Update api.py

api.py

@@ -215,6 +215,17 @@ def extract_main_query(text):
         last = re.sub(rf'\b(\w+){p}\b', r'\1', last)
     return last.strip()
 
+import re
+import requests
+import numpy as np
+from sklearn.feature_extraction.text import TfidfVectorizer
+from sklearn.metrics.pairwise import cosine_similarity
+
+# 1. 입력에서 핵심 쿼리 추출 (간단히 영어/한글만 남기기)
+def extract_main_query(text):
+    return re.sub(r'[^\w가-힣]', '', text.strip())
+
+# 2. 위키백과 요약 가져오기
 def get_wikipedia_summary(query):
     cleaned_query = extract_main_query(query)
     url = f"https://ko.wikipedia.org/api/rest_v1/page/summary/{cleaned_query}"
@@ -224,6 +235,45 @@ def get_wikipedia_summary(query):
     else:
         return "위키백과에서 정보를 가져올 수 없습니다."
 
+# 3. TextRank 요약기
+def textrank_summarize(text, top_n=3):
+    sentences = re.split(r'(?<=[.!?])\s+', text.strip())
+    sentences = [s.strip() for s in sentences if len(s.strip()) > 10]
+
+    if len(sentences) <= top_n:
+        return text  # 문장이 너무 적으면 원문 반환
+
+    vectorizer = TfidfVectorizer()
+    tfidf_matrix = vectorizer.fit_transform(sentences)
+    sim_matrix = cosine_similarity(tfidf_matrix)
+    np.fill_diagonal(sim_matrix, 0)
+
+    def pagerank(matrix, damping=0.85, max_iter=100, tol=1e-4):
+        N = matrix.shape[0]
+        ranks = np.ones(N) / N
+        row_sums = np.sum(matrix, axis=1)
+        row_sums[row_sums == 0] = 1  # NaN 방지용 처리
+        for _ in range(max_iter):
+            prev_ranks = ranks.copy()
+            for i in range(N):
+                incoming = matrix[:, i]
+                ranks[i] = (1 - damping) / N + damping * np.sum(incoming * prev_ranks / row_sums)
+            if np.linalg.norm(ranks - prev_ranks) < tol:
+                break
+        return ranks
+
+    scores = pagerank(sim_matrix)
+    ranked_idx = np.argsort(scores)[::-1]
+    selected_idx = sorted(ranked_idx[:top_n])
+    summary = ' '.join([sentences[i] for i in selected_idx])
+
+    return summary
+
+# 4. 전체 파이프라인
+def summarize_from_wikipedia(query, top_n=3):
+    raw_summary = get_wikipedia_summary(query)
+    return textrank_summarize(raw_summary, top_n=top_n)
+
 def simple_intent_classifier(text):
     text = text.lower()
     greet_keywords = ["안녕", "반가워", "이름", "누구", "소개", "어디서 왔", "정체", "몇 살", "너 뭐야"]
@@ -268,7 +318,7 @@ def respond(input_text):
         keyword = re.sub(r"(에 대해|에 대한|에 대해서)?\s*(설명해줘|알려줘|뭐야|개념|정의|정보)?", "", input_text).strip()
         if not keyword:
             return "어떤 주제에 대해 궁금한가요?"
-        summary = get_wikipedia_summary(keyword)
+        summary = summarize_from_wikipedia(keyword)
         return f"{summary}\n다른 궁금한 점 있으신가요?"
 
     return generate_text_topkp(model, input_text)