O Claude, um modelo de linguagem grande (LLM) da Anthropic, é um dos modelos de IA mais usados em todo o mundo. Depois de aprender a raspar sites, a maior parte do seu tempo será gasta escrevendo analisadores.
Com os modelos modernos de IA, podemos de fato automatizar esse processo. Em vez de passar horas analisando um site difícil, um LLM pode analisá-lo para você em menos de cinco minutos.
Temos outros tutoriais sobre o uso de IA para gerar código rapidamente, mas aqui, de fato, integraremos o Claude ao nosso ambiente Python. Acompanhe o tutorial e automatize a parte mais tediosa de seu trabalho.
Criação de uma conta antrópica
Para obter acesso à API do Claude, você precisa criar uma conta no Anthropic. Você pode fazer isso usando seu e-mail ou sua conta do Google.
Depois de criar uma conta, clique na guia “API Keys” e você poderá criar uma chave de API. Depois de criá-la, proteja essa chave com toda a sua vida. Você não poderá visualizá-la uma segunda vez.
Guarde essa chave em um local seguro, pois você não poderá usar a API sem ela.
Como fazer uma solicitação básica
Faremos nossa primeira solicitação ao Quotes to Scrape. Esse site não muda muito e foi criado para raspagem educacional. Isso nos dá uma página estática para testar as respostas do Claude.
Primeiro, precisamos instalar o antropic.
pip install anthropic
A configuração de uma instância de cliente é bastante simples.
#set up the client
client = anthropic.Anthropic(
api_key=ANTHROPIC_API_KEY,
)
Aqui está a função em que alimentamos tudo no Claude.
#takes in http response and sends its text to claude for processing
def extract_with_claude(response, token_limit=200000, max_tokens_per_chunk=1024):
message = client.messages.create(
model="claude-3-5-haiku-20241022",
max_tokens=2048,
messages=[
{
"role": "user",
"content": f"""
Hello, please parse this chunk of the HTML page and convert it to JSON.
Make sure to strip newlines, remove escape characters, and whitespace:
{response.text}
"""
}
]
)
text = message.to_dict()["content"][0]["text"]
return text
modelo: O modelo que desejamos usar. Estamos usandoclaude-3-5-haiku-20241022.max_tokensrepresenta o número máximo de tokens que queremos que o Claude use para a resposta.- Por padrão, a API retorna um objeto
Message. Usamosto_dict()para convertê-lo em pares de valores-chave que são fáceis de trabalhar.
import re
import requests
import anthropic
import json
ANTHROPIC_API_KEY = "YOUR-ANTHROPIC-API-KEY"
#set up the client
client = anthropic.Anthropic(
api_key=ANTHROPIC_API_KEY,
)
#takes in http response and sends its text to claude for processing
def extract_with_claude(response, token_limit=200000, max_tokens_per_chunk=1024):
message = client.messages.create(
model="claude-3-5-haiku-20241022",
max_tokens=2048,
messages=[
{
"role": "user",
"content": f"""
Hello, please parse this chunk of the HTML page and convert it to JSON.
Make sure to strip newlines, remove escape characters, and whitespace:
{response.text}
"""
}
]
)
text = message.to_dict()["content"][0]["text"]
return text
if __name__ == "__main__":
TARGET_URL = "https://quotes.toscrape.com"
response = requests.get(TARGET_URL)
print(extract_with_claude(response))
Compreensão das respostas de Claude
Como mencionamos acima, por padrão, o Claude retorna um objeto Message. O método to_dict() torna nossa resposta um pouco mais fácil de ser integrada ao nosso ambiente Python. Entretanto, ela ainda não está pronta para ser usada. Dê uma olhada na resposta abaixo.
I'll help you parse this HTML into a JSON format. I'll focus on extracting the quotes, their authors, and tags. Here's the resulting JSON:
```json
{
"quotes": [
{
"text": "The world as we have created it is a process of our thinking. It cannot be changed without changing our thinking.",
"author": "Albert Einstein",
"tags": ["change", "deep-thoughts", "thinking", "world"]
},
{
"text": "It is our choices, Harry, that show what we truly are, far more than our abilities.",
"author": "J.K. Rowling",
"tags": ["abilities", "choices"]
},
{
"text": "There are only two ways to live your life. One is as though nothing is a miracle. The other is as though everything is a miracle.",
"author": "Albert Einstein",
"tags": ["inspirational", "life", "live", "miracle", "miracles"]
},
{
"text": "The person, be it gentleman or lady, who has not pleasure in a good novel, must be intolerably stupid.",
"author": "Jane Austen",
"tags": ["aliteracy", "books", "classic", "humor"]
},
{
"text": "Imperfection is beauty, madness is genius and it's better to be absolutely ridiculous than absolutely boring.",
"author": "Marilyn Monroe",
"tags": ["be-yourself", "inspirational"]
},
{
"text": "Try not to become a man of success. Rather become a man of value.",
"author": "Albert Einstein",
"tags": ["adulthood", "success", "value"]
},
{
"text": "It is better to be hated for what you are than to be loved for what you are not.",
"author": "André Gide",
"tags": ["life", "love"]
},
{
"text": "I have not failed. I've just found 10,000 ways that won't work.",
"author": "Thomas A. Edison",
"tags": ["edison", "failure", "inspirational", "paraphrased"]
},
{
"text": "A woman is like a tea bag; you never know how strong it is until it's in hot water.",
"author": "Eleanor Roosevelt",
"tags": ["misattributed-eleanor-roosevelt"]
},
{
"text": "A day without sunshine is like, you know, night.",
"author": "Steve Martin",
"tags": ["humor", "obvious", "simile"]
}
],
"top_tags": [
{"tag": "love", "size": 28},
{"tag": "inspirational", "size": 26},
{"tag": "life", "size": 26},
{"tag": "humor", "size": 24},
{"tag": "books", "size": 22},
{"tag": "reading", "size": 14},
{"tag": "friendship", "size": 10},
{"tag": "friends", "size": 8},
{"tag": "truth", "size": 8},
{"tag": "simile", "size": 6}
]
}
```
I've extracted:
1. The quotes, their text, authors, and associated tags
2. The top tags with their relative sizes
The JSON is clean, without newlines or escape characters, and follows a clear structure. Would you like me to modify the JSON in any way?
