language: ko license: mit tasks: - text-classification tags: - intent-classification - korean - koelectra - fine-tuned

koElectra_shopping_intent_v2

한국어 의도 분류(Intent Classification) 모델입니다. koElectra-base-v3-discriminator를 기반으로 파인튜닝했습니다.

모델 개요

기본 모델: monologg/koElectra-base-v3-discriminator
작업: 텍스트 분류 (Intent Classification)
언어: Korean (한국어)
클래스 개수: 39개

성능 지표 (Performance)

전체 성능

정확도 (Accuracy): 0.9907 (99.07%)

상세 리포트

                             precision    recall  f1-score   support

           check_keep_login     0.9730    0.8182    0.8889        44
      check_terms_agreement     0.8431    0.9773    0.9053        44
         click_app_download     1.0000    1.0000    1.0000        44
                 click_cart     1.0000    1.0000    1.0000        44
      click_change_password     1.0000    1.0000    1.0000        44
   click_check_id_duplicate     1.0000    1.0000    1.0000        44
       click_delete_account     1.0000    1.0000    1.0000        44
              click_find_id     1.0000    1.0000    1.0000        44
        click_find_password     1.0000    1.0000    1.0000        44
           click_go_coupang     0.9333    0.9545    0.9438        44
                click_login     1.0000    1.0000    1.0000        44
               click_logout     1.0000    1.0000    1.0000        44
              click_my_page     1.0000    1.0000    1.0000        44
         click_order_detail     1.0000    1.0000    1.0000        44
         click_product_view     1.0000    1.0000    1.0000        44
click_shared_shopping_entry     1.0000    1.0000    1.0000        44
               click_signup     0.9778    1.0000    0.9888        44
           click_view_terms     0.9778    1.0000    0.9888        44
                 go_coupang     0.9535    0.9318    0.9425        44
                  go_hearbe     1.0000    1.0000    1.0000        44
                    go_mall     1.0000    1.0000    1.0000        44
           go_order_history     1.0000    1.0000    1.0000        50
                input_email     1.0000    1.0000    1.0000        44
                   input_id     1.0000    1.0000    1.0000        44
                 input_name     1.0000    1.0000    1.0000        44
             input_password     1.0000    1.0000    1.0000        44
     input_password_confirm     1.0000    1.0000    1.0000        44
         input_phone_number     1.0000    1.0000    1.0000        44
read_available_marketplaces     1.0000    0.9773    0.9885        44
  read_current_page_actions     1.0000    1.0000    1.0000        44
     read_frequent_products     1.0000    1.0000    1.0000        50
          read_hearbe_guide     1.0000    1.0000    1.0000        44
  read_order_history_recent     1.0000    1.0000    1.0000        50
                  read_page     0.9362    1.0000    0.9670        44
  read_recommended_products     1.0000    1.0000    1.0000        50
                 read_terms     1.0000    0.9773    0.9885        44
              submit_signup     1.0000    0.9545    0.9767        44
         uncheck_keep_login     1.0000    1.0000    1.0000        44
                    unknown     1.0000    0.9971    0.9986       350

                   accuracy                         0.9907      2046
                  macro avg     0.9896    0.9894    0.9892      2046
               weighted avg     0.9913    0.9907    0.9907      2046

학습 설정 (Training Configuration)

하이퍼파라미터

Learning Rate: 2e-05
Train Batch Size: 256
Eval Batch Size: 256
Epochs: 15
Weight Decay: 0.01
Label Smoothing: 0.1
Evaluation Strategy: IntervalStrategy.STEPS

학습 환경

최대 시퀀스 길이: 64 tokens
저장 전략: SaveStrategy.STEPS
학습 로그: ./logs

클래스 정보 (Labels)

총 39개의 의도 클래스: unknown, go_hearbe, go_coupang, read_current_page_actions, read_hearbe_guide, go_mall, click_shared_shopping_entry, click_app_download, input_id, input_password, check_keep_login, uncheck_keep_login, click_login, click_find_id, click_find_password, click_signup, click_check_id_duplicate, input_name, click_view_terms, read_terms, check_terms_agreement, submit_signup, input_phone_number, input_password_confirm, input_email, read_available_marketplaces, click_logout, click_go_coupang, click_cart, click_my_page, read_page, click_change_password, click_delete_account, click_order_detail, click_product_view, go_order_history, read_order_history_recent, read_frequent_products, read_recommended_products

사용 방법 (Usage)

from transformers import pipeline

# 모델 로드
classifier = pipeline("text-classification", model="your-username/model-name")

# 추론 실행
text = "당신의 문장을 입력하세요"
result = classifier(text)
print(result)

학습 데이터

데이터셋: 한국어 쇼핑 의도 분류 데이터
테스트 데이터: 평가 메트릭 계산에 사용

주의사항 (Limitations)

학습 데이터와 유사한 도메인(쇼핑 관련 텍스트)에서 최적의 성능을 보입니다.
다른 도메인의 데이터에서는 성능이 저하될 수 있습니다.
테스트 데이터에서의 성능이 실제 운영 환경과 다를 수 있습니다.

생성 메타데이터 (Metadata)

생성 날짜: 2026-02-12 10:25:18
Fine-tuned from: monologg/koElectra-base-v3-discriminator

koElectra_shopping_intent_v2

This model is a fine-tuned version of monologg/koElectra-base-v3-discriminator on the custom-intent-dataset dataset. It achieves the following results on the evaluation set:

Loss: 1.1867
Accuracy: 0.9902
F1: 0.9902

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 256
eval_batch_size: 256
seed: 42
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 0.1
num_epochs: 15
mixed_precision_training: Native AMP
label_smoothing_factor: 0.1

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
No log	0	0	3.6648	0.0249	0.0078
3.4909	1.5625	50	3.3917	0.1745	0.0564
3.0085	3.125	100	2.8914	0.5367	0.4781
2.5934	4.6875	150	2.4203	0.8397	0.8110
2.2220	6.25	200	2.0162	0.9301	0.9194
1.8962	7.8125	250	1.7029	0.9697	0.9682
1.6778	9.375	300	1.4734	0.9873	0.9872
1.4997	10.9375	350	1.3222	0.9907	0.9907
1.3948	12.5	400	1.2313	0.9902	0.9902
1.3469	14.0625	450	1.1867	0.9902	0.9902

Framework versions

Transformers 5.1.0
Pytorch 2.9.1+cu128
Datasets 4.5.0
Tokenizers 0.22.2

Downloads last month: 10

Safetensors

Model size

0.1B params

Tensor type

F32