Floke
16f5f77967
Korrektes XPath mit normalize-space() zur besseren Erkennung von Branche/Umsatz Bessere Trefferqualität durch Score-Gewichtung (Namensähnlichkeit + Domain) Kein Rückfall mehr auf Kategorien bei fehlender Infobox Spalte Q enthält nun die Versionsnummer
228 lines
8.6 KiB
Python
228 lines
8.6 KiB
Python
# Schritt 1: Nur Wikipedia-Daten extrahieren und in Google Sheet schreiben
|
||
|
||
import os
|
||
import time
|
||
import re
|
||
import gspread
|
||
import wikipedia
|
||
import requests
|
||
import openai
|
||
import csv
|
||
from bs4 import BeautifulSoup
|
||
from oauth2client.service_account import ServiceAccountCredentials
|
||
from datetime import datetime
|
||
from difflib import SequenceMatcher
|
||
from lxml import html as lh
|
||
|
||
# === KONFIGURATION ===
|
||
VERSION = "1.0.8-wiki-api"
|
||
LANG = "de"
|
||
CREDENTIALS = "service_account.json"
|
||
SHEET_URL = "https://docs.google.com/spreadsheets/d/1u_gHr9JUfmV1-iviRzbSe3575QEp7KLhK5jFV_gJcgo"
|
||
DURCHLÄUFE = int(input("Wieviele Zeilen sollen überprüft werden? "))
|
||
MAX_RETRIES = 3
|
||
RETRY_DELAY = 5
|
||
LOG_CSV = "gpt_antworten_log.csv"
|
||
|
||
# === OpenAI API-KEY LADEN ===
|
||
with open("api_key.txt", "r") as f:
|
||
openai.api_key = f.read().strip()
|
||
|
||
# === GOOGLE SHEET VERBINDUNG ===
|
||
scope = ["https://www.googleapis.com/auth/spreadsheets"]
|
||
creds = ServiceAccountCredentials.from_json_keyfile_name(CREDENTIALS, scope)
|
||
sheet = gspread.authorize(creds).open_by_url(SHEET_URL).sheet1
|
||
sheet_values = sheet.get_all_values()
|
||
|
||
# === STARTINDEX SUCHEN (Spalte N = Index 13) ===
|
||
filled_n = [row[13] if len(row) > 13 else '' for row in sheet_values[1:]]
|
||
start = next((i + 1 for i, v in enumerate(filled_n, start=1) if not str(v).strip()), len(filled_n) + 1)
|
||
print(f"Starte bei Zeile {start+1}")
|
||
|
||
wikipedia.set_lang(LANG)
|
||
|
||
# === DOMAIN SCHLÜSSEL ===
|
||
def extract_domain_key(url):
|
||
if not url:
|
||
return ""
|
||
clean_url = url.replace("https://", "").replace("http://", "").split("/")[0]
|
||
parts = clean_url.split(".")
|
||
return parts[0] if len(parts) > 1 else ""
|
||
|
||
# === INFOBOX-PARSING MIT XPATH ===
|
||
def parse_infobox_xpath(html_text):
|
||
doc = lh.fromstring(html_text)
|
||
branche = "k.A."
|
||
umsatz = "k.A."
|
||
try:
|
||
branche_xpath = doc.xpath("//table[contains(@class, 'infobox')]//tr[th[contains(normalize-space(), 'Branche') or contains(normalize-space(), 'Tätigkeitsfeld')]]/td/text()")
|
||
umsatz_xpath = doc.xpath("//table[contains(@class, 'infobox')]//tr[th[contains(translate(normalize-space(), 'UMSATZ', 'umsatz'), 'umsatz')]]/td/text()")
|
||
if branche_xpath:
|
||
branche = branche_xpath[0].strip()
|
||
if umsatz_xpath:
|
||
umsatz_raw = umsatz_xpath[0].strip()
|
||
if "mio" in umsatz_raw.lower() or "millionen" in umsatz_raw.lower():
|
||
match = re.search(r"(\d+[.,]?\d*)", umsatz_raw)
|
||
if match:
|
||
umsatz = match.group(1).replace(",", ".")
