Update app.py

app.py

@@ -539,8 +539,6 @@ def clean_item_query(text: str) -> str:
         "عنده", "عندها",
         "فيه", "فيها",
         "بيقدم", "بتقدم"
-        "نفسي", "عندو",
-
     ]
 
     for w in remove_words:
@@ -648,11 +646,11 @@ def name_search(query, top_k=10, category=None):
     matched = df[df["name_score"] > 0].copy()
 
     if category and not matched.empty:
-        cat_c     = clean_text(category)
-        cat_match = matched[matched["category_clean"].astype(str).str.contains(
-                        re.escape(cat_c), na=False)]
-        if not cat_match.empty:
-            matched = cat_match
+        target_cat = normalize_category(category)
+        cat_match = matched[matched["category"].astype(str).apply(lambda x: _category_match_value(x, target_cat))]
+        if cat_match.empty and "category_clean" in matched.columns:
+            cat_match = matched[matched["category_clean"].astype(str).apply(lambda x: _category_match_value(x, target_cat))]
+        matched = cat_match
 
     return matched.sort_values("name_score", ascending=False).head(top_k)
 
@@ -1059,11 +1057,11 @@ def build_response(place, intent, category=None, total_results=1):
 
 def handle_detail(text, place):
     if not place:
-        return "مش فاكر إحنا اتكلمنا عن مكان. ممكن تسألني من الأول؟"
+        return "تقصد أنهي مكان؟ اكتب اسم المكان أو اسألني الأول عن مطعم/صيدلية/كافيه. 😊"
     t    = norm(text)
     name = place.get("name", "المكان")
 
-    if any(w in t for w in ["بيفتح","بتفتح","مواعيد","امتى","امتي","بيقفل","بتقفل"]):
+    if any(w in t for w in ["بيفتح","بتفتح","مواعيد","مواعيده","مواعيدها","امتى","امتي","بيقفل","بتقفل"]):
         h = place.get("opening_hours","")
         if not h or str(h).strip() in ("","nan","none"):
             return f"معندناش معلومات عن مواعيد *{name}*."
@@ -1071,6 +1069,15 @@ def handle_detail(text, place):
         status = "🟢 مفتوح دلوقتي." if on == 1 else ("🔴 مغلق دلوقتي." if on == 0 else "")
         return f"{fmt_hours(h)}{status}"
 
+    # Phone must be checked before address because queries like "فين رقم تليفونه؟" contain "فين".
+    if any(w in t for w in ["تليفون","تلفون","رقم","رقمه","رقمها","phone","اتصل"]):
+        ph = place.get("phone","")
+        if isinstance(ph, (list, tuple)):
+            ph = ", ".join(str(x) for x in ph if x)
+        elif isinstance(ph, str) and ph.startswith("[") and ph.endswith("]"):
+            ph = ph.strip("[]").replace("'", "").replace('"', "").strip()
+        return f"📞 {name}: {ph}" if ph and ph not in ("","nan") else f"مش عندنا رقم *{name}*."
+
     if any(w in t for w in ["عنوان","فين","فينه","فينها","مكانه","مكانها","العنوان",
                              "لوكيشن","الموقع","ازاي اروحه","ازاي اروح","map","الخريطة"]):
         addr = place.get("address",""); loc = place.get("location","")
@@ -1084,10 +1091,6 @@ def handle_detail(text, place):
             parts.append(f"🗺️ https://maps.google.com/?q={lat},{lon}")
         return "\n".join(parts) if parts else f"عنوان {name}: غير متاح."
 
-    if any(w in t for w in ["تليفون","رقم","phone","اتصل"]):
-        ph = place.get("phone","")
-        return f"📞 {name}: {ph}" if ph and ph not in ("","nan") else f"مش عندنا رقم *{name}*."
-
     if any(w in t for w in ["تقييم","كام نجمة","rating","نجوم"]):
         r = fmt_rating(place.get("rating_norm", place.get("rating", 0)))
         return r or "مش عندنا تقييم لهذا المكان."
@@ -1623,6 +1626,13 @@ def chat(text: str, session, user_lat=None, user_lon=None):
         session.add(text, reply, result["intent"], {}, session.last_place, [])
         return result
 
