BERT

# BERT를 활용한 감정 분류 Fine-tuning 예제
from transformers import BertTokenizer, BertForSequenceClassification
from transformers import Trainer, TrainingArguments
from datasets import load_dataset
import torch

# 1. 모델과 토크나이저 로드
model_name = "bert-base-uncased"
tokenizer = BertTokenizer.from_pretrained(model_name)
model = BertForSequenceClassification.from_pretrained(
    model_name,
    num_labels=2  # 긍정/부정 이진 분류
)

# 2. 데이터셋 로드 및 전처리
dataset = load_dataset("imdb")

def tokenize_function(examples):
    return tokenizer(
        examples["text"],
        padding="max_length",
        truncation=True,
        max_length=512
    )

tokenized_datasets = dataset.map(tokenize_function, batched=True)

# 3. 학습 설정
training_args = TrainingArguments(
    output_dir="./bert-imdb",
    num_train_epochs=3,
    per_device_train_batch_size=8,
    per_device_eval_batch_size=16,
    warmup_steps=500,
    learning_rate=2e-5,  # BERT 권장 학습률
    evaluation_strategy="epoch",
    save_strategy="epoch",
    load_best_model_at_end=True
)

# 4. Trainer로 Fine-tuning
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_datasets["train"],
    eval_dataset=tokenized_datasets["test"]
)

trainer.train()

# 5. 추론 예제
def predict_sentiment(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
    with torch.no_grad():
        outputs = model(**inputs)
    probs = torch.softmax(outputs.logits, dim=-1)
    return "긍정" if probs[0][1] > 0.5 else "부정"