C++ TensorRT 部署 SAM：从 .pth 到实时分割引擎
2026/10/3 4:18:19
引言：学术数据采集的重要性和挑战
在当今的数字化研究时代，学术论文摘要作为科研成果的核心浓缩，对研究人员、学术机构和企业研发部门具有不可估量的价值。传统的手动收集方式效率低下，无法满足大规模数据需求，而动态网页技术和反爬机制的普及使得数据采集变得愈发复杂。本文将深入探讨如何构建一个高效、稳定且智能化的学术论文摘要采集系统，采用最新的Python爬虫技术栈。
技术架构演进：从传统爬虫到现代化解决方案
1. 技术选型对比
传统方案：Requests + BeautifulSoup，适合静态页面
动态渲染挑战：越来越多的学术网站采用JavaScript动态加载
现代化方案：Playwright + Asyncio + Scrapy框架，完美应对复杂场景
2. 核心技术组件
Playwright：微软开源的自动化测试工具，支持多浏览器、无头模式
Asyncio：Python原生异步I/O框架，实现高并发采集
Scrapy框架：成熟的爬虫框架，提供完整的爬虫生命周期管理
Redis：分布式任务队列和去重存储
MongoDB：非结构化数据存储，适应多变的论文数据结构
完整代码实现：智能学术论文摘要采集系统
环境配置与依赖安装
python
# requirements.txt playwright==1.40.0 scrapy==2.11.0 scrapy-playwright==0.0.33 asyncio==3.4.3 aiohttp==3.9.1 redis==5.0.1 pymongo==4.6.0 pydantic==2.5.0 selenium==4.16.0 beautifulsoup4==4.12.2 pandas==2.1.4 numpy==1.26.2 lxml==4.9.3 fake-useragent==1.4.0 rate-limiter==1.0.0
1. 基于Playwright的智能爬虫核心类
python
import asyncio import logging from typing import List, Dict, Optional, Any from dataclasses import dataclass from datetime import datetime from urllib.parse import urlparse, urljoin, urlencode import hashlib from playwright.async_api import async_playwright, Page, BrowserContext from pydantic import BaseModel, Field, validator from fake_useragent import UserAgent import aiohttp import redis from pymongo import MongoClient # 数据模型定义 class AcademicPaper(BaseModel): """学术论文数据模型""" paper_id: str = Field(..., description="论文唯一标识") title: str = Field(..., description="论文标题") abstract: str = Field(..., description="论文摘要") authors: List[str] = Field(default_factory=list, description="作者列表") keywords: List[str] = Field(default_factory=list, description="关键词") publication_date: Optional[str] = Field(None, description="发表日期") journal: Optional[str] = Field(None, description="期刊名称") doi: Optional[str] = Field(None, description="DOI标识") url: str = Field(..., description="原文链接") citations: Optional[int] = Field(None, description="引用次数") download_count: Optional[int] = Field(None, description="下载次数") crawl_timestamp: datetime = Field(default_factory=datetime.now, description="爬取时间") @validator('abstract') def validate_abstract_length(cls, v): if len(v) < 50: raise ValueError('Abstract too short') return v.strip() def generate_id(self) -> str: """生成论文唯一ID""" content = f"{self.title}{self.doi if self.doi else self.url}" return hashlib.md5(content.encode()).hexdigest() # 异步浏览器管理器 class AsyncBrowserManager: """管理Playwright浏览器实例的异步上下文管理器""" def __init__(self, headless: bool = True, proxy: Optional[str] = None): self.headless = headless self.proxy = proxy self.playwright = None self.browser = None self.context = None async def __aenter__(self): self.playwright = await async_playwright().start() launch_options = { 'headless': self.headless, 'proxy': {'server': self.proxy} if self.proxy else None, 'args': [ '--disable-blink-features=AutomationControlled', '--disable-dev-shm-usage', '--no-sandbox', '--disable-setuid-sandbox', '--disable-web-security', '--disable-features=IsolateOrigins,site-per-process', ] } self.browser = await self.playwright.chromium.launch(**launch_options) # 创建上下文，模拟真实浏览器 self.context = await self.browser.new_context( viewport={'width': 1920, 'height': 1080}, user_agent=UserAgent().random, locale='en-US', timezone_id='America/New_York', permissions=['geolocation'], ignore_https_errors=True ) # 添加Stealth.js防检测 await self.context.add_init_script(""" // 覆盖webdriver属性 Object.defineProperty(navigator, 'webdriver', { get: () => undefined }); // 覆盖chrome属性 window.chrome = { runtime: {}, loadTimes: function() {}, csi: function() {}, app: {} }; // 覆盖plugins属性 Object.defineProperty(navigator, 'plugins', { get: () => [1, 2, 3, 4, 5] }); // 覆盖languages属性 Object.defineProperty(navigator, 'languages', { get: () => ['en-US', 'en'] }); """) return self.context async def __aexit__(self, exc_type, exc_val, exc_tb): if self.context: await self.context.close() if self.browser: await self.browser.close() if self.playwright: await