Recebemos o JSON que queremos, mas ele está incorporado em uma cadeia de caracteres maior. Precisamos extrair do texto os dados extraídos de Claude.
Extração de dados da resposta
O JSON está embutido em backticks, “`, exatamente como um bloco de código markdown. Para obter nossa resposta, usaremos um regex para encontrar o início e o fim do JSON.
def pull_json_data(claude_text):
json_match = re.search(r"```json\n(.*?)\n```", claude_text, re.DOTALL)
if json_match:
# Extract the JSON and load it into a Python dictionary
parsed_json = json.loads(json_match.group(1))
return parsed_json # Pretty-print the JSON
else:
print("Could not find JSON in the response.")
Aqui está um exemplo de código completo que imprime apenas os dados extraídos da página.
import re
import requests
import anthropic
import json
ANTHROPIC_API_KEY = "YOUR-ANTHROPIC-API-KEY"
# Set up the Claude client
client = anthropic.Anthropic(
api_key=ANTHROPIC_API_KEY,
)
def pull_json_data(claude_text):
json_match = re.search(r"```json\n(.*?)\n```", claude_text, re.DOTALL)
if json_match:
# Extract the JSON and load it into a Python dictionary
parsed_json = json.loads(json_match.group(1))
return parsed_json # Pretty-print the JSON
else:
print("Could not find JSON in the response.")
def extract_with_claude(response, token_limit=200000, max_tokens_per_chunk=1024):
message = client.messages.create(
model="claude-3-5-haiku-20241022",
max_tokens=2048,
messages=[
{
"role": "user",
"content": f"""
Hello, please parse this chunk of the HTML page and convert it to JSON.
Make sure to strip newlines, remove escape characters, and whitespace:
{response.text}
"""
}
]
)
text = message.to_dict()["content"][0]["text"]
return pull_json_data(text)
if __name__ == "__main__":
TARGET_URL = "https://quotes.toscrape.com"
response = requests.get(TARGET_URL)
print(extract_with_claude(response))
Aqui estão nossos dados extraídos, livres da saída de conversação do Claude.
{'quotes': [{'text': 'The world as we have created it is a process of our thinking. It cannot be changed without changing our thinking.', 'author': 'Albert Einstein', 'tags': ['change', 'deep-thoughts', 'thinking', 'world']}, {'text': 'It is our choices, Harry, that show what we truly are, far more than our abilities.', 'author': 'J.K. Rowling', 'tags': ['abilities', 'choices']}, {'text': 'There are only two ways to live your life. One is as though nothing is a miracle. The other is as though everything is a miracle.', 'author': 'Albert Einstein', 'tags': ['inspirational', 'life', 'live', 'miracle', 'miracles']}, {'text': 'The person, be it gentleman or lady, who has not pleasure in a good novel, must be intolerably stupid.', 'author': 'Jane Austen', 'tags': ['aliteracy', 'books', 'classic', 'humor']}, {'text': "Imperfection is beauty, madness is genius and it's better to be absolutely ridiculous than absolutely boring.", 'author': 'Marilyn Monroe', 'tags': ['be-yourself', 'inspirational']}, {'text': 'Try not to become a man of success. Rather become a man of value.', 'author': 'Albert Einstein', 'tags': ['adulthood', 'success', 'value']}, {'text': 'It is better to be hated for what you are than to be loved for what you are not.', 'author': 'André Gide', 'tags': ['life', 'love']}, {'text': "I have not failed. I've just found 10,000 ways that won't work.", 'author': 'Thomas A. Edison', 'tags': ['edison', 'failure', 'inspirational', 'paraphrased']}, {'text': "A woman is like a tea bag; you never know how strong it is until it's in hot water.", 'author': 'Eleanor Roosevelt', 'tags': ['misattributed-eleanor-roosevelt']}, {'text': 'A day without sunshine is like, you know, night.', 'author': 'Steve Martin', 'tags': ['humor', 'obvious', 'simile']}], 'topTags': ['love', 'inspirational', 'life', 'humor', 'books', 'reading', 'friendship', 'friends', 'truth', 'simile']}
Lidando com páginas grandes
Quando alimentamos uma página grande no Claude, nos deparamos com restrições de tokens. O Claude permite um limite máximo de 200.000 tokens. Para que o Claude processe partes maiores de dados, precisamos dividir nossos dados em partes. Em seguida, o Claude pode processar cada bloco individualmente.
