File size: 12,527 Bytes
841be26
 
 
 
 
 
 
 
 
 
 
 
03b034f
 
841be26
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2576589
 
841be26
 
07c826b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
841be26
07c826b
 
 
 
 
841be26
07c826b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
841be26
 
07c826b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
841be26
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be1bd4b
841be26
 
 
 
 
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
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
---
license: cc-by-4.0
language:
- az
base_model:
- FacebookAI/xlm-roberta-base
pipeline_tag: token-classification
tags:
- personally identifiable information
- pii
- ner
- azerbaijan
datasets:
- LocalDoc/pii_ner_azerbaijani
---


# PII NER Azerbaijani v2

**PII NER Azerbaijani** is a second version of fine-tuned Named Entity Recognition (NER) model (First version: <a target="_blank" href="https://huggingface.co/LocalDoc/private_ner_azerbaijani">PII NER Azerbaijani</a>) based on XLM-RoBERTa. 
It is trained on Azerbaijani pii data for classification personally identifiable information such as names, dates of birth, cities, addresses, and phone numbers from text.

## Model Details

- **Base Model:** XLM-RoBERTa
- **Training Metrics:**
- 
| Epoch | Training Loss | Validation Loss | Precision | Recall  | F1       |
|-------|----------------|------------------|-----------|---------|----------|
| 1     | 0.029100       | 0.025319         | 0.963367  | 0.962449| 0.962907 |
| 2     | 0.019900       | 0.023291         | 0.964567  | 0.968474| 0.966517 |
| 3     | 0.015400       | 0.018993         | 0.969536  | 0.967555| 0.968544 |
| 4     | 0.012700       | 0.017730         | 0.971919  | 0.969768| 0.970842 |
| 5     | 0.011100       | 0.018095         | 0.973056  | 0.970075| 0.971563 |



- **Test Metrics:**  

- **Precision:** 0.9760  
- **Recall:** 0.9732  
- **F1 Score:** 0.9746  


## Detailed Test Classification Report

| Entity              | Precision | Recall | F1-score | Support |
|---------------------|-----------|--------|----------|---------|
| AGE                 | 0.98      | 0.98   | 0.98     | 509     |
| BUILDINGNUM         | 0.97      | 0.75   | 0.85     | 1285    |
| CITY                | 1.00      | 1.00   | 1.00     | 2100    |
| CREDITCARDNUMBER    | 0.99      | 0.98   | 0.99     | 249     |
| DATE                | 0.85      | 0.92   | 0.88     | 1576    |
| DRIVERLICENSENUM    | 0.98      | 0.98   | 0.98     | 258     |
| EMAIL               | 0.98      | 1.00   | 0.99     | 1485    |
| GIVENNAME           | 0.99      | 1.00   | 0.99     | 9926    |
| IDCARDNUM           | 0.99      | 0.99   | 0.99     | 1174    |
| PASSPORTNUM         | 0.99      | 0.99   | 0.99     | 426     |
| STREET              | 0.94      | 0.98   | 0.96     | 1480    |
| SURNAME             | 1.00      | 1.00   | 1.00     | 3357    |
| TAXNUM              | 0.99      | 1.00   | 0.99     | 240     |
| TELEPHONENUM        | 0.97      | 0.95   | 0.96     | 2175    |
| TIME                | 0.96      | 0.96   | 0.96     | 2216    |
| ZIPCODE             | 0.97      | 0.97   | 0.97     | 520     |


### Averages

| Metric        | Precision | Recall | F1-score | Support |
|---------------|-----------|--------|----------|---------|
| **Micro avg** | 0.98      | 0.97   | 0.97     | 28976   |
| **Macro avg** | 0.97      | 0.96   | 0.97     | 28976   |
| **Weighted avg** | 0.98   | 0.97   | 0.97     | 28976   |


## A list of entities that the model is able to recognize.

```python
[
    "AGE",
    "BUILDINGNUM",
    "CITY",
    "CREDITCARDNUMBER",
    "DATE",
    "DRIVERLICENSENUM",
    "EMAIL",
    "GIVENNAME",
    "IDCARDNUM",
    "PASSPORTNUM",
    "STREET",
    "SURNAME",
    "TAXNUM",
    "TELEPHONENUM",
    "TIME",
    "ZIPCODE"
]

```

## Usage

To use the model for spell correction:

