feat : dev 파일들 배포에 추가 #1 (#4)

* fix : requirements.txt 수정

* feat : CI/CD 설정 및 프로토타입 세팅 (#2)

* feat : 프로토타입 추가 #1

* feat : 프로토타입 추가 및 환경 수정 #1

* feat : 테스트용 deploy 수정 #1

* feat : test #1

* feat : test22 #1

* feat : deploy 설정변경 #1

* fix : deploy 설정 변경 #1

* style : 주석 제거 #1

* style : 주석 제거 #1

* style : 주석 추가 제거 #1

Author: minhyuk2

.github/workflows/deploy.yml

@@ -1,12 +1,14 @@
 name: Deploy AI Server (FastAPI)
 
 on:
-  push:
-    branches: [ "dev" ]
+  pull_request:
+    types: [ closed ]
+    branches: [ main ]
 
 jobs:
   deploy-ai-server:
     runs-on: ubuntu-latest
+    if: github.event.pull_request.merged == true
 
     steps:
       - name: ✅ Checkout FastAPI 코드
@@ -28,5 +30,5 @@ jobs:
           key: ${{ secrets.EC2_KEY }}
           script: |
             cd ~/my-app
-            docker-compose pull ai-server
-            docker-compose up -d ai-server
+            docker compose pull ai-server
+            docker compose up -d ai-server

app/main.py

@@ -0,0 +1,11 @@
+from fastapi import FastAPI
+from routes import skt_classify
+
+app = FastAPI(title="SKT KoBERT Text Classification API with URL Extraction")
+
+# API 라우터 등록
+app.include_router(skt_classify.router, prefix="/api", tags=["SKT KoBERT Classification"])
+
+@app.get("/")
+def root():
+    return {"message": "FastAPI SKT KoBERT 문장 분류 API (URL 추출 포함)"}

app/models/kobert_classifier.py

@@ -0,0 +1,15 @@
+import torch
+import numpy as np
+from transformers import AutoModel, AutoTokenizer
+
+# KO-BERT 모델 로드
+model_name = "monologg/kobert"
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+model = AutoModel.from_pretrained(model_name)
+
+# 문장 임베딩 생성 함수
+def get_sentence_embedding(text: str):
+    inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
+    with torch.no_grad():
+        outputs = model(**inputs)
+    return outputs.last_hidden_state[:, 0, :].squeeze().numpy()

app/models/skt_kobert.py

@@ -0,0 +1,19 @@
+import torch
+from transformers import AutoTokenizer, AutoModel
+
+tokenizer = AutoTokenizer.from_pretrained("skt/kobert-base-v1")
+model = AutoModel.from_pretrained("skt/kobert-base-v1")
+
+def get_sentence_embedding(text):
+    inputs = tokenizer(
+        text,
+        return_tensors="pt",
+        padding=True,
+        truncation=True,
+        max_length=512
+    )
+    inputs.pop("token_type_ids", None)
+
+    with torch.no_grad():
+        outputs = model(**inputs)
+    return outputs.last_hidden_state[:, 0, :].squeeze().numpy()

app/models/url_classifier.py

@@ -0,0 +1,20 @@
+import re
+
+def separate_text_and_urls(text):
+    # URL 정규 표현식 패턴 (도메인과 경로 부분만 추출)
+    url_pattern = re.compile(r'https?://[a-zA-Z0-9.-]+\.[a-zA-Z]{2,6}(/\S*)?')
+
+    # URL 추출 (finditer 사용)
+    urls = [match.group() for match in url_pattern.finditer(text)]
+
+    # URL을 제외한 일반 텍스트 추출
+    text_without_urls = url_pattern.sub('', text).strip()
+
+    return text_without_urls, urls
+
+# 테스트
+sample_text = "이 영화는 너무 재미있어요! 예매는http://ticket.com 에서 할 수 있어요."
+text, urls = separate_text_and_urls(sample_text)
+
+print("📌 일반 텍스트:", text)
+print("🔗 추출된 URL:", urls)

app/routes/classify.py

@@ -0,0 +1,16 @@
+from fastapi import APIRouter
+from pydantic import BaseModel
+from typing import List
+from app.services.text_processing import classify_or_create_category, category_embeddings
+
+router = APIRouter()
+
+# 요청 데이터 형식 정의
+class SentenceRequest(BaseModel):
+    sentences: List[str]  # 문자열 리스트 받기
+
+# 문자열 입력 후 문장 자동 분류 API
+@router.post("/classify")
+async def classify_text(sentences: SentenceRequest):
+    categorized_results = classify_or_create_category(sentences.sentences)
+    return {"categorized_sentences": categorized_results, "category_list": list(category_embeddings.keys())}

app/routes/skt_classify.py

@@ -0,0 +1,13 @@
+from fastapi import APIRouter
+from pydantic import BaseModel
+from app.services.skt_text_processing import classify_paragraph
+
+router = APIRouter()
+
+class ParagraphRequest(BaseModel):
+    paragraph: str  # 단일 문단 받기
+
+@router.post("/classify_paragraph")
+async def classify_paragraph_api(request: ParagraphRequest):
+    result = classify_paragraph(request.paragraph)
+    return result