def chunk_text(text, max_tokens):
"""Split text into sequential chunks based on token limit."""
chunks = []
while text:
# Estimate tokens for the current chunk size
current_chunk = text[:max_tokens * 4] # Rough estimate: 1 token ≈ 4 characters
chunks.append(current_chunk)
text = text[len(current_chunk):] # Move to the next chunk
return chunks
O código acima nos dá um algoritmo primitivo de chunking para trabalhar. Cada pedaço será enviado ao Claude individualmente para processamento.
Claude With Web Unlocker e Residential Proxies
No próximo exemplo, integraremos nosso raspador com inteligência artificial aos proxies da Bright Data para contornar o sistema de bloqueio da Amazon. Isso dá muito menos trabalho do que raspar manualmente a Amazon. Você pode usar o raspador abaixo com o Web Unlocker ou com nossos Residential Proxies.
import re
import requests
from bs4 import BeautifulSoup
import anthropic
import json
ANTHROPIC_API_KEY = "YOUR-ANTHROPIC-API-KEY"
client = anthropic.Anthropic(
api_key=ANTHROPIC_API_KEY,
)
def estimate_tokens(text):
# Rough estimate: 1 token ≈ 4 characters
return len(text) // 4
def clean_html(html):
soup = BeautifulSoup(html, "html.parser")
# Remove script and style elements
for script_or_style in soup(["script", "style"]):
script_or_style.decompose()
# Get the text content only
return soup.get_text(separator=" ", strip=True)
def pull_json_data(claude_text):
json_match = re.search(r"```json\n(.*?)\n```", claude_text, re.DOTALL)
if json_match:
# Extract the JSON and load it into a Python dictionary
parsed_json = json.loads(json_match.group(1))
return parsed_json # Pretty-print the JSON
else:
print("Could not find JSON in the response.")
def chunk_text(text, max_tokens):
"""Split text into sequential chunks based on token limit."""
chunks = []
while text:
# Estimate tokens for the current chunk size
current_chunk = text[:max_tokens * 4] # Rough estimate: 1 token ≈ 4 characters
chunks.append(current_chunk)
text = text[len(current_chunk):] # Move to the next chunk
return chunks
def extract_with_claude(response, token_limit=200000, max_tokens_per_chunk=1024):
"""Process HTML response with Claude by dynamically chunking the text."""
# Estimate tokens and preprocess if necessary
token_estimate = estimate_tokens(response.text)
page_to_parse = response.text
# Clean HTML if it exceeds the token limit
if token_estimate > token_limit:
page_to_parse = clean_html(page_to_parse)
# Chunk the cleaned text
chunks = chunk_text(page_to_parse, max_tokens_per_chunk)
print(f"Chunks to process: {len(chunks)}")
# Process each chunk
results = []
for i, chunk in enumerate(chunks):
print(f"Processing chunk {i + 1}/{len(chunks)}...")
message = client.messages.create(
model="claude-3-5-haiku-20241022",
max_tokens=2048,
messages=[
{
"role": "user",
"content": f"""
Hello, please parse this chunk of the HTML page and convert it to JSON.
Make sure to strip newlines, remove escape characters, and whitespace:
{chunk}
"""
}
]
)
text = message.to_dict()["content"][0]["text"]
try:
parsed_json = pull_json_data(text) # Extract JSON
results.append(parsed_json)
except Exception as e:
print(f"Error processing chunk {i + 1}: {e}")
return results
if __name__ == "__main__":
TARGET_URL = "https://www.amazon.com/s?k=laptops"
PROXY_URL = "http://brd-customer-<YOUR-USERNAME>-zone-<YOUR-ZONE-NAME>:<YOUR-PASSWORD>@brd.superproxy.io:33335"
proxies = {
"http": PROXY_URL,
"https": PROXY_URL
}
response = requests.get(TARGET_URL, proxies=proxies, verify="brd.crt")
json_data = extract_with_claude(response)
with open("output.json", "w") as file:
try:
json.dump(json_data, file, indent=4)
except:
print("Failed to save JSON data")
Esse exemplo é um pouco mais refinado do que o nosso raspador de cotações.
- A Amazon nos fornece páginas de respostas massivas. Usamos nosso algoritmo de fragmentação para dividir a página.
- Alimentamos cada bloco para o Claude por meio do cliente de API para processamento.
