address-ner-ru

Address NER model to find address parts from string

https://huggingface.co/aidarmusin/address-ner-ru

Dataset

5K raw addresses dataset

90% for training and 10% for evaluation

Evaluation

Metric	Value
eval_overall_precision	0.9550486413955048
eval_overall_recall	0.9644308943089431
eval_overall_f1	0.9597168380246082
eval_overall_accuracy	0.9770456798596813
eval_Apartment_f1	0.9663865546218489
eval_Apartment_number	352
eval_Building_precision	0.8695652173913043
eval_Building_recall	0.9195402298850575
eval_Building_f1	0.8938547486033519
eval_Building_number	87
eval_Country_precision	0.9950738916256158
eval_Country_recall	0.9805825242718447
eval_Country_f1	0.9877750611246944
eval_Country_number	206
eval_District_precision	0.9562043795620438
eval_District_recall	0.9924242424242424
eval_District_f1	0.9739776951672863
eval_District_number	132
eval_House_precision	0.9702380952380952
eval_House_recall	0.9760479041916168
eval_House_f1	0.973134328358209
eval_House_number	501
eval_Region_precision	0.9826989619377162
eval_Region_recall	0.9861111111111112
eval_Region_f1	0.9844020797227037
eval_Region_number	288
eval_Settlement_precision	0.9599271402550091
eval_Settlement_recall	0.9547101449275363
eval_Settlement_f1	0.9573115349682106
eval_Settlement_number	552
eval_Street_precision	0.9424603174603174
eval_Street_recall	0.9615384615384616
eval_Street_f1	0.9519038076152305
eval_Street_number	494
eval_ZipCode_precision	0.9208211143695014
eval_ZipCode_recall	0.9235294117647059
eval_ZipCode_f1	0.9221732745961821
eval_ZipCode_number	340

Example

from transformers import pipeline
import torch
import logging

device = "cuda:0" if torch.cuda.is_available() else "cpu"
logging.info(f"using device: {device}")

address_ner_pipeline = pipeline("ner", model="aidarmusin/address-ner-ru", device=device)

address = "628672,,,, Автономный Округ Ханты-Мансийский Автономный Округ - Югра,, Г. Лангепас, Ул. Солнечная, Д.21"
entities = address_ner_pipeline(address)
print(entities)