kenzy-2312 commited on
Commit
1a242c8
·
verified ·
1 Parent(s): 37b48df

Update services/academic_rag_service.py

Browse files
Files changed (1) hide show
  1. services/academic_rag_service.py +29 -10
services/academic_rag_service.py CHANGED
@@ -256,22 +256,41 @@ class AcademicRAGService:
256
  # ============================================================
257
  # 1️⃣ Prerequisite Chain
258
  # ============================================================
259
- if intent == "courses" and course_subintent == "prerequisite":
260
- print("🔄 [Recursive Retrieval] Fetching full prerequisite chain chunks...")
261
- primary_docs = self.ensemble_retriever.invoke(searched_query)[:6]
 
 
 
262
  combined_content = [doc.page_content for doc in primary_docs]
263
-
264
- extracted_codes = []
 
265
  for doc in primary_docs:
266
  matches = re.findall(r"02-24-\d+", doc.page_content)
267
- extracted_codes.extend(matches)
268
-
269
- for code in list(set(extracted_codes))[:3]:
270
- secondary_docs = self.vectordb.similarity_search(code, k=2)
 
 
 
 
 
271
  for s_doc in secondary_docs:
272
  if s_doc.page_content not in combined_content:
273
  combined_content.append(s_doc.page_content)
274
-
 
 
 
 
 
 
 
 
 
 
275
  return "\n\n".join(combined_content)
276
 
277
  # ============================================================
 
256
  # ============================================================
257
  # 1️⃣ Prerequisite Chain
258
  # ============================================================
259
+ # ============================================================
260
+ elif intent == "courses" and course_subintent == "prerequisite":
261
+ print("🔄 [Recursive Retrieval] Fetching comprehensive prerequisite chain chunks...")
262
+
263
+ # 1. جلب المستندات الأساسية للمادة المطلوبة (تقليل الـ k لـ 4 لضغط التوكنز وحماية الـ Rate Limit)
264
+ primary_docs = self.ensemble_retriever.invoke(searched_query)[:4]
265
  combined_content = [doc.page_content for doc in primary_docs]
266
+
267
+ # نستخدم set لتخزين الأكواد المكتشفة لمنع التكرار نهائياً
268
+ discovered_codes = set()
269
  for doc in primary_docs:
270
  matches = re.findall(r"02-24-\d+", doc.page_content)
271
+ discovered_codes.update(matches)
272
+
273
+ # 2. خطوة التتبع الأعمق: نمر على الأكواد المستخرجة ونبحث عن شروطها هي أيضاً
274
+ # نأخذ أول 4 أكواد مستخرجة لضمان جلب السلسلة كاملة
275
+ for code in list(discovered_codes)[:4]:
276
+ # 🚀 التعديل الجوهري: دمج الكود مع كلمة prerequisite لتوجيه قاعدة البيانات لجلب فقرة الشروط الحقيقية فقط
277
+ steering_query = f"{code} prerequisite requirement"
278
+ secondary_docs = self.vectordb.similarity_search(steering_query, k=2)
279
+
280
  for s_doc in secondary_docs:
281
  if s_doc.page_content not in combined_content:
282
  combined_content.append(s_doc.page_content)
283
+ # استخراج أكواد جديدة من البحث الثاني (لتوسيع الشجرة ديناميكياً)
284
+ new_matches = re.findall(r"02-24-\d+", s_doc.page_content)
285
+
286
+ # إذا وجدنا كوداً جديداً لم يكن معنا، نبحث عنه كدعم أخير لضمان اكتمال السلسلة
287
+ for n_code in new_matches:
288
+ if n_code not in discovered_codes:
289
+ deep_docs = self.vectordb.similarity_search(f"{n_code} prerequisite", k=1)
290
+ for d_doc in deep_docs:
291
+ if d_doc.page_content not in combined_content:
292
+ combined_content.append(d_doc.page_content)
293
+
294
  return "\n\n".join(combined_content)
295
 
296
  # ============================================================