saved as CSV files in data/outputs/DECODING/LANG
6. To get the NER tags of the 3 sentences present in the slides launch decode.py. The result will be printed to console in the CoNLL-U format

Project Report

Link to the report of the project