import re import requests import sqlite3 import pandas as pd from concurrent.futures import ThreadPoolExecutor from bs4 import BeautifulSoup, SoupStrainer from sys import argv regexes = { 'artist': re.compile(r'SplitScreenContentHeaderArtist-\w*'), 'album': re.compile(r'SplitScreenContentHeaderHed-\w*'), 'year_released': re.compile(r'SplitScreenContentHeaderReleaseYear-\w*'), 'rating': re.compile(r'Rating-\w*'), 'small_text': re.compile(r'SplitScreenContentHeaderDekDown-\w*'), 'review': re.compile(r'body__inner-container'), 'reviewer': re.compile(r'BylineName'), 'genre': re.compile(r'SplitScreenContentHeaderInfoSlice-\w*'), 'label': re.compile(r'SplitScreenContentHeaderInfoSlice-\w*'), 'reviewed': re.compile(r'SplitScreenContentHeaderInfoSlice-\w*'), 'album_art_url': re.compile(r'SplitScreenContentHeaderImage-\w*'), } def fetch(args): ''' Fetch a single url and return a dictionary of data from a Pitchfork review ''' url, session = args response = session.get(url) if response.status_code == 200: soup_strainer = SoupStrainer('article', {'data-testid': 'ReviewPageArticle'}) soup = BeautifulSoup(response.content, 'lxml', parse_only=soup_strainer) if soup.find('article', {'data-testid': 'ReviewPageArticle'}) is None: with open('not_done.txt', 'a') as f: f.write(url + '\n') return None print('.', end='', flush=True) result = data_from_soup(soup) if result is None: with open('not_done.txt', 'a') as f: f.write(url + '\n') return result else: with open('errors.txt', 'a') as f: f.write(url + '\n') return None def get_reviews(urls): ''' Return a list of review data dictionaries from the provided urls ''' reviews = [] session = requests.Session() with ThreadPoolExecutor() as executor: for result in executor.map(fetch, ((url, session) for url in urls)): if result: # Check if result is not None reviews.append(result) print() return reviews def data_from_soup(soup): ''' Return a dictionary of data from a Pitchfork review ''' artist = soup.find('div', {'class': regexes['artist']}).text.strip() album = soup.find('h1', {'class': regexes['album']}).text.strip() year_released = soup.find('time', {'class': regexes['year_released']}) if year_released: year_released = int(year_released.text.strip()) else: return None rating = float(soup.find('p', {'class': regexes['rating']}).text.strip()) small_text = soup.find('div', {'class': regexes['small_text']}) small_text = small_text.text.strip() if small_text else 'N/A' review = "".join(e.text for e in soup.find('div', {'class': regexes['review']}).descendants if e.name == 'p') reviewer = soup.find('span', {'data-testid': regexes['reviewer']}) reviewer = reviewer.text.strip()[3:] if reviewer else 'N/A' misc = [e.text for e in soup.find('div', {'class': regexes['genre']}).descendants if e.name == 'li'] misc = {'genre': 'N/A', 'label': 'N/A', 'reviewed': 'N/A'} | {e.split(':')[0].strip().lower(): e.split(':')[1].strip() for e in misc} album_art_url = soup.find('source', {'media': '(max-width: 767px)'}) album_art_url = album_art_url['srcset'].split(',')[-2].strip() if album_art_url else 'N/A' return { 'artist': artist, 'album': album, 'year_released': year_released, 'rating': rating, 'small_text': small_text, 'review': review, 'reviewer': reviewer, 'genre': misc['genre'], 'label': misc['label'], 'reviewed': misc['reviewed'], 'album_art_url': album_art_url, } def insert_into_db(data, cursor): ''' Insert data into a sqlite3 database ''' for review in data: artist = review.get('artist') album = review.get('album') year_released = review.get('year_released') rating = review.get('rating') small_text = review.get('small_text') review_text = review.get('review') # 'review' is a reserved word in Python reviewer = review.get('reviewer') genre = review.get('genre') label = review.get('label') reviewed = review.get('reviewed') album_art_url = review.get('album_art_url') cursor.execute('SELECT * FROM reviews WHERE artist=? AND album=?', (artist, album)) result = cursor.fetchone() if result is None: # Insert new review into database cursor.execute(''' INSERT INTO reviews ( artist, album, year_released, rating, small_text, review, reviewer, genre, label, reviewed, album_art_url ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?) ''', ( artist, album, year_released, rating, small_text, review_text, reviewer, genre, label, reviewed, album_art_url )) def main(start, end): conn = sqlite3.connect('reviews.db') c = conn.cursor() # Create table with all necessary fields c.execute(''' CREATE TABLE IF NOT EXISTS reviews ( artist TEXT, album TEXT, year_released INTEGER, rating REAL, small_text TEXT, review TEXT, reviewer TEXT, genre TEXT, label TEXT, reviewed TEXT, album_art_url TEXT ) ''') # Read URLs from a CSV file into a list df = pd.read_csv('urls.csv') urls = df['url'].tolist() # replace 'url' with your actual column name start, end = max(0, start), min(len(urls), end) urls = urls[start:end] print(f'Fetching {len(urls)} reviews') data = get_reviews(urls) print(f'Fetching complete. Inserting into database') insert_into_db(data, c) print('Done') conn.commit() conn.close() if __name__ == '__main__': main(int(argv[1]), int(argv[2]))