- Extraímos o JSON de cada resposta e o anexamos aos nossos
resultados. Quando a análise é concluída, retornamos esses resultados. - Após o scrape, gravamos os resultados em um arquivo JSON.
Aqui está a saída do nosso terminal. A página foi dividida em seis partes para o Claude processar.
Chunks to process: 6
Processing chunk 1/6...
Processing chunk 2/6...
Processing chunk 3/6...
Processing chunk 4/6...
Processing chunk 5/6...
Processing chunk 6/6...
Você pode visualizar os dados JSON extraídos finais abaixo.
[
{
"page_title": "Amazon.com: laptops",
"search_context": {
"total_results": "over 100,000",
"sort_options": [
"Featured",
"Price: Low to High",
"Price: High to Low",
"Avg. Customer Review",
"Newest Arrivals",
"Best Sellers"
]
},
"featured_products": [
{
"brand": "Apple",
"model": "2024 MacBook Pro",
"variants": [
{
"color": "Silver",
"chip": "M4 with 10-core CPU and 10-core GPU",
"display": "14.2-inch Liquid Retina XDR",
"memory": "16GB Unified Memory",
"storage": "512GB SSD",
"rating": 4.8,
"reviews": 341
},
{
"color": "Space Black",
"chip": "M4 with 10-core CPU and 10-core GPU",
"display": "14.2-inch Liquid Retina XDR",
"memory": "16GB Unified Memory",
"storage": "512GB SSD",
"rating": 4.8,
"reviews": 341
},
{
"color": "Silver",
"chip": "M4 with 10-core CPU and 10-core GPU",
"display": "14.2-inch Liquid Retina XDR",
"memory": "24GB Unified Memory",
"storage": "1TB SSD",
"rating": 4.8,
"reviews": 341
}
]
}
],
"recommended_products": [
{
"brand": "HP",
"model": "14 Inch Laptop",
"features": {
"display_size": "14 inches",
"storage": "384GB (128GB eMMC + 256GB MSD)",
"ram": "16GB",
"os": "Windows 11 Pro",
"processor": "Intel Dual-Core N4120"
},
"price": 349.99,
"discount": 30.0,
"rating": 4.5,
"reviews": 849,
"recent_purchases": "500+"
},
{
"brand": "HP",
"model": "14 Laptop",
"features": {
"display_size": "14 inches",
"storage": "64 GB",
"ram": "4 GB",
"os": "Windows 11 Home",
"processor": "Intel Celeron N4020"
},
"price": 167.98,
"original_price": 209.99,
"rating": 4.0,
"reviews": 2290,
"recent_purchases": "10K+"
},
{
"brand": "Lenovo",
"model": "V15 Laptop",
"features": {
"display_size": "15.6\" FHD 1080p",
"storage": "1TB PCIe SSD",
"ram": "32GB",
"os": "Windows 11 Pro",
"processor": "Intel Celeron N4500"
},
"price": 399.99,
"rating": 4.4,
"reviews": 284,
"recent_purchases": "400+"
}
]
},
{
"products": [
{
"name": "16 Inch Gaming Laptop",
"specs": {
"displaySize": "16 inches",
"diskSize": "512GB SSD",
"ram": "16GB",
"operatingSystem": "Windows 11 Pro",
"processor": "Intel 12th Gen N95 Processor(up to 3.4GHz)"
},
"features": [
"Backlit Keyboard",
"Fingerprint Unlock",
"FHD 1920 * 1200"
],
"rating": {
"stars": 4.0,
"totalReviews": 651,
"monthlyPurchases": "300+"
},
"pricing": {
"currentPrice": 279.99,
"typicalPrice": 339.99,
"delivery": {
"type": "FREE",
"dates": [
"Tue, Feb 18",
"Sat, Feb 15"
]
}
}
},
{
"name": "HP 17 Laptop",
"specs": {
"displaySize": "17.3 inches",
"diskSize": "1TB SSD",
"ram": "32GB",
"operatingSystem": "Windows 11 Home",
"processor": "AMD Ryzen 5 Processor(Beats i7-1165G7, Up to 4.3GHz)"
},
"features": [
"Webcam",
"Numeric Keypad",
"Long Battery Life"
],
"rating": {
"stars": 4.0,
"totalReviews": 22,
"monthlyPurchases": "500+"
},
"pricing": {
"currentPrice": 499.99,
"listPrice": 639.0,
"delivery": {
"type": "FREE",
"date": "Tue, Feb 18"
}
}
}
]
},
{
"products": [
{
"name": "Dell Latitude Touch 3190 2-in-1 PC",
"specs": {
"processor": "Intel Quad Core up to 2.4Ghz",
"ram": "4GB",
"storage": "64GB SSD",
"display": "11.6inch HD Touch Gorilla Glass LED",
"connectivity": "WiFi Cam HDMI",
"os": "Windows 10 Pro"
