doc id problem fixed

index_retriever.py

@@ -31,20 +31,21 @@ def keyword_filter_nodes(query, nodes, min_keyword_matches=1):
 
 
 def normalize_doc_id(doc_id: str) -> str:
-    """Normalize document ID for consistent comparison."""
+    """Normalize document ID - KEEP dots for numeric parts"""
     doc_id = doc_id.upper().strip()
-    doc_id = re.sub(r'[^\w\d\.]+', '', doc_id)  # remove spaces, dashes, etc.
+    doc_id = re.sub(r'\s+', '', doc_id)  # Remove spaces only
     doc_id = doc_id.replace("ГОСТР", "ГОСТ")
     doc_id = doc_id.replace("GOSTR", "ГОСТ")
     return doc_id
 
 def base_number(doc_id: str) -> str:
-    """Extract base numeric pattern (e.g., '59023.4' from 'ГОСТ Р 59023.4-2020')."""
-    m = re.search(r'(\d+(?:\.\d+)+)', doc_id)
+    """Extract full numeric pattern including all parts (e.g., '59023.6' from 'ГОСТ 59023.6')"""
+    # Match: 59023.6 or 59023.4 or 50.05.01 etc.
+    m = re.search(r'(\d+(?:\.\d+)*)', doc_id)
     return m.group(1) if m else ""
 
-def filter_nodes_by_doc_id(nodes, doc_ids, threshold=0.75):
-    """Filter nodes by normalized document ID with fallback to fuzzy numeric match."""
+def filter_nodes_by_doc_id(nodes, doc_ids, threshold=0.85):
+    """Filter nodes by document ID with strict numeric matching"""
     if not doc_ids:
         return nodes
 
@@ -57,22 +58,17 @@ def filter_nodes_by_doc_id(nodes, doc_ids, threshold=0.75):
         node_base = base_number(node_doc_id)
 
         for q_doc, q_base in zip(doc_ids_norm, doc_ids_base):
-            # Strong match: same base number (e.g., 59023.4)
+            # STRICT: base number must match exactly
             if q_base and node_base and q_base == node_base:
                 filtered.append(node)
                 break
-
-            # Medium match: similarity ratio > threshold
-            if SequenceMatcher(None, node_doc_id, q_doc).ratio() >= threshold:
-                filtered.append(node)
-                break
-
-            # Weak fallback: contains or partial substring
-            if q_base in node_doc_id or q_doc in node_doc_id:
+            
+            # STRICT: full normalized ID must match exactly or have very high similarity
+            elif SequenceMatcher(None, node_doc_id, q_doc).ratio() >= threshold:
                 filtered.append(node)
                 break
 
-    return filtered if filtered else nodes  # Fallback: keep all if none matched
+    return filtered if filtered else nodes
 
 
 def extract_doc_id_from_query(query):