+    if ref == "detail" and not session.last_place:
+        reply = "تقصد أنهي مكان؟ اكتب اسم المكان أو اسألني الأول عن مطعم/صيدلية/كافيه. 😊"
+        result.update(reply=reply, intent="missing_info", confidence=1.0,
+                      best_place=None, all_results=[])
+        session.add(text, reply, "missing_info", {}, None, [])
+        return result
+
     # ── next result ───────────────────────────────────────────────
     if ref == "next" and session.last_results:
         ptr = session.result_pointer + 1
@@ -1659,6 +1669,21 @@ def chat(text: str, session, user_lat=None, user_lon=None):
     ir     = predict_intent(text)
     intent = ir["intent"]
     conf   = ir["confidence"]
+    # Deterministic category intent override: the model/context must not turn
+    # "عايزة صيدلية" into restaurant or keep old category from previous turns.
+    current_text_cat = infer_category(text)
+    if current_text_cat:
+        forced_intent_map = {
+            "restaurant": "nearest_restaurant",
+            "pharmacy": "nearest_pharmacy",
+            "cafe": "nearest_cafe",
+            "supermarket": "nearest_supermarket",
+            "housing": "housing_search",
+        }
+        if intent in ("fallback", "recommend_place", "open_now", "place_details") or intent in SEARCH_INTENTS:
+            intent = forced_intent_map.get(current_text_cat, intent)
+            conf = max(float(conf or 0), 0.96)
+
     result["intent"]     = intent
     result["confidence"] = conf
 
@@ -1740,17 +1765,24 @@ def chat(text: str, session, user_lat=None, user_lon=None):
         session.context_slots = {}
         log.info("Context cleared for new category")
 
+    # Merge context carefully: explicit category in the current user message wins.
+    # If the user changes from restaurant → pharmacy/supermarket/etc, old context is cleared.
+    current_text_cat = infer_category(text) or INTENT_TO_CATEGORY.get(intent)
+    old_ctx_cat = normalize_category(session.context_slots.get("category")) if session.context_slots else None
+    if current_text_cat and old_ctx_cat and normalize_category(current_text_cat) != old_ctx_cat:
+        session.context_slots = {}
+        log.info(f"Context cleared: category changed {old_ctx_cat} → {current_text_cat}")
+
     merged = session.merge(ents)
 
-    category    = normalize_category(
-        merged.get("category") or INTENT_TO_CATEGORY.get(intent) or infer_category(text)
-    )
-    sub_cat     = merged.get("sub_category")
+    category = normalize_category(current_text_cat or merged.get("category"))
+    sub_cat     = ents.get("sub_category") or (merged.get("sub_category") if not current_text_cat else None)
     if sub_cat and normalize_category(sub_cat):
         sub_cat = None
-    location    = merged.get("location")
-    price_range = merged.get("price")
-    place_name  = merged.get("place_name")
+    # inherit location only for true continuations, not for a new category question
+    location    = ents.get("location") or (merged.get("location") if not current_text_cat else None)
+    price_range = ents.get("price") or merged.get("price")
+    place_name  = ents.get("place_name") or (merged.get("place_name") if not current_text_cat else None)
     open_only   = (intent == "open_now")
 
     # ── extract place name — مع منع proximity words ────────────
@@ -1954,18 +1986,21 @@ _ITEM_SEARCH_PATTERNS = [
     "مكان عنده", "مكان عندها", "اماكن عندها", "أماكن عندها",
     "مطعم عنده", "كافيه عنده", "عنده", "عندها",
     "بيعمل", "بتعمل", "بيقدم", "بتقدم", "فيه", "فيها",
-    "عايز", "عايزة", "عاوزه", "عاوز", "نفسي في", "بدور على", "فين",
+    "عايز", "عايزة", "عايزه", "عاوزه", "عاوز", "محتاج", "محتاجة", "نفسي في", "بدور على", "فين",
     "place has", "places have", "has", "serves", "serve", "where can i find"
 ]
 
 def _looks_like_item_search(text: str) -> bool:
     """
-    Detect item/menu-item search without stealing normal category requests.
-
-    Important fixes:
-    - "عايزة سوبرماركت" is nearest_supermarket, not search_by_item.
-    - "عايزة احسن مطعم" is recommendation/search, not item_query=احسن.
-    - "عايزة برجر" / "مكان عنده اسموزي خوخ" stays search_by_item.
+    Detect item/menu-item search without breaking normal category searches.
+
+    Rules:
+    - Category-only requests stay category search:
+      "عايزة صيدلية" → nearest_pharmacy, "عايزة سوبرماركت" → nearest_supermarket.
+    - Ranking/category requests stay place search:
+      "عايزة أحسن مطعم" → recommend/restaurant, not item=أحسن.
+    - Known or unknown item requests become search_by_item:
+      "عايزة برجر", "عايزة فتة", "مكان عنده اسموزي خوخ".
     """
     t = norm(text)
 
@@ -1973,7 +2008,7 @@ def _looks_like_item_search(text: str) -> bool:
     item_context_words = [
         "عنده", "عندها", "بيعمل", "بتعمل", "بيقدم", "بتقدم",
         "فيه", "فيها", "مكان عنده", "مكان عندها", "مطعم عنده", "كافيه عنده",
-        "has", "serves", "serve"
+        "محل عنده", "مطعم فيه", "كافيه فيه", "has", "serves", "serve"
     ]
     if any(kw in t for kw in _MENU_TRIGGERS) and not any(norm(p) in t for p in item_context_words):
         return False
@@ -1982,44 +2017,48 @@ def _looks_like_item_search(text: str) -> bool:
     candidate = extract_item_query(text)
     cand_clean = clean_text(candidate or "")
 