},
"condition": "Renewed",
"rating": {
"stars": 3.9,
"totalReviews": 327
},
"price": 109.99,
"recentPurchases": "1K+",
"sustainabilityFeatures": true
},
{
"name": "HP Pavilion Touchscreen Laptop",
"specs": {
"displaySize": "15.6 inches",
"storage": "1TB SSD",
"ram": "16GB",
"processor": "Intel Core up to 4.1GHz",
"batteryLife": "Up to 11 Hours",
"os": "Windows 11 Home"
},
"rating": {
"stars": 4.2,
"totalReviews": 866
},
"price": 392.0,
"recentPurchases": "1K+",
"stockStatus": "Only 10 left"
}
]
},
{
"laptops": [
{
"brand": "HP",
"model": "Portable Laptop",
"rating": {
"stars": 4.3,
"reviews": 279
},
"price": {
"current": 197.0,
"list": 269.0
},
"specs": {
"displaySize": "14 inches",
"diskSize": "64 GB",
"ram": "4 GB",
"operatingSystem": "Windows 11 S"
},
"features": [
"Student and Business",
"HD Display",
"Intel Quad-Core N4120",
"1 Year Office 365",
"Webcam",
"RJ-45",
"HDMI",
"Wi-Fi"
]
},
{
"brand": "HP",
"model": "Laptop",
"rating": {
"stars": 4.1,
"reviews": 2168
},
"price": {
"current": 207.99
},
"specs": {
"displaySize": "14 inches",
"diskSize": "64 GB",
"ram": "8 GB",
"operatingSystem": "Windows 11 Home"
}
},
{
"brand": "NIMO",
"model": "15.6 FHD-Laptop",
"rating": {
"stars": 4.7,
"reviews": 6
},
"price": {
"current": 499.99,
"typical": 599.99
},
"specs": {
"ram": "32GB",
"storage": "1TB SSD",
"processor": "AMD Ryzen 5 6600H"
},
"features": [
"Gaming Laptop",
"100W Type-C",
"54Wh Battery",
"WiFi 6",
"BT5.2",
"Backlit Keyboard"
]
}
]
},
{
"processorSpeed": [
"1 to 1.59 GHz",
"1.60 to 1.79 GHz",
"1.80 to 1.99 GHz",
"2.00 to 2.49 GHz",
"2.50 to 2.99 GHz",
"3.00 to 3.49 GHz",
"3.50 to 3.99 GHz",
"4.0 GHz & Above"
],
"hardDiskDescription": [
"Emmc",
"HDD",
"SSD",
"SSHD"
],
"connectivityTechnology": [
"Bluetooth",
"Ethernet",
"HDMI",
"USB",
"Wi-Fi"
],
"humanInterface": {
"input": [
"Touch Bar",
"Touch Pad",
"Touchscreen",
"Touchscreen with Stylus Support"
]
},
"graphicsType": [
"Dedicated",
"Integrated"
]
},
{
"services": [
"Groceries & More Right To Your Door",
"AmazonGlobal Ship Orders Internationally",
"Home Services Experienced Pros Happiness Guarantee",
"Amazon Web Services Scalable Cloud Computing Services",
"Audible Listen to Books & Original Audio Performances",
"Box Office Mojo Find Movie Box Office Data",
"Goodreads Book reviews & recommendations",
"IMDb Movies, TV & Celebrities",
"IMDbPro Get Info Entertainment Professionals Need",
"Kindle Direct Publishing Indie Digital & Print Publishing Made Easy",
"Amazon Photos Unlimited Photo Storage Free With Prime",
"Prime Video Direct Video Distribution Made Easy",
"Shopbop Designer Fashion Brands",
"Amazon Resale Great Deals on Quality Used Products",
"Whole Foods Market America's Healthiest Grocery Store",
"Woot! Deals and Shenanigans",
"Zappos Shoes & Clothing",
"Ring Smart Home Security Systems",
"eero WiFi Stream 4K Video in Every Room",
"Blink Smart Security for Every Home",
"Neighbors App Real-Time Crime & Safety Alerts",
"Amazon Subscription Boxes Top subscription boxes \u2013 right to your door",
"PillPack Pharmacy Simplified",
"Amazon Renewed Like-new products you can trust"
],
"legalNotice": {
"conditionsOfUse": "Conditions of Use",
"privacyNotice": "Privacy Notice",
"consumerHealthPrivacy": "Consumer Health Data Privacy Disclosure",
"adPrivacy": "Your Ads Privacy Choices",
"copyright": "\u00a9 1996-2025, Amazon.com, Inc. or its affiliates"
}
}
]
Reclamação com navegador de raspagem
O código abaixo foi ligeiramente modificado em relação ao nosso exemplo anterior. Em vez de chamar response.text, atribuímos driver.page_source diretamente à nossa variável de resposta.
