Spaces:
Sleeping
Sleeping
File size: 14,305 Bytes
57d41f3 af86bc3 57d41f3 9b280de 57d41f3 9541be4 cc86d95 9b280de 6c46177 9b280de cc86d95 9b280de cc86d95 57d41f3 9541be4 cc86d95 9541be4 cc86d95 9541be4 cc86d95 57d41f3 9b280de 57d41f3 9541be4 cc86d95 9541be4 cc86d95 6c46177 cc86d95 6c46177 cc86d95 57d41f3 af86bc3 9b280de af86bc3 9b280de af86bc3 9b280de af86bc3 9b280de af86bc3 57d41f3 6c46177 9541be4 57d41f3 9541be4 57d41f3 9541be4 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 | import os
import json
import re
UNKNOWN_FALLBACK = "Thank you for your inquiry. Unfortunately, I am unable to provide an answer to your question at this time. For accurate and up-to-date information, please contact the admissions office at admission@nu.edu.eg"
def _coerce_embedding(vec):
if hasattr(vec, "tolist"):
vec = vec.tolist()
if isinstance(vec, list):
return [float(v) for v in vec]
return None
# =========================
# Category Detection
# =========================
def detect_category(question: str) -> str:
"""Detect the category of a question based on keywords."""
query_lower = question.lower()
# Category keywords based on Categories.txt
category_keywords = {
"Admissions": ["apply", "admission", "accept", "requirements", "application", "enroll"],
"Fees": ["fee", "tuition", "cost", "payment", "credit", "price", "pay", "refund"],
"Academics": ["gpa", "grades", "scores", "grade", "cgpa", "dean"],
"Academic Advising": ["advisor", "track", "course", "major", "register", "summer course"],
"IT & Systems": ["portal", "moodle", "login", "system", "technical", "support"],
"Emails": ["email", "gmail", "outlook", "mail", "inbox", "address", "contact email"],
}
# Lightweight Arabic keyword support for common student queries.
arabic_category_keywords = {
"Admissions": ["تقديم", "قبول", "التحاق", "شروط", "متطلبات", "مستندات", "اوراق", "اختبار", "placement", "معادله", "تحويل"],
"Fees": ["رسوم", "مصاريف", "سعر", "تكلفة", "قسط", "ساعه", "ساعة", "credit", "tuition", "refund", "منحه", "منح"],
"Academics": ["معدل", "gpa", "cgpa", "درجات", "انسحاب", "drop", "withdraw", "حضور", "غياب", "اختبار", "امتحان"],
"Academic Advising": ["ادفيزور", "مرشد", "ارشاد", "مقررات", "تسجيل", "ماجور", "تخصص", "خطة", "تراك", "self service"],
"IT & Systems": ["مودل", "moodle", "بورتال", "بوابه", "بوابة", "سيستم", "تسجيل الدخول", "portal", "حساب", "تقني", "itsupport"],
"Emails": ["ايميل", "بريد", "outlook", "email", "admission@", "nu.edu.eg"],
}
# Count keyword matches for each category
category_scores = {}
for category, keywords in category_keywords.items():
score = sum(1 for keyword in keywords if keyword in query_lower)
if score > 0:
category_scores[category] = score
# Return category with highest score, or None if no matches
if category_scores:
return max(category_scores, key=category_scores.get)
for category, keywords in arabic_category_keywords.items():
score = sum(1 for keyword in keywords if keyword in query_lower)
if score > 0:
category_scores[category] = score
if category_scores:
return max(category_scores, key=category_scores.get)
return None
# =========================
# Retrieval (Chroma / Local)
# =========================
def retrieve_with_chroma(query_embedding, top_k=5, category_filter=None):
"""Retrieve relevant Q&A pairs from Chroma with optional category filtering."""
