diff --git a/duplicate_checker.py b/duplicate_checker.py index 51ff913c..05de2aab 100644 --- a/duplicate_checker.py +++ b/duplicate_checker.py @@ -1,4 +1,4 @@ -# duplicate_checker.py (v2.0 - mit Blocking-Strategie) +# duplicate_checker.py (v2.5 - Final Hybrid Approach) import logging import pandas as pd @@ -6,48 +6,68 @@ from thefuzz import fuzz from config import Config from helpers import normalize_company_name, simple_normalize_url from google_sheet_handler import GoogleSheetHandler +from collections import defaultdict +import time # --- Konfiguration --- CRM_SHEET_NAME = "CRM_Accounts" MATCHING_SHEET_NAME = "Matching_Accounts" -SCORE_THRESHOLD = 80 +SCORE_THRESHOLD = 85 # Zeigt nur Treffer an, die diesen Score erreichen oder übertreffen + +# Erweiterte Liste von generischen Wörtern, die für das Blocking ignoriert werden +BLOCKING_STOP_WORDS = { + 'gmbh', 'ag', 'co', 'kg', 'se', 'holding', 'gruppe', 'industries', 'systems', 'technik', 'service', + 'services', 'solutions', 'management', 'international', 'und', 'germany', 'deutschland', 'gbr', + 'mbh', 'company', 'limited', 'logistics', 'construction', 'products', 'group' +} logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s') -def calculate_similarity(record1, record2): - """Berechnet einen gewichteten Ähnlichkeits-Score zwischen zwei Datensätzen.""" - total_score = 0 - if record1['normalized_domain'] and record1['normalized_domain'] == record2['normalized_domain']: - total_score += 100 - if record1['normalized_name'] and record2['normalized_name']: - name_similarity = fuzz.token_set_ratio(record1['normalized_name'], record2['normalized_name']) - total_score += name_similarity * 0.7 - if record1['CRM Ort'] and record1['CRM Ort'] == record2['CRM Ort']: - if record1['CRM Land'] and record1['CRM Land'] == record2['CRM Land']: - total_score += 20 - return round(total_score) +def calculate_similarity_details(record1, record2): + """Berechnet einen gewichteten Ähnlichkeits-Score und gibt die Details zurück.""" + scores = {'name': 0, 'location': 0, 'domain': 0} + + if record1.get('normalized_domain') and record1['normalized_domain'] != 'k.a.' and record1['normalized_domain'] == record2.get('normalized_domain'): + scores['domain'] = 100 + + if record1.get('normalized_name') and record2.get('normalized_name'): + # Wir verwenden token_sort_ratio für eine gute Balance zwischen Wortreihenfolge und Inhalt + scores['name'] = round(fuzz.token_sort_ratio(record1['normalized_name'], record2['normalized_name']) * 0.85) + + 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 + + total_score = sum(scores.values()) + return {'total': total_score, 'details': scores} + +def create_blocking_keys(name): + """Erstellt Blocking Keys aus allen signifikanten Wörtern eines Namens.""" + if not name: + return [] + # Filtere Stop-Wörter und sehr kurze Wörter (z.B. '&') aus der Wortliste + significant_words = {word for word in name.split() if word not in BLOCKING_STOP_WORDS and len(word) > 2} + return list(significant_words) def main(): - """Hauptfunktion zum Laden, Vergleichen und Schreiben der Daten.""" - logging.info("Starte den Duplikats-Check (v2.0 mit Blocking)...") + start_time = time.time() + logging.info("Starte den Duplikats-Check (v2.5 - Final Hybrid Approach)...") + # ... (Initialisierung und Laden der Daten bleibt gleich) ... try: sheet_handler = GoogleSheetHandler() except Exception as e: - logging.critical(f"FEHLER bei Initialisierung des GoogleSheetHandler: {e}") + logging.critical(f"FEHLER bei Initialisierung: {e}") return logging.info(f"Lade Master-Daten aus '{CRM_SHEET_NAME}'...") crm_df = sheet_handler.get_sheet_as_dataframe(CRM_SHEET_NAME) - if crm_df is None or crm_df.empty: - logging.critical(f"Konnte keine Daten aus '{CRM_SHEET_NAME}' laden. Breche