|
||
except:
|
||
pass
|
||
return branche, umsatz
|
||
|
||
# === WIKIPEDIA DATEN ===
|
||
WHITELIST_KATEGORIEN = [
|
||
"unternehmen", "hersteller", "produktion", "industrie",
|
||
"maschinenbau", "technik", "dienstleistung", "chemie",
|
||
"pharma", "elektro", "medizin", "bau", "energie",
|
||
"logistik", "automobil"
|
||
]
|
||
|
||
def similarity(a, b):
|
||
return SequenceMatcher(None, a.lower(), b.lower()).ratio()
|
||
|
||
def validate_wikipedia_page(content, title, name, domain_key):
|
||
name_fragments = name.lower().split()[:2]
|
||
title_check = any(frag in title.lower() for frag in name_fragments)
|
||
content_check = any(frag in content.lower() for frag in name_fragments)
|
||
domain_check = domain_key and domain_key.lower() in content.lower()
|
||
sim_check = similarity(name, title) > 0.5
|
||
return (title_check or content_check or domain_check or sim_check)
|
||
|
||
def get_wikipedia_data(name, website_hint=""):
|
||
begriffe = [name.strip(), " ".join(name.split()[:2])]
|
||
domain_key = extract_domain_key(website_hint)
|
||
if domain_key:
|
||
begriffe.append(domain_key)
|
||
|
||
best_score = 0
|
||
best_result = ("", "k.A.", "k.A.")
|
||
|
||
for suchbegriff in begriffe:
|
||
if not suchbegriff:
|
||
continue
|
||
for attempt in range(MAX_RETRIES):
|
||
try:
|
||
results = wikipedia.search(suchbegriff, results=5)
|
||
for title in results:
|
||
try:
|
||
page = wikipedia.page(title, auto_suggest=False)
|
||
html_text = requests.get(page.url, timeout=10).text
|
||
if not validate_wikipedia_page(page.content, title, name, domain_key):
|
||
continue
|
||
branche, umsatz = parse_infobox_xpath(html_text)
|
||
score = similarity(name, title)
|
||
if branche != "k.A.":
|
||
score += 0.1
|
||
if domain_key and domain_key in page.content.lower():
|
||
score += 0.1
|
||
if score > best_score:
|
||
best_score = score
|
||
best_result = (page.url, branche or "k.A.", umsatz or "k.A.")
|
||
except:
|
||
continue
|
||
except Exception as e:
|
||
print(f"⚠️ Wikipedia-Fehler ({suchbegriff}, Versuch {attempt+1}): {str(e)[:100]}")
|
||
time.sleep(RETRY_DELAY)
|
||
|
||
return best_result
|
||
|
||
# === SCHRITT 1: WIKIPEDIA VERARBEITUNG ===
|
||
for i in range(start, min(start + DURCHLÄUFE, len(sheet_values))):
|
||
row = sheet_values[i]
|
||
print(f"\n[{datetime.now().strftime('%H:%M:%S')}] Verarbeite Zeile {i+1}: {row[0]}")
|
||
url, wiki_branche, umsatz = get_wikipedia_data(row[0], row[1])
|
||
wiki_final = wiki_branche if url else "k.A."
|
||
umsatz_final = umsatz if url else "k.A."
|
||
values = [
|
||
wiki_final,
|
||
"k.A.", # LinkedIn-Branche leer
|
||
umsatz_final,
|
||
"k.A.", "k.A.", "k.A.",
|
||
url,
|
||
datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
||
"k.A.", "k.A.",
|
||
VERSION
|
||
]
|
||
sheet.update(range_name=f"G{i+1}:Q{i+1}", values=[values])
|
||
print(f"✅ Aktualisiert: {values[:3]}...")