try:
import chromadb
# Try to get a ChromaDB client
client = None
CHROMA_PERSIST_DIR = os.getenv("CHROMA_PERSIST_DIR", "./chroma_db")
try:
client = chromadb.PersistentClient(path=CHROMA_PERSIST_DIR)
except Exception:
try:
from chromadb.config import Settings
client = chromadb.Client(Settings(persist_directory=CHROMA_PERSIST_DIR))
except Exception:
try:
client = chromadb.EphemeralClient()
except Exception:
return [], []
# Get qa_knowledge collection (stores JSON Q&A data)
col = None
try:
col = client.get_collection("qa_knowledge")
except Exception:
return [], []
# Apply category filter if provided (e.g., only Fees, Admissions, etc.)
if category_filter:
results = col.query(
query_embeddings=[query_embedding],
n_results=top_k,
where={"category": category_filter}
)
else:
results = col.query(
query_embeddings=[query_embedding],
n_results=top_k
)
docs = results.get("documents", [[]])[0]
metas = results.get("metadatas", [[]])[0]
return docs, metas
except Exception as e:
# If there's an error, print it for debugging and return empty
print(f"ChromaDB error: {e}")
return [], []
def initialize_chroma_from_json(embed_fn, collection_name="qa_knowledge"):
"""Ensure Chroma has indexed Q&A entries from data.json."""
try:
import chromadb
except Exception as e:
print(f"ChromaDB import error: {e}")
return False
cwd = os.getcwd()
json_path = os.path.join(cwd, "data.json")
if not os.path.exists(json_path):
print("Chroma init skipped: data.json not found")
return False
try:
CHROMA_PERSIST_DIR = os.getenv("CHROMA_PERSIST_DIR", "./chroma_db")
try:
client = chromadb.PersistentClient(path=CHROMA_PERSIST_DIR)
except Exception:
from chromadb.config import Settings
client = chromadb.Client(Settings(persist_directory=CHROMA_PERSIST_DIR))
# Always rebuild from current data.json to avoid stale vectors.
try:
client.delete_collection(collection_name)
except Exception:
pass
col = client.create_collection(collection_name)
with open(json_path, "r", encoding="utf-8") as f:
data = json.load(f)
ids = []
docs = []
metas = []
for idx, entry in enumerate(data):
qtext = str(entry.get("question", "")).strip()
atext = str(entry.get("answer", "")).strip()
if not qtext and not atext:
continue
qa_id = str(entry.get("id", idx + 1))
category = entry.get("category", "General")
ids.append(f"qa_{qa_id}_{idx}")
docs.append(f"Question: {qtext}\nAnswer: {atext}")
metas.append({
"source": json_path,
"qa_id": qa_id,
"category": category,
"question": qtext,
"answer": atext,
})
if not docs:
print("Chroma init skipped: no valid Q&A entries")
return False
embeddings = []
batch_size = 32
for i in range(0, len(docs), batch_size):
batch_docs = docs[i:i + batch_size]
batch_emb = embed_fn(batch_docs)
if not isinstance(batch_emb, list) or len(batch_emb) != len(batch_docs):
# Fallback to per-item embedding if backend returns unexpected shape.
batch_emb = [embed_fn([d])[0] for d in batch_docs]
for emb in batch_emb:
emb_vec = _coerce_embedding(emb)
if emb_vec is None:
print("Chroma init aborted: invalid embedding vector")
return False
embeddings.append(emb_vec)
if len(embeddings) != len(docs):
print("Chroma init aborted: embedding count mismatch")
return False
col.add(ids=ids, documents=docs, metadatas=metas, embeddings=embeddings)
print(f"Chroma initialized with {len(docs)} entries")
return True
except Exception as e:
print(f"Chroma initialization error: {e}")
return False
def local_retrieve(question, top_k=3, category_filter=None):
"""Keyword-overlap retrieval over JSON data on disk with optional category filtering."""
cwd = os.getcwd()
json_path = os.path.join(cwd, "data.json")
items = []
metadatas = []
if os.path.exists(json_path):
try:
with open(json_path, "r", encoding="utf-8") as f:
data = json.load(f)
for entry in data:
category = entry.get("category", "General")
# Apply category filter if provided
if category_filter and category != category_filter:
continue
qa_id = entry.get("id", "")
qtext = str(entry.get("question", "")).strip()
atext = str(entry.get("answer", "")).strip()
if qtext or atext:
combined = f"Question: {qtext}\nAnswer: {atext}"
items.append(combined)
# Store metadata with each item
metadatas.append({
"source": json_path,
"qa_id": str(qa_id),
"category": category,
"question": qtext,
"answer": atext
})
except Exception as e:
print(f"Error reading data.json: {e}")
items = []
metadatas = []
# Use items directly (already have metadata)
if not items:
return [], []