-    # If the remaining candidate is only a category or a ranking adjective, do NOT treat it as food item.
     category_words = set()
     for words in CATEGORY_KEYWORDS.values():
         category_words.update(clean_text(w) for w in words)
     category_words.update(clean_text(k) for k in _CAT_MAP.keys())
+    category_words.update({"restaurant", "restaurants", "cafe", "coffee", "pharmacy", "supermarket", "market", "housing", "hotel"})
 
     ranking_words = {
         "احسن", "افضل", "الاحسن", "الافضل", "أحسن", "أفضل",
         "كويس", "كويسة", "حلو", "حلوة", "ترشيح", "رشحلي",
-        "best", "good", "recommended", "recommend"
+        "best", "good", "recommended", "recommend", "اقرب", "قريب", "nearest"
     }
     ranking_words = {clean_text(w) for w in ranking_words}
 
+    generic_words = {"مكان", "اماكن", "حاجه", "حاجة", "شي", "اي", "اى", "فين", "عندي", "عندك"}
     cand_tokens = [w for w in cand_clean.split() if w]
-    if cand_tokens and all((w in category_words or w in ranking_words) for w in cand_tokens):
+
+    # Candidate is only category/ranking/generic → not item.
+    if not cand_tokens or all((w in category_words or w in ranking_words or w in generic_words) for w in cand_tokens):
         return False
 
-    # Pure category requests must go to nearest_* intents, not item search.
+    # Pure pharmacy/supermarket/housing requests are never item searches.
     if inferred_cat in ("supermarket", "pharmacy", "housing"):
         return False
-    if inferred_cat in ("restaurant", "cafe") and cand_tokens and all(w in category_words or w in ranking_words for w in cand_tokens):
+
+    # Restaurant/cafe requests with only category/ranking words are place search.
+    if inferred_cat in ("restaurant", "cafe") and all(w in category_words or w in ranking_words or w in generic_words for w in cand_tokens):
         return False
 
     has_item_keyword = any(norm(kw) in t for kw in _ITEM_TRIGGERS)
-    has_item_pattern = any(norm(p) in t for p in _ITEM_SEARCH_PATTERNS)
     has_strong_item_context = any(norm(p) in t for p in item_context_words)
 
-    # Known item names are enough: "عايزة برجر".
     if has_item_keyword:
         return True
 
-    # Unknown item names need strong item context: "مكان عنده فتة" / "كافيه بيعمل موهيتو".
-    if has_strong_item_context and candidate and len(cand_clean) >= 3:
+    # Unknown item with strong context: "مكان عنده فتة".
+    if has_strong_item_context:
         return True
 
-    # Avoid broad patterns like "عايزة" stealing normal searches.
-    if has_item_pattern and inferred_cat in ("restaurant", "cafe") and candidate and len(cand_clean) >= 3:
-        return False
+    # Unknown item without category: "عايزة فتة" / "عايز موهيتو".
+    demand_words = ["عايز", "عايزة", "عايزه", "عاوز", "محتاج", "محتاجة", "نفسي", "بدور على", "فين"]
+    if inferred_cat is None and any(norm(w) in t for w in demand_words) and len(cand_clean) >= 3:
+        return True
 
     return False
 
@@ -2426,7 +2465,7 @@ def _infer_item_kind(item_query: str) -> str:
     return "unknown"
 
 
-def scan_menus_for_item(item_query: str, max_places_to_scan: int = 120) -> list[dict]:
+def scan_menus_for_item(item_query: str, max_places_to_scan: int = 300) -> list[dict]:
     """
     Last-resort but important fallback:
     Search inside actual menus when /search-by-item returns empty or too strict.
@@ -2793,6 +2832,11 @@ def clean_place(place: dict) -> dict:
 
     image = _first_value(place, "image", "image_url", "main_image_url", "photo", "photo_url", default="")
     favorite_raw = _first_value(place, "is_favorited", "is_favorite", default=False)
+    phone_val = _first_value(place, "phone", "mobile", "telephone", default="")
+    if isinstance(phone_val, (list, tuple)):
+        phone_val = ", ".join(str(x) for x in phone_val if x)
+    elif isinstance(phone_val, str) and phone_val.startswith("[") and phone_val.endswith("]"):
+        phone_val = phone_val.strip("[]").replace("'", "").replace('"', "").strip()
 
     return {
         "id": str(_first_value(place, "id", "place_id", default="")),
@@ -2808,7 +2852,7 @@ def clean_place(place: dict) -> dict:
         "opening_hours": _first_value(place, "opening_hours", "openning_hours", default=""),
         "distance_km": distance,
         "image": image or "",
-        "phone": _first_value(place, "phone", "mobile", "telephone", default=""),
+        "phone": phone_val,
         "lat": lat,
         "lon": lon,
         "is_favorited": bool(favorite_raw),