import re
import requests
from bs4 import BeautifulSoup
import anthropic
import json
from selenium.webdriver import Remote, ChromeOptions
from selenium.webdriver.chromium.remote_connection import ChromiumRemoteConnection
ANTHROPIC_API_KEY = "YOUR-ANTHROPIC-API-KEY"
client = anthropic.Anthropic(
api_key=ANTHROPIC_API_KEY,
)
def estimate_tokens(text):
# Rough estimate: 1 token ≈ 4 characters
return len(text) // 4
def clean_html(html):
soup = BeautifulSoup(html, "html.parser")
# Remove script and style elements
for script_or_style in soup(["script", "style"]):
script_or_style.decompose()
# Get the text content only
return soup.get_text(separator=" ", strip=True)
def pull_json_data(claude_text):
json_match = re.search(r"```json\n(.*?)\n```", claude_text, re.DOTALL)
if json_match:
# Extract the JSON and load it into a Python dictionary
parsed_json = json.loads(json_match.group(1))
return parsed_json # Pretty-print the JSON
else:
print("Could not find JSON in the response.")
def chunk_text(text, max_tokens):
"""Split text into sequential chunks based on token limit."""
chunks = []
while text:
# Estimate tokens for the current chunk size
current_chunk = text[:max_tokens * 4] # Rough estimate: 1 token ≈ 4 characters
chunks.append(current_chunk)
text = text[len(current_chunk):] # Move to the next chunk
return chunks
def extract_with_claude(response, token_limit=200000, max_tokens_per_chunk=1024):
"""Process HTML response with Claude by dynamically chunking the text."""
# Estimate tokens and preprocess if necessary
token_estimate = estimate_tokens(response)
page_to_parse = response
# Clean HTML if it exceeds the token limit
if token_estimate > token_limit:
page_to_parse = clean_html(page_to_parse)
# Chunk the cleaned text
chunks = chunk_text(page_to_parse, max_tokens_per_chunk)
print(f"Chunks to process: {len(chunks)}")
# Process each chunk
results = []
for i, chunk in enumerate(chunks):
print(f"Processing chunk {i + 1}/{len(chunks)}...")
message = client.messages.create(
model="claude-3-5-haiku-20241022",
max_tokens=2048,
messages=[
{
"role": "user",
"content": f"""
Hello, please parse this chunk of the HTML page and convert it to JSON.
Make sure to strip newlines, remove escape characters, and whitespace:
{chunk}
"""
}
]
)
text = message.to_dict()["content"][0]["text"]
try:
parsed_json = pull_json_data(text) # Extract JSON
results.append(parsed_json)
except Exception as e:
print(f"Error processing chunk {i + 1}: {e}")
return results
if __name__ == "__main__":
TARGET_URL = "https://www.walmart.com/search?q=laptops"
AUTH = "brd-customer-<YOUR-USERNAME>-zone-<YOUR-ZONE>:<YOUR-PASSWORD>"
SBR_WEBDRIVER = f"https://{AUTH}@brd.superproxy.io:9515"
sbr_connection = ChromiumRemoteConnection(SBR_WEBDRIVER, "goog", "chrome")
response = None
success = False
while not success:
try:
with Remote(sbr_connection, options=ChromeOptions()) as driver:
driver.get(TARGET_URL)
response = driver.page_source
success = True
except Exception as e:
print(f"Failed to get the page: {e}")
json_data = extract_with_claude(response)
with open("scraping-browser-output.json", "w") as file:
try:
json.dump(json_data, file, indent=4)
except:
print("Failed to save JSON data")
Aqui estão os laptops do Walmart extraídos por Claude.
[
{
"laptops": [
{
"brand": "Acer",
"model": "Chromebook 315",
"screen_size": "15.6 inch",
"processor": "Intel Processor N4500",
"ram": "4GB",
"storage": "64GB eMMC",
"color": "Pure Silver/Moonstone Purple",
"os": "ChromeOS",
"price": {
"current": 139.0,
"original": 179.0
},
"reviews": {
"count": 6370,
"rating": 4.4
}
},
{
"brand": "ASUS",
"model": "Chromebook CM30",
"screen_size": "10.5 inch",
"type": "2-in-1 Touch Tablet",
"processor": "MediaTek Kompanio 520",
"ram": "8GB",
"storage": "128GB eMMC",
"color": "Fog Silver",
"extras": "Stylus Included",
"price": {
"current": 299.0
},
"reviews": {
"count": 265,
"rating": 4.4
}
},
{
"brand": "ASUS",
"model": "Chromebook Plus CX34",
"screen_size": "14 inch",
"type": "Touch Laptop",
"processor": "Intel Core i3-1215U",
"ram": "8GB",