app/server.py

@@ -0,0 +1,5 @@
+import uvicorn
+
+if __name__ == "__main__":
+	# 서버 구동 이후 디버깅을 위해 reload=True 옵션을 추가한다
+    uvicorn.run("app.main:app", host="0.0.0.0", port=8000, reload=True)

app/services/skt_text_processing.py

@@ -0,0 +1,84 @@
+import re
+from sklearn.metrics.pairwise import cosine_similarity
+from typing import List, Dict
+import numpy as np
+from app.models.skt_kobert import get_sentence_embedding
+
+
+CATEGORY_LABELS = {
+    "여행": ["도쿄 여행", "유럽 배낭 여행", "국내 캠핑", "비행기 예약", "숙소 추천"],
+    "교통": ["지하철 환승", "고속열차 이용", "비행기 탑승", "대중교통 이용", "렌터카 예약"],
+    "쇼핑": ["면세점 할인", "백화점 세일", "전자제품 구매", "패션 브랜드 쇼핑", "기념품 구매"],
+    "음식": ["일본 라멘 맛집", "프랑스 빵 추천", "한식당 방문", "카페 탐방", "길거리 음식"],
+    "기타": ["문화 체험", "미술관 방문", "박물관 견학", "자연 탐방", "테마파크 방문"]
+}
+
+url_pattern = re.compile(r'https?://[a-zA-Z0-9./?=&_%:-]+')
+
+# ✅ 문장에서 URL을 추출하고 문장 내에서 분리하는 함수
+def extract_urls_from_sentences(sentence: str):
+    urls = url_pattern.findall(sentence)
+    text_without_urls = url_pattern.sub(' ', sentence).strip()
+    text_without_urls = re.sub(r'\s+', ' ', text_without_urls)
+    return text_without_urls, urls
+
+category_embeddings = {
+    category: np.mean([get_sentence_embedding(example) for example in examples], axis=0)
+    for category, examples in CATEGORY_LABELS.items()
+}
+
+def generate_new_category(text: str):
+    words = text.split()
+    for word in words:
+        if len(word) > 1:
+            return word
+    return "기타"
+
+def classify_paragraph(paragraph: str, threshold: float = 0.7):
+    global category_embeddings
+    sentences = paragraph.split("\n")
+    processed_sentences = []
+    paragraph_embedding = get_sentence_embedding(paragraph)
+
+
+    best_category = None
+    best_similarity = 0
+
+    for category, category_vector in category_embeddings.items():
+        similarity = cosine_similarity([paragraph_embedding], [category_vector])[0][0]
+        if similarity > best_similarity:
+            best_similarity = similarity
+            best_category = category
+
+    # ✅ 기존 카테고리에 없으면 "category": "no", 추천 카테고리 제공
+    if best_similarity < threshold:
+        recommend_category = generate_new_category(paragraph)
+        if recommend_category not in category_embeddings:
+            category_embeddings[recommend_category] = paragraph_embedding
+        return_category = "no"
+    else:
+        recommend_category = best_category
+        return_category = best_category
+
+
+    for sentence in sentences:
+        text_part, urls = extract_urls_from_sentences(sentence)
+        sentence_embedding = get_sentence_embedding(text_part) if text_part else None
+
+
+        if urls:
+            sub_category = "관련 링크"
+        else:
+            sub_category = recommend_category
+
+        processed_sentences.append({
+            "text": text_part if text_part else "URL 포함 문장",
+            "sub_category": sub_category,
+            "urls": urls if urls else None
+        })
+
+    return {
+        "category": return_category,
+        "recommend_category": recommend_category,
+        "sentences": processed_sentences
+    }

app/services/text_processing.py

@@ -0,0 +1,36 @@
+from sklearn.metrics.pairwise import cosine_similarity
+from typing import List, Dict
+import numpy as np
+from app.models.kobert_classifier import get_sentence_embedding
+
+# 기존 카테고리 저장소 (초기 없음)
+category_embeddings: Dict[str, List[np.ndarray]] = {}
+
+# 문장 분류 함수 (유사 카테고리 확인 및 생성)
+def classify_or_create_category(sentences: List[str], threshold: float = 0.7):
+    global category_embeddings
+    categorized_sentences = {}
+
+    for sentence in sentences:
+        sentence_embedding = get_sentence_embedding(sentence)
+
+        best_category = None
+        best_similarity = 0
+
+        # 기존 카테고리와 유사도 비교
+        for category, embeddings in category_embeddings.items():
+            similarities = cosine_similarity([sentence_embedding], embeddings).mean()
+            if similarities > best_similarity:
+                best_similarity = similarities
+                best_category = category
+
+        # 기존 카테고리 존재 시 추가, 없으면 새 카테고리 생성
+        if best_similarity >= threshold and best_category:
+            categorized_sentences.setdefault(best_category, []).append(sentence)
+            category_embeddings[best_category].append(sentence_embedding)
+        else:
+            new_category = f"Category_{len(category_embeddings) + 1}"
+            category_embeddings[new_category] = [sentence_embedding]
+            categorized_sentences.setdefault(new_category, []).append(sentence)
+
+    return categorized_sentences