revoce 2

duplicate_checker.py

@@ -1,4 +1,4 @@
-# duplicate_checker.py (v2.4 - Optimized Brute-Force)
+# duplicate_checker.py (v2.2 - Multi-Key Blocking & optimiertes Scoring)
 
 import logging
 import pandas as pd
@@ -6,12 +6,12 @@ from thefuzz import fuzz
 from config import Config
 from helpers import normalize_company_name, simple_normalize_url
 from google_sheet_handler import GoogleSheetHandler
-import time
+from collections import defaultdict
 
 # --- Konfiguration ---
 CRM_SHEET_NAME = "CRM_Accounts"
 MATCHING_SHEET_NAME = "Matching_Accounts"
-SCORE_THRESHOLD = 85 # Treffer unter diesem Wert werden nicht als "potenzieller Treffer" angezeigt
+SCORE_THRESHOLD = 85 # Etwas höherer Schwellenwert für bessere Präzision
 
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
@@ -19,15 +19,13 @@ def calculate_similarity_details(record1, record2):
     """Berechnet einen gewichteten Ähnlichkeits-Score und gibt die Details zurück."""
     scores = {'name': 0, 'location': 0, 'domain': 0}
     
-    # Domain-Match gibt 100 Punkte
     if record1.get('normalized_domain') and record1['normalized_domain'] != 'k.a.' and record1['normalized_domain'] == record2.get('normalized_domain'):
         scores['domain'] = 100
     
-    # Namensähnlichkeit (85% Gewichtung)
+    # Höhere Gewichtung für den Namen, da die Website oft fehlt
     if record1.get('normalized_name') and record2.get('normalized_name'):
         scores['name'] = round(fuzz.token_set_ratio(record1['normalized_name'], record2['normalized_name']) * 0.85)
 
-    # Standort-Bonus (20 Punkte)
     if record1.get('CRM Ort') and record1['CRM Ort'] == record2.get('CRM Ort'):
         if record1.get('CRM Land') and record1['CRM Land'] == record2.get('CRM Land'):
             scores['location'] = 20
@@ -35,9 +33,28 @@ def calculate_similarity_details(record1, record2):
     total_score = sum(scores.values())
     return {'total': total_score, 'details': scores}
 
+def create_blocking_keys(name):
+    """Erstellt mehrere Blocking Keys für einen Namen, um die Sensitivität zu erhöhen."""
+    if not name:
+        return []
+    
+    words = name.split()
+    keys = set()
+    
+    # 1. Erstes Wort
+    if len(words) > 0:
+        keys.add(words[0])
+    # 2. Zweites Wort (falls vorhanden)
+    if len(words) > 1:
+        keys.add(words[1])
+    # 3. Erste 4 Buchstaben des ersten Wortes
+    if len(words) > 0 and len(words[0]) >= 4:
+        keys.add(words[0][:4])
+        
+    return list(keys)
+
 def main():
-    start_time = time.time()
-    logging.info("Starte den Duplikats-Check (v2.4 - Optimized Brute-Force)...")
+    logging.info("Starte den Duplikats-Check (v2.2 mit Multi-Key Blocking)...")
     
     try:
         sheet_handler = GoogleSheetHandler()
@@ -60,22 +77,32 @@ def main():
         df['normalized_domain'] = df['CRM Website'].astype(str).apply(simple_normalize_url)
         df['CRM Ort'] = df['CRM Ort'].astype(str).str.lower().str.strip()
         df['CRM Land'] = df['CRM Land'].astype(str).str.lower().str.strip()
+        df['block_keys'] = df['normalized_name'].apply(create_blocking_keys)
 
-    # Konvertiere DataFrames in Listen von Dictionaries für schnelleren Zugriff in der Schleife
-    crm_records = crm_df.to_dict('records')
-    matching_records = matching_df.to_dict('records')
+    logging.info("Erstelle Index für CRM-Daten zur Beschleunigung...")
+    crm_index = defaultdict(list)
+    for record in crm_df.to_dict('records'):
+        for key in record['block_keys']:
+            crm_index[key].append(record)
     
-    logging.info(f"Starte Matching-Prozess: {len(matching_records)} Einträge werden mit {len(crm_records)} CRM-Einträgen verglichen...")
+    logging.info("Starte Matching-Prozess...")
     results = []
     
-    for i, match_record in enumerate(matching_records):
+    for match_record in matching_df.to_dict('records'):
         best_score_info = {'total': 0, 'details': {'name': 0, 'location': 0, 'domain': 0}}
         best_match_name = ""
         
-        logging.info(f"Prüfe {i + 1}/{len(matching_records)}: {match_record['CRM Name']}...")
+        logging.info(f"Prüfe: {match_record['CRM Name']}...")
+
+        candidate_pool = {}
+        for key in match_record['block_keys']:
+            for crm_record in crm_index.get(key, []):
+                candidate_pool[crm_record['CRM Name']] = crm_record
         
-        # Der Brute-Force-Ansatz: Vergleiche mit jedem einzelnen CRM-Eintrag
-        for crm_record in crm_records:
+        if not candidate_pool:
+            logging.debug("  -> Keine Kandidaten im Index gefunden.")
+        
+        for crm_record in candidate_pool.values():
             score_info = calculate_similarity_details(match_record, crm_record)
             if score_info['total'] > best_score_info['total']:
                 best_score_info = score_info
@@ -102,8 +129,5 @@ def main():
     else:
         logging.error("FEHLER beim Schreiben der Ergebnisse ins Google Sheet.")
 
-    end_time = time.time()
-    logging.info(f"Gesamtdauer des Duplikats-Checks: {end_time - start_time:.2f} Sekunden.")
-
 if __name__ == "__main__":
     main()