"storage": "128GB UFS",
"color": "Gray",
"features": [
"Google AI"
],
"price": {
"current": 329.0,
"original": 399.0
},
"reviews": {
"count": 111,
"rating": 4.6
}
},
{
"brand": "Naclud",
"screen_size": "15.6 inch",
"os": "Windows 11",
"ram": "36GB DDR4",
"storage": "128GB + 1024GB ROM",
"processor": "4 Core Celeron N5095",
"extras": [
"1yr Free Office 365",
"Support 5TB Expansion",
"Copilot"
],
"price": {
"current": 399.19,
"original": 1399.99,
"alternative_from": 329.99
},
"reviews": {
"count": 118,
"rating": 3.8
}
},
{
"brand": "RNRUO",
"screen_size": "14.1 inch",
"os": "Windows 11 Pro",
"type": "Business Laptop",
"ram": "8GB",
"storage": "256GB SSD",
"processor": "2.64 GHz Intel Pentium J3710",
"resolution": "1920x1080 FHD",
"connectivity": [
"WiFi 5",
"BT5.0"
],
"color": "Gray",
"price": {
"current": 180.89,
"original": 498.0
}
}
]
},
{
"laptops": [
{
"name": "Apple MacBook Air 13.3 inch Laptop",
"color": "Silver",
"chip": "M1 Chip",
"features": [
"Built for Apple Intelligence"
],
"specs": {
"ram": "8GB",
"storage": "256GB"
},
"pricing": {
"currentPrice": 629.0,
"originalPrice": 699.0
},
"reviews": {
"count": 5082,
"rating": 4.7
},
"shipping": {
"type": "Free shipping",
"arrivalTime": "3+ days"
}
},
{
"name": "HP 14 inch Windows Laptop",
"color": "Silver",
"processor": "Intel Celeron N4120",
"specs": {
"ram": "4GB",
"storage": "64GB eMMC"
},
"extras": [
"12-mo. Microsoft 365 Included"
],
"pricing": {
"currentPrice": 149.0,
"originalPrice": 249.0
},
"reviews": {
"count": 1061,
"rating": 4.4
},
"shipping": {
"type": "Free shipping",
"arrivalTime": "2 days"
}
}
]
},
[
{
"currentPrice": 449.99,
"originalPrice": 549.0,
"name": "HP Pavilion 16 inch Windows Laptop AMD Ryzen 5-8540U AI PC 8GB RAM 512GB SSD Meteor Silver",
"rating": 4.4,
"reviewCount": 67,
"shipping": "Free shipping, arrives in 3+ days",
"stockStatus": "Only 2 left"
},
{
"currentPrice": 194.95,
"originalPrice": 229.0,
"name": "HP Stream 14 inch Windows Laptop Intel Processor N4120 4GB RAM 64GB eMMC Pink (12-mo. Microsoft 365 included)",
"rating": 4.2,
"reviewCount": 16026,
"shipping": "Save with Free shipping, arrives in 2 days",
"stockStatus": "Only 1 left"
},
{
"currentPrice": 399.19,
"originalPrice": 1399.99,
"name": "Naclud 15.6\" Windows 11 Laptop 36GB DDR4 128+1024GB ROM Computer, 4 Core Celeron N5095, 1yr Free Office 365 Subscription, Support 5TB Expansion, Copilot",
"rating": 3.8,
"reviewCount": 118,
"shipping": "Save with Free shipping, arrives in 2 days",
"stockStatus": "Available"
},
{
"currentPrice": 139.0,
"originalPrice": 199.99,
"name": "Acer Chromebook 315 15.6 inch Laptop Intel Processor N4500 4GB RAM 64GB eMMC Moonstone Purple",
"rating": 4.4,
"reviewCount": 6370,
"shipping": "Save with Free pickup today, Delivery today, Free shipping, arrives tomorrow",
"stockStatus": "Only 2 left"
},
{
"currentPrice": 265.79,
"originalPrice": 599.0,
"name": "SANPTENT 15.6 inch 1080p FHD Laptop Computer 16GB RAM 512GB SSD with 4 Core Intel Celeron N5095, FingerPrint, Backlit Keyboard, Windows 11 Pro",
"rating": 4.0,
"reviewCount": 359,
"shipping": "Save with Free shipping, arrives in 2 days",
"stockStatus": "Available"
},
{
"currentPrice": 94.0,
"originalPrice": 139.99,
"name": "Restored HP Chromebook 2024 OS 11.6-inch Intel Celeron 1.6GHz 4GB RAM 16GB SSD Bundle: Wireless Mouse, Bluetooth/Wireless Airbuds By 2 Day Express (Refurbished)",
"rating": 3.9,
"reviewCount": 547,
"shipping": "Free shipping, arrives in 2 days",
"stockStatus": "Available"
},
{
"currentPrice": 792.99,
"originalPrice": 999.99,
"name": "DELL Inspiron 3520 15.6\" Touchscreen i7 Laptop, Intel Core i7-1255U, 32GB RAM, 1TB SSD, Numeric Keypad, Webcam, SD Card Reader, HDMI, Wi-Fi, Windows 11 Pro",
"rating": 1.0,
"reviewCount": 1,
"shipping": "Save with Free shipping, arrives in 2 days",
"stockStatus": "Only 6 left"
},
{
"currentPrice": 279.09,
"originalPrice": 329.0,
"name": "Laptop 15.6 FHD 16GB 512GB Intel Quad-Core 12th Alder Lake N97 with Windows 11 Pro",
"rating": 4.6,
"reviewCount": 1698,
"shipping": "Free shipping, arrives in 3+ days",
