File size: 1,337 Bytes
d087963
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
---
language: sq
widget:
- text: "Varianti AY.4.2 është më i lehtë për t'u transmetuar, thotë Francois Balu, drejtor i Institutit të Gjenetikës në Londër."
---
# Albanian Named Entity Recognition (NER) Model
This model is the fine-tuned model of "bert-base-multilingual-cased" 
using the famous WikiANN dataset presented
in the "Cross-lingual Name Tagging and Linking for 282 Languages" [paper](https://aclanthology.org/P17-1178.pdf).
# Fine-tuning parameters:
```
task = "ner"
model_checkpoint = "bert-base-multilingual-cased"
batch_size = 8 
label_list = ['O', 'B-PER', 'I-PER', 'B-ORG', 'I-ORG', 'B-LOC', 'I-LOC']
max_length = 512 
learning_rate = 2e-5 
num_train_epochs = 3 
weight_decay = 0.01 
```
# How to use: 
```
model = AutoModelForTokenClassification.from_pretrained("akdeniz27/mbert-base-albanian-cased-ner")
tokenizer = AutoTokenizer.from_pretrained("akdeniz27/mbert-base-albanian-cased-ner")
ner = pipeline('ner', model=model, tokenizer=tokenizer, aggregation_strategy="first")
ner("<your text here>")
```
Pls refer "https://huggingface.co/transformers/_modules/transformers/pipelines/token_classification.html" for entity grouping with aggregation_strategy parameter.
# Reference test results:
* accuracy: 0.9719268816143276
* f1: 0.9192366826444787
* precision: 0.9171629669734704
* recall: 0.9213197969543148