|
||
time.sleep(RETRY_DELAY)
|
||
|
||
print("\n✅ Wikipedia-Auswertung abgeschlossen")
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
# === SCHRITT 2: GPT-BEWERTUNG ===
|
||
def classify_company(row, wikipedia_url=""):
|
||
user_prompt = {
|
||
"role": "user",
|
||
"content": f"{row[0]};{row[1]};{row[2]};{row[4]};{row[5]}\nWikipedia-Link: {wikipedia_url}"
|
||
}
|
||
for attempt in range(MAX_RETRIES):
|
||
try:
|
||
response = openai.chat.completions.create(
|
||
model="gpt-3.5-turbo",
|
||
messages=[
|
||
{
|
||
"role": "system",
|
||
"content": (
|
||
"Du bist ein Experte für Brancheneinstufung und FSM-Potenzialbewertung.\n"
|
||
"Bitte beziehe dich ausschließlich auf das konkret genannte Unternehmen.\n"
|
||
"FSM steht für Field Service Management. Ziel ist es, Unternehmen mit >50 Technikern im Außendienst zu identifizieren.\n\n"
|
||
"Struktur: Firmenname; Website; Ort; Aktuelle Einstufung; Beschreibung der Branche Extern\n\n"
|
||
"Gib deine Antwort im CSV-Format zurück (1 Zeile, 8 Spalten):\n"
|
||
"Wikipedia-Branche;LinkedIn-Branche;Umsatz (Mio €);Empfohlene Neueinstufung;Begründung;FSM-Relevanz;Techniker-Einschätzung;Techniker-Begründung"
|
||
)
|
||
},
|
||
user_prompt
|
||
],
|
||
temperature=0,
|
||
timeout=15
|
||
)
|
||
full_text = response.choices[0].message.content.strip()
|
||
break
|
||
except Exception as e:
|
||
print(f"⚠️ GPT-Fehler (Versuch {attempt+1}): {str(e)[:100]}")
|
||
time.sleep(RETRY_DELAY)
|
||
else:
|
||
print("❌ GPT 3x fehlgeschlagen – Standardwerte")
|
||
full_text = "k.A.;k.A.;k.A.;k.A.;k.A.;k.A.;k.A.;k.A."
|
||
|
||
lines = full_text.splitlines()
|
||
csv_line = next((l for l in lines if ";" in l), "")
|
||
parts = [v.strip() for v in csv_line.split(";")] if csv_line else ["k.A."] * 8
|
||
|
||
with open(LOG_CSV, "a", newline="", encoding="utf-8") as log:
|
||
writer = csv.writer(log, delimiter=";")
|
||
writer.writerow([datetime.now().strftime("%Y-%m-%d %H:%M:%S"), row[0], *parts, full_text])
|
||
|
||
return parts
|
||
|
||
# === SCHRITT 2 DURCHFÜHREN ===
|
||
for i in range(start, min(start + DURCHLÄUFE, len(sheet_values))):
|
||
row = sheet_values[i]
|
||
print(f"\n[{datetime.now().strftime('%H:%M:%S')}] GPT-Bewertung für Zeile {i+1}: {row[0]}")
|
||
wiki_url = row[12] if len(row) > 12 else ""
|
||
wiki, linkedin, umsatz_chat, new_cat, reason, fsm, techniker, techniker_reason = classify_company(row, wikipedia_url=wiki_url)
|
||
values = [
|
||
wiki,
|
||
linkedin,
|
||
umsatz_chat,
|
||
new_cat,
|
||
reason,
|
||
fsm,
|
||
wiki_url,
|
||
datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
||
techniker,
|
||
techniker_reason
|
||
]
|
||
sheet.update(range_name=f"G{i+1}:P{i+1}", values=[values])
|
||
time.sleep(RETRY_DELAY)
|
||
|
||
print("\n✅ GPT-Bewertung abgeschlossen")
|