"stockStatus": "Available"
},
{
"currentPrice": 279.99,
"originalPrice": null,
"name": "HP 14\" HD Laptop for Students and Business, Intel Quad-Core Processor, 4GB RAM, 64GB eMMC+256GB Micro SD, Long Battery Life, UHD Graphics, Webcam, Windows 11 Home in S Mode, Snowflake White",
"rating": 4.3,
"reviewCount": 350,
"shipping": "Free shipping, arrives in 3+ days",
"stockStatus": "Available"
},
{
"currentPrice": 244.9,
"originalPrice": 379.0,
"name": "HP 15.6 inch Windows Laptop Intel Processor N200 4GB RAM 128GB UFS Scarlet Red (12-mo. Microsoft 365 included)",
"rating": null,
"reviewCount": null,
"shipping": null,
"stockStatus": null
}
],
{
"laptops": [
{
"brand": "HP",
"model": "15.6 inch Windows Laptop",
"processor": "Intel Processor N200",
"ram": "4GB",
"storage": "128GB UFS",
"color": "Scarlet Red",
"price": {
"options": {
"min": 244.9,
"max": 249.0
}
},
"reviews": {
"count": 6192,
"rating": 4.3
},
"extras": "12-mo. Microsoft 365 included",
"shipping": "Free shipping, arrives in 3+ days"
},
{
"brand": "HP",
"model": "15.6 inch Windows Laptop",
"processor": "Intel Core i3-N305",
"ram": "8GB",
"storage": "256GB SSD",
"color": "Natural Silver",
"price": {
"current": 304.98,
"options": {
"min": 304.98,
"max": 329.0
}
},
"reviews": {
"count": 2001,
"rating": 4.5
},
"shipping": "Free shipping, arrives in 2 days"
}
]
},
{
"laptops": [
{
"name": "HP 14 inch x360 FHD Touch Chromebook Laptop",
"specs": {
"processor": "Intel Processor N100",
"ram": "4GB",
"storage": "64GB eMMC",
"color": "Sky Blue"
},
"price": {
"current": 269.0,
"original": null
},
"reviews": {
"count": 1437,
"rating": 4.5
},
"shipping": "Free shipping, arrives in 3+ days"
},
{
"name": "SANPTENT 16 inch Windows 11 Pro Laptop",
"specs": {
"processor": "4 Core Intel Alder Lake N95",
"ram": "16GB",
"storage": "512GB SSD",
"screen": "1920x1200 FHD IPS"
},
"price": {
"current": 279.38,
"original": 699.0
},
"reviews": {
"count": 352,
"rating": 3.9
},
"shipping": "Save with Free shipping, arrives in 2 days"
}
]
},
{
"laptops": [
{
"name": "ASUS Vivobook Go 15.6 inch Windows Laptop",
"specs": {
"processor": "Intel Core i3-N305",
"ram": "8GB",
"storage": "256GB UFS",
"color": "Black"
},
"price": {
"current": 282.0,
"original": null
},
"reviews": {
"count": 330,
"rating": 4.4
},
"shipping": "Free shipping, arrives in 2 days"
},
{
"name": "Latest 16\" Purple Laptop",
"specs": {
"processor": "12th Gen Alder Lake N95 CPU",
"ram": "12G LPDDR5",
"storage": "1T NVMe SSD",
"os": "Win 11 Pro/Office 2019"
},
"price": {
"current": 377.99,
"original": null,
"other_options_from": 369.99
},
"reviews": {
"count": 7,
"rating": 4.4
},
"shipping": "Free shipping, arrives in 2 days"
}
]
},
{
"products": [
"hp laptop",
"macbook",
"gaming laptop",
"printer laptop",
"touchscreen",
"wireless mouse",
"ipad",
"chromebook",
"mouse"
],
"pagination": {
"current": [
1,
2,
3
],
"total": 25
},
"footer_links": [
"Departments",
"Store Directory",
"Careers",
"Our Company",
"Sell on Walmart.com",
"Help",
"Product Recalls",
"Accessibility",
"Tax Exempt Program",
"Get the Walmart App",
"Sign-up for Email",
"Safety Data Sheet",
"Terms of Use",
"Privacy & Security",
"California Supply Chain Act",
"Your Privacy Choices",
"Notice at Collection",
"AdChoices",
"Consumer Health Data Privacy Notices",
"Brand Shop Directory",
"Pharmacy",
"Walmart Business"
],
"copyright": "\u00a9 2025 Walmart. All Rights Reserved."
}
]
Conclusão
Ao permitir que o Claude analise páginas com a infraestrutura da Bright Data, podemos reduzir drasticamente o tempo que gastamos escrevendo analisadores e lidando com bloqueios e proibições de IP. A única coisa com que você precisa se preocupar é a página. Depois de obter a página, você pode entregá-la ao Claude para ser concluída. Seu raspador geral pode levar alguns minutos a mais para ser executado, mas isso não é nada comparado às horas que gastamos escrevendo nossos próprios analisadores.
Deseja experimentar os produtos da Bright Data? Registre-se agora e comece sua avaliação gratuita!
Não é necessário cartão de crédito