# Normalized keyword-overlap scoring over both question and answer fields.
q_tokens = set(_normalize_question(question).split())
if not q_tokens:
return [], []
scores = []
for meta in metadatas:
qtext = str(meta.get("question", ""))
atext = str(meta.get("answer", ""))
q_field_tokens = set(_normalize_question(qtext).split())
a_field_tokens = set(_normalize_question(atext).split())
# Weight question overlap higher than answer overlap.
q_overlap = len(q_tokens & q_field_tokens)
a_overlap = len(q_tokens & a_field_tokens)
score = (2.0 * q_overlap) + (1.0 * a_overlap)
scores.append(score)
ranked = sorted(enumerate(scores), key=lambda x: x[1], reverse=True)
top = [items[i] for i, s in ranked[:top_k] if s > 0]
top_meta = [metadatas[i] for i, s in ranked[:top_k] if s > 0]
return top, top_meta
def _normalize_question(text: str) -> str:
"""Normalize questions for deterministic exact-match lookup."""
text = (text or "").lower()
# Arabic orthographic normalization to improve matching robustness.
text = re.sub("[إأآا]", "ا", text)
text = re.sub("ى", "ي", text)
text = re.sub("ؤ", "ء", text)
text = re.sub("ئ", "ء", text)
text = re.sub("ة", "ه", text)
# Keep both Latin and Arabic word characters so exact match works bilingually.
return " ".join(re.findall(r"[\w\u0600-\u06FF]+", text, flags=re.UNICODE))
def local_exact_match(question, category_filter=None):
"""Return exact question match from data.json if available."""
cwd = os.getcwd()
json_path = os.path.join(cwd, "data.json")
if not os.path.exists(json_path):
return None, None
target = _normalize_question(question)
if not target:
return None, None
try:
with open(json_path, "r", encoding="utf-8") as f:
data = json.load(f)
candidates = []
for entry in data:
category = entry.get("category", "General")
if category_filter and category != category_filter:
continue
qtext = str(entry.get("question", "")).strip()
if _normalize_question(qtext) == target:
atext = str(entry.get("answer", "")).strip()
doc = f"Question: {qtext}\nAnswer: {atext}"
meta = {
"source": json_path,
"qa_id": str(entry.get("id", "")),
"category": category,
"question": qtext,
"answer": atext,
}
return doc, meta
candidates.append(entry)
# Near-exact fallback for paraphrases/translations.
target_tokens = set(target.split())
if not target_tokens:
return None, None
best_entry = None
best_score = 0.0
for entry in candidates:
qtext = str(entry.get("question", "")).strip()
q_norm = _normalize_question(qtext)
if not q_norm:
continue
q_tokens = set(q_norm.split())
if not q_tokens:
continue
overlap = len(target_tokens & q_tokens)
union = len(target_tokens | q_tokens)
score = overlap / max(union, 1)
if score > best_score:
best_score = score
best_entry = entry
if best_entry is not None and best_score >= 0.45:
qtext = str(best_entry.get("question", "")).strip()
atext = str(best_entry.get("answer", "")).strip()
doc = f"Question: {qtext}\nAnswer: {atext}"
meta = {
"source": json_path,
"qa_id": str(best_entry.get("id", "")),
"category": best_entry.get("category", "General"),
"question": qtext,
"answer": atext,
}
return doc, meta
except Exception as e:
print(f"Error in exact match lookup: {e}")
return None, None
# =========================
# Context Formatting
# =========================
def format_context(docs, _metas):
"""Format retrieved documents with metadata."""
formatted = []
for i, doc in enumerate(docs):
formatted.append(f"- {doc}")
return "\n".join(formatted)
# =========================
# Save Results (Optional)
# =========================
def save_result(query, answer, sources):
"""Save query results to a log file."""
os.makedirs("logs", exist_ok=True)
record = {
"query": query,
"answer": answer,
"sources": sources
}
with open("logs/history.jsonl", "a", encoding="utf-8") as f:
f.write(json.dumps(record, ensure_ascii=False) + "\n") |