Repository for CS60075 Course Project
Link to Project Report: https://drive.google.com/file/d/1uTS5oGIhPWyqAVL7R44m1E0sGo0WDDWW/view?usp=share_link
Team Members:
| Name | Roll Number |
|---|---|
| Sagnik Chakraborty | 22CS60R08 |
| Tanishq Sharma | 22CS60R27 |
| Tarun Mohanty | 22CS60R28 |
| Vatsal Venkatkrishna | 19CH30037 |
The Legal_NER_DataFormatting.ipynb notebook contains the code to the required data format. The source data formatting is of the format:
"id": <ID>,
"annotations": [
{
"result": [
{
"value": {
"start": <start token 1>,
"end": <end token 1>,
"text": "<text 1>",
"labels": [
"<NER Label 1>"
]
},
},
{
"value": {
"start": <start token 2>,
"end": <end token 2>,
"text": "<text 2>",
"labels": [
"<NER Label 2>"
]
},
...
},
"data": {
"text": "<Full Text>"
},
"meta": {
"source": "<source>"
}
The annotation is performed using the Spacy tokenizer. We hence extract the text and tokenize it using spacy, after which we use BIO-tag based formatting to extract the NER labels corresponding to each token in the text.
We use the simpletransformers Python module to obtain the results for our tasks. Legal_NER_BERT.ipynb contains the code for obtaining the baseline results using the models described in the report, While Model_Fusing_Method_[1-3].ipynb contains the result for the three model fusion methods.