ab.") - return + if crm_df is None or crm_df.empty: return logging.info(f"Lade zu prüfende Daten aus '{MATCHING_SHEET_NAME}'...") matching_df = sheet_handler.get_sheet_as_dataframe(MATCHING_SHEET_NAME) - if matching_df is None or matching_df.empty: - logging.critical(f"Konnte keine Daten aus '{MATCHING_SHEET_NAME}' laden. Breche ab.") - return + if matching_df is None or matching_df.empty: return + original_matching_df = matching_df.copy() logging.info("Normalisiere Daten für den Vergleich...") for df in [crm_df, matching_df]: @@ -55,60 +75,59 @@ 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() - # Blocking Key: Das erste Wort des normalisierten Namens - df['block_key'] = df['normalized_name'].apply(lambda x: x.split()[0] if x else None) + df['block_keys'] = df['normalized_name'].apply(create_blocking_keys) - # --- NEUE, SCHNELLE BLOCKING-STRATEGIE --- logging.info("Erstelle Index für CRM-Daten zur Beschleunigung...") - crm_index = {} - for index, row in crm_df.iterrows(): - key = row['block_key'] - if key: - if key not in crm_index: - crm_index[key] = [] - crm_index[key].append(row) + crm_index = defaultdict(list) + crm_records = crm_df.to_dict('records') + for record in crm_records: + for key in record['block_keys']: + crm_index[key].append(record) logging.info("Starte Matching-Prozess...") results = [] - total_matches = len(matching_df) - for index, match_row in matching_df.iterrows(): - best_score = 0 + 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 {index + 1}/{total_matches}: {match_row['CRM Name']}...") + logging.info(f"Prüfe: {match_record['CRM Name']}...") - # Finde den Block von Kandidaten - block_key = match_row['block_key'] - candidates = crm_index.get(block_key, []) + 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 - # Führe den teuren Vergleich nur für die Kandidaten in diesem Block durch - for crm_row in candidates: - score = calculate_similarity(match_row, crm_row) - if score > best_score: - best_score = score - best_match_name = crm_row['CRM Name'] + # Brute-Force-Vergleich innerhalb des intelligenten Blocks + 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 + best_match_name = crm_record['CRM Name'] - if best_score >= SCORE_THRESHOLD: - results.append({'Potenzieller Treffer im CRM': best_match_name, 'Ähnlichkeits-Score': best_score}) - else: - # Wenn nichts im Block gefunden wurde, trotzdem den besten Treffer (kann 0 sein) anzeigen - results.append({'Potenzieller Treffer im CRM': '' if not best_match_name else best_match_name, 'Ähnlichkeits-Score': best_score}) + results.append({ + 'Potenzieller Treffer im CRM': best_match_name if best_score_info['total'] >= SCORE_THRESHOLD else "", + 'Score (Gesamt)': best_score_info['total'], + 'Score (Name)': best_score_info['details']['name'], + 'Bonus (Standort)': best_score_info['details']['location'], + 'Bonus (Domain)': best_score_info['details']['domain'] + }) logging.info("Matching abgeschlossen. Schreibe Ergebnisse zurück ins Sheet...") result_df = pd.DataFrame(results) - # Die ursprünglichen Spalten aus matching_df für die Ausgabe nehmen - output_df = matching_df[['CRM Name', 'CRM Website', 'CRM Ort', 'CRM Land']].copy() - output_df = pd.concat([output_df.reset_index(drop=True), result_df], axis=1) + output_df = pd.concat([original_matching_df.reset_index(drop=True), result_df], axis=1) data_to_write = [output_df.columns.values.tolist()] + output_df.values.tolist() success = sheet_handler.clear_and_write_data(MATCHING_SHEET_NAME, data_to_write) if success: - logging.info(f"Ergebnisse erfolgreich in das Tabellenblatt '{MATCHING_SHEET_NAME}' geschrieben.") + logging.info(f"Ergebnisse erfolgreich in '{MATCHING_SHEET_NAME}' geschrieben.") 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() \ No newline at end of file