The model is trained to work with lowercase text. This code automatically normalizes the text. If you use custom code, keep this in mind.

```python
import torch
from transformers import AutoModelForTokenClassification, XLMRobertaTokenizerFast
import numpy as np
from typing import List, Dict, Tuple

class AzerbaijaniNER:
    def __init__(self, model_name_or_path="LocalDoc/private_ner_azerbaijani_v2"):
        self.model = AutoModelForTokenClassification.from_pretrained(model_name_or_path)
        self.tokenizer = XLMRobertaTokenizerFast.from_pretrained("xlm-roberta-base")
        
        self.model.eval()
        
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self.model.to(self.device)
        
        self.id_to_label = {
            0: "O",
            1: "B-AGE", 2: "B-BUILDINGNUM", 3: "B-CITY", 4: "B-CREDITCARDNUMBER",
            5: "B-DATE", 6: "B-DRIVERLICENSENUM", 7: "B-EMAIL", 8: "B-GIVENNAME",
            9: "B-IDCARDNUM", 10: "B-PASSPORTNUM", 11: "B-STREET", 12: "B-SURNAME",
            13: "B-TAXNUM", 14: "B-TELEPHONENUM", 15: "B-TIME", 16: "B-ZIPCODE",
            17: "I-AGE", 18: "I-BUILDINGNUM", 19: "I-CITY", 20: "I-CREDITCARDNUMBER",
            21: "I-DATE", 22: "I-DRIVERLICENSENUM", 23: "I-EMAIL", 24: "I-GIVENNAME", 
            25: "I-IDCARDNUM", 26: "I-PASSPORTNUM", 27: "I-STREET", 28: "I-SURNAME",
            29: "I-TAXNUM", 30: "I-TELEPHONENUM", 31: "I-TIME", 32: "I-ZIPCODE"
        }
        
        self.entity_types = {
            "AGE": "Age",
            "BUILDINGNUM": "Building Number",
            "CITY": "City",
            "CREDITCARDNUMBER": "Credit Card Number",
            "DATE": "Date",
            "DRIVERLICENSENUM": "Driver License Number",
            "EMAIL": "Email",
            "GIVENNAME": "Given Name",
            "IDCARDNUM": "ID Card Number",
            "PASSPORTNUM": "Passport Number",
            "STREET": "Street",
            "SURNAME": "Surname",
            "TAXNUM": "Tax ID Number",
            "TELEPHONENUM": "Phone Number",
            "TIME": "Time",
            "ZIPCODE": "Zip Code"
        }
    
    def predict(self, text: str, max_length: int = 512) -> List[Dict]:
        text = text.lower()
        
        inputs = self.tokenizer(
            text,
            return_tensors="pt",
            max_length=max_length,
            padding="max_length",
            truncation=True,
            return_offsets_mapping=True
        )
        
        offset_mapping = inputs.pop("offset_mapping").numpy()[0]
        
        inputs = {k: v.to(self.device) for k, v in inputs.items()}
        
        with torch.no_grad():
            outputs = self.model(**inputs)
            predictions = outputs.logits.argmax(dim=2)
        
        predictions = predictions[0].cpu().numpy()
        
        entities = []
        current_entity = None
        
        for idx, (offset, pred_id) in enumerate(zip(offset_mapping, predictions)):
            if offset[0] == 0 and offset[1] == 0:
                continue
                
            pred_label = self.id_to_label[pred_id]
            
            if pred_label.startswith("B-"):
                if current_entity:
                    entities.append(current_entity)
                
                entity_type = pred_label[2:]
                current_entity = {
                    "label": entity_type,
                    "name": self.entity_types.get(entity_type, entity_type),
                    "start": int(offset[0]),
                    "end": int(offset[1]),
                    "value": text[offset[0]:offset[1]]
                }
            
            elif pred_label.startswith("I-") and current_entity is not None:
                entity_type = pred_label[2:]
                
                if entity_type == current_entity["label"]:
                    current_entity["end"] = int(offset[1])
                    current_entity["value"] = text[current_entity["start"]:current_entity["end"]]
                else:
                    entities.append(current_entity)
                    current_entity = None
            
            elif pred_label == "O" and current_entity is not None:
                entities.append(current_entity)
                current_entity = None
        
        if current_entity:
            entities.append(current_entity)
        
        return entities
    
    def anonymize_text(self, text: str, replacement_char: str = "X") -> Tuple[str, List[Dict]]:
        entities = self.predict(text)
        
        if not entities:
            return text, []
        
        entities.sort(key=lambda x: x["start"], reverse=True)
        
        anonymized_text = text
        for entity in entities:
            start = entity["start"]
            end = entity["end"]
            length = end - start
            anonymized_text = anonymized_text[:start] + replacement_char * length + anonymized_text[end:]
        
        entities.sort(key=lambda x: x["start"])
        
        return anonymized_text, entities

    def highlight_entities(self, text: str) -> str:
        entities = self.predict(text)
        
        if not entities:
            return text
        
        entities.sort(key=lambda x: x["start"], reverse=True)
        
        highlighted_text = text
        for entity in entities:
            start = entity["start"]
            end = entity["end"]
            entity_value = entity["value"]
            entity_type = entity["name"]
            
            highlighted_text = (
                highlighted_text[:start] + 
                f"[{entity_type}: {entity_value}]" + 
                highlighted_text[end:]
            )
        
        return highlighted_text

if __name__ == "__main__":
    ner = AzerbaijaniNER()
    
    test_text = """Salam, mənim adım Əli Hüseynovdu. Doğum tarixim 15.05.1990-dır. Bakı şəhərində, 28 may küçəsi 4 ünvanında yaşayıram. Telefon nömrəm +994552345678-dir. Mən 4169741358254152 nömrəli kartdan ödəniş etmişəm. Sifarişim nə vaxt çatdırılcaq ?"""
    
    print("=== Original Text ===")
    print(test_text)
    print("\n=== Found Entities ===")
    
    entities = ner.predict(test_text)
    for entity in entities:
        print(f"{entity['name']}: {entity['value']} (positions {entity['start']}-{entity['end']})")
    
    print("\n=== Text with Highlighted Entities ===")
    highlighted_text = ner.highlight_entities(test_text)
    print(highlighted_text)
    
    print("\n=== Anonymized Text ===")
    anonymized_text, _ = ner.anonymize_text(test_text)
    print(anonymized_text)
```

```
=== Original Text ===
Salam, mənim adım Əli Hüseynovdu. Doğum tarixim 15.05.1990-dır. Bakı şəhərində, 28 may küçəsi 4 ünvanında yaşayıram. Telefon nömrəm +994552345678-dir. Mən 4169741358254152 nömrəli kartdan ödəniş etmişəm. Sifarişim nə vaxt çatdırılcaq ?

=== Found Entities ===
Given Name: əli (positions 18-21)
Surname: hüseynov (positions 22-30)
Date: 15.05.1990 (positions 48-58)
City: bakı (positions 64-68)
Street: 28 may küçəsi (positions 80-93)
Building Number: 4 (positions 94-95)
Phone Number: +994552345678 (positions 132-145)
Credit Card Number: 4169741358254152 (positions 155-171)

=== Text with Highlighted Entities ===
Salam, mənim adım [Given Name: əli] [Surname: hüseynov]du. Doğum tarixim [Date: 15.05.1990]-dır. [City: bakı] şəhərində, [Street: 28 may küçəsi] [Building Number: 4] ünvanında yaşayıram. Telefon nömrəm [Phone Number: +994552345678]-dir. Mən [Credit Card Number: 4169741358254152] nömrəli kartdan ödəniş etmişəm. Sifarişim nə vaxt çatdırılcaq ?

=== Anonymized Text ===
Salam, mənim adım XXX XXXXXXXXdu. Doğum tarixim XXXXXXXXXX-dır. XXXX şəhərində, XXXXXXXXXXXXX X ünvanında yaşayıram. Telefon nömrəm XXXXXXXXXXXXX-dir. Mən XXXXXXXXXXXXXXXX nömrəli kartdan ödəniş etmişəm. Sifarişim nə vaxt çatdırılcaq ?
```


## CC BY 4.0 License — What It Allows

The **Creative Commons Attribution 4.0 International (CC BY 4.0)** license allows:

### ✅ You Can:
- **Use** the model for any purpose, including commercial use.
- **Share** it — copy and redistribute in any medium or format.
- **Adapt** it — remix, transform, and build upon it for any purpose, even commercially.

### 📝 You Must:
- **Give appropriate credit** — Attribute the original creator (e.g., name, link to the license, and indicate if changes were made).
- **Not imply endorsement** — Do not suggest the original author endorses you or your use.

### ❌ You Cannot:
- Apply legal terms or technological measures that legally restrict others from doing anything the license permits (no DRM or additional restrictions).


### Summary:
You are free to use, modify, and distribute the model — even for commercial purposes — as long as you give proper credit to the original creator.


For more information, please refer to the <a target="_blank" href="https://creativecommons.org/licenses/by/4.0/deed.en">CC BY 4.0 license</a>.


## Contact

For more information, questions, or issues, please contact LocalDoc at [[email protected]].