self.playwright.stop() # 智能页面解析器 class AcademicPageParser: """智能解析学术论文页面的解析器""" @staticmethod async def parse_abstract_from_page(page: Page, url: str) -> Optional[AcademicPaper]: """从页面解析论文摘要信息""" try: # 等待页面关键元素加载 await page.wait_for_load_state('networkidle') await page.wait_for_timeout(2000) # 额外等待动态内容 # 尝试多种选择器定位摘要 selectors = [ 'div.abstract', '.abstract-text', 'section#abstract', '[class*="abstract"]', 'div.article-abstract', 'meta[name="description"]', 'meta[property="og:description"]' ] abstract = None for selector in selectors: try: element = await page.query_selector(selector) if element: abstract = await element.text_content() if abstract and len(abstract) > 50: break except: continue # 获取论文标题 title_selectors = ['h1.title', 'h1.article-title', 'title', 'meta[property="og:title"]'] title = None for selector in title_selectors: try: element = await page.query_selector(selector) if element: title = await element.text_content() if selector != 'meta[property="og:title"]' \ else await element.get_attribute('content') if title: break except: continue # 获取作者信息 authors = [] author_selectors = ['.authors', '.author-list', 'meta[name="citation_authors"]'] for selector in author_selectors: try: elements = await page.query_selector_all(selector) if elements: for elem in elements: author_text = await elem.text_content() if author_text: authors.extend([a.strip() for a in author_text.split(',')]) if authors: break except: continue # 获取关键词 keywords = [] keyword_selectors = ['.keywords', 'meta[name="keywords"]', 'meta[name="citation_keywords"]'] for selector in keyword_selectors: try: element = await page.query_selector(selector) if element: keyword_text = await element.text_content() if not selector.startswith('meta') \ else await element.get_attribute('content') if keyword_text: keywords = [k.strip() for k in keyword_text.split(',')] break except: continue if abstract and title: paper = AcademicPaper( paper_id=hashlib.md5(f"{title}{url}".encode()).hexdigest(), title=title.strip(), abstract=abstract.strip(), authors=[a for a in authors if a], keywords=keywords, url=url ) # 提取额外元数据 await AcademicPageParser.extract_metadata(page, paper) return paper except Exception as e: logging.error(f"解析页面 {url} 时出错: {str(e)}") return None @staticmethod async def extract_metadata(page: Page, paper: AcademicPaper): """提取论文的元数据信息""" try: # 提取DOI doi_selectors = ['meta[name="citation_doi"]', 'meta[name="doi"]', '.doi'] for selector in doi_selectors: element = await page.query_selector(selector) if element: doi = await element.get_attribute('content') if selector.startswith('meta') \ else await element.text_content() if doi: paper.doi = doi.strip() break # 提取发表日期 date_selectors = ['meta[name="citation_publication_date"]', 'meta[name="date"]', '.publication-date'] for selector in date_selectors: element = await page.query_selector(selector) if element: date_str = await element.get_attribute('content') if selector.startswith('meta') \ else await element.text_content() if date_str: paper.publication_date = date_str.strip() break # 提取期刊信息 journal_selectors = ['meta[name="citation_journal_title"]', '.journal-title'] for selector in journal_selectors: element = await page.query_selector(selector) if element: journal = await element.get_attribute('content') if selector.startswith('meta') \ else await element.text_content() if journal: paper.journal = journal.strip() break except Exception as e: logging.warning(f"提取元数据失败: {str(e)}") # 分布式任务队列 class RedisTaskQueue: """基于Redis的分布式任务队列""" def __init__(self, redis_url: str = 'redis://localhost:6379'): self.redis_client = redis.from_url(redis_url, decode_responses=True) self.queue_key = 'academic:crawler:queue' self.visited_key = 'academic:crawler:visited' self.failed_key = 'academic:crawler:failed' async def add_url(self, url: str, priority: int = 0): """添加URL到任务队列""" score = time.time() - priority * 1000 # 优先级越高，分数越小 await self.redis_client.zadd(self.queue_key, {url: score}) async def get_next_url(self) -> Optional[str]: """获取下一个要爬取的URL""" # 使用原子操作获取并移除最高优先级URL script = """ local result = redis.call('ZRANGE', KEYS[1], 0, 0) if result[1] then redis.call('ZREM', KEYS[1], result[1]) return result[1] end return nil """ return await self.redis_client.eval(script, 1, self.queue_key) async def mark_visited(self, url: str): """标记URL为已访问""" await self.redis_client.sadd(self.visited_key, url) async def is_visited(self, url: str) -> bool: """检查URL是否已访问""" return await self.redis_client.sismember(self.visited_key, url) async def mark_failed(self, url: str, reason: str): """标记失败的URL""" await self.redis_client.hset(self.failed_key, url, reason) # 主爬虫类 class AcademicPaperCrawler: """学术论文摘要爬虫主类""" def __init__(self, config: Dict[str, Any]): self.config = config self.task_queue = RedisTaskQueue(config.get('redis_url')) self.mongo_client = MongoClient(config.get('mongo_url')) self.db = self.mongo_client[config.get('database', 'academic_papers')] self.collection = self.db.papers # 统计信息 self.stats = { 'total_crawled': 0, 'successful': 0, 'failed': 0, 'start_time': datetime.now() } # 配置浏览器 self.browser_config = { 'headless': config.get('headless', True), 'proxy': config.get('proxy'), 'timeout': config.get('timeout', 30000) } # 请求限流 self.rate_limiter = asyncio.Semaphore(config.get('concurrent_limit', 5)) # 设置日志 logging.basicConfig( level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s', handlers=[ logging.FileHandler('crawler.log'), logging.StreamHandler() ] ) self.logger = logging.getLogger(__name__) async def crawl_url(self, url: str) -> Optional[AcademicPaper]: """爬取单个URL""" # 检查是否已访问 if await self.task_queue.is_visited(url): self.logger.info(f"URL已访问: {url}") return None async with self.rate_limiter: async with AsyncBrowserManager( headless=self.browser_config['headless'], proxy=self.browser_config['proxy'] ) as context: try: # 创建新页面 page = await context.new_page() # 设置超时 page.set_default_timeout(self.browser_config['timeout']) # 导航到URL self.logger.info(f"正在爬取: {url}") response = await page.goto(url, wait_until='networkidle') if response.status >= 400: self.logger.error(f"HTTP错误 {response.status}: {url}") await self.task_queue.mark_failed(url, f"HTTP {response.status}") return None # 解析页面 parser = AcademicPageParser() paper = await parser.parse_abstract_from_page(page, url) if paper: # 保存到数据库 await self.save_paper(paper) await self.task_queue.mark_visited(url) self.stats['successful'] += 1 self.logger.info(f"成功爬取: {paper.title}") return paper else: self.logger.warning(f"无法解析页面内容: {url}") await self.task_queue.mark_failed(url, "解析失败") self.stats['failed'] += 1 await page.close() except Exception as e: self.logger.error(f"爬取失败 {url}: {str(e)}") await self.task_queue.mark_failed(url, str(e)) self.stats['failed'] += 1 return None return None async def save_paper(self, paper: AcademicPaper): """保存论文到MongoDB""" try: # 检查是否已存在 existing = await self.collection.find_one({'paper_id': paper.paper_id}) if existing: # 更新现有记录 await self.collection.update_one( {'paper_id': paper.paper_id}, {'$set': paper.dict()} ) self.logger.debug(f"更新论文: {paper.title}") else: # 插入新记录 await self.collection.insert_one(paper.dict()) self.logger.debug(f"保存新论文: {paper.title}") except Exception as e: self.logger.error(f"保存论文失败 {paper.title}: {str(e)}") async def discover_links(self, page: Page, base_url: str) -> List[str]: """从页面发现新的论文链接""" links = [] try: # 查找可能的论文链接 link_selectors = [ 'a[href*="abstract"]', 'a[href*="paper"]', 'a[href*="article"]', 'a[href*="doi.org"]', '.article-title a', '.title a' ] for selector in link_selectors: elements = await page.query_selector_all(selector) for elem in elements: href = await elem.get_attribute('href') if href: full_url = urljoin(base_url, href) # 过滤非学术链接 if self.is_academic_url(full_url): links.append(full_url) except Exception as e: self.logger.error(f"发现链接失败: {str(e)}") return list(set(links)) # 去重 def is_academic_url(self, url: str) -> bool: """判断是否为学术URL""" academic_domains = [ 'arxiv.org', 'ieee.org', 'springer.com', 'sciencedirect.com', 'nature.com', 'science.org', 'wiley.com', 'tandfonline.com', 'aclweb.org', 'acm.org', 'researchgate.net', 'semanticscholar.org', 'arxiv-vanity.com', 'doi.org', 'pubmed.ncbi.nlm.nih.gov' ] domain = urlparse(url).netloc.lower() return any(academic_domain in domain for academic_domain in academic_domains) async def run(self, start_urls: List[str], max_pages: int = 100): """运行爬虫主循环""" # 添加起始URL到队列 for url in start_urls: await self.task_queue.add_url(url, priority=10) self.logger.info(f"爬虫启动，目标{max_pages}个页面") while self.stats['total_crawled'] < max_pages: url = await self.task_queue.get_next_url() if not url: self.logger.info("任务队列为空") break # 爬取页面 paper = await self.crawl_url(url) self.stats['total_crawled'] += 1 # 进度显示 if self.stats['total_crawled'] % 10 == 0: self.print_progress() # 发现新链接 if paper and self.stats['total_crawled'] < max_pages: async with AsyncBrowserManager() as context: page = await context.new_page() try: await page.goto(url, wait_until='networkidle') new_links = await self.discover_links(page, url) # 添加新链接到队列 for link in new_links[:5]: # 限制每个页面发现的数量 if not await self.task_queue.is_visited(link): await self.task_queue.add_url(link) except Exception as e: self.logger.error(f"发现新链接失败: {str(e)}") finally: await page.close() # 避免请求过于频繁 await asyncio.sleep(self.config.get('delay', 1)) self.print_final_stats() def print_progress(self): """打印爬取进度""" elapsed = datetime.now() - self.stats['start_time'] rate = self.stats['total_crawled'] / elapsed.total_seconds() * 60 print(f"\n{'='*50}") print(f"已爬取: {self.stats['total_crawled']} 个页面") print(f"成功: {self.stats['successful']}, 失败: {self.stats['failed']}") print(f"爬取速率: {rate:.2f} 页/分钟") print(f"运行时间: {elapsed}") print(f"{'='*50}") def print_final_stats(self): """打印最终统计信息""" elapsed = datetime.now() - self.stats['start_time'] print(f"\n{'#'*60}") print(f"爬虫运行完成!") print(f"{'#'*60}") print(f"总运行时间: {elapsed}") print(f"总爬取页面: {self.stats['total_crawled']}") print(f"成功爬取: {self.stats['successful']}") print(f"失败: {self.stats['failed']}") print(f"成功率: {self.stats['successful']/self.stats['total_crawled']*100:.2f}%") print(f"{'#'*60}") # Scrapy集成版本 import scrapy from scrapy_playwright.page import PageMethod from scrapy.http import Response, Request class AcademicSpider(scrapy.Spider): """Scrapy版本的学术论文爬虫""" name = 'academic_papers' custom_settings = { 'DOWNLOAD_HANDLERS': { "http": "scrapy_playwright.handler.ScrapyPlaywrightDownloadHandler", "https": "scrapy_playwright.handler.ScrapyPlaywrightDownloadHandler", }, 'TWISTED_REACTOR': "twisted.internet.asyncioreactor.AsyncioSelectorReactor", 'PLAYWRIGHT_BROWSER_TYPE': 'chromium', 'PLAYWRIGHT_LAUNCH_OPTIONS': { 'headless': True, 'args': ['--disable-blink-features=AutomationControlled'] }, 'CONCURRENT_REQUESTS': 10, 'AUTOTHROTTLE_ENABLED': True, 'AUTOTHROTTLE_START_DELAY': 1, 'DOWNLOAD_DELAY': 2, 'ITEM_PIPELINES': { 'academic_pipeline.MongoDBPipeline': 300, } } def start_requests(self): """起始请求""" urls = [ 'https://arxiv.org/list/cs.AI/recent', 'https://dl.acm.org/', 'https://ieeexplore.ieee.org/Xplore/home.jsp' ] for url in urls: yield scrapy.Request( url, callback=self.parse, meta={ 'playwright': True, 'playwright_page_methods': [ PageMethod('wait_for_selector', 'div.abstract'), ], 'playwright_include_page': True } ) async def parse(self, response: Response): """解析页面""" page = response.meta['playwright_page'] # 使用Playwright页面对象 papers = await page.query_selector_all('.paper-item') for paper_elem in papers: # 提取论文信息 title = await paper_elem.query_selector('.title') abstract = await paper_elem.query_selector('.abstract') if title and abstract: item = { 'title': await title.text_content(), 'abstract': await abstract.text_content(), 'url': response.url } yield item # 发现分页链接 next_page = await page.query_selector('.next-page') if next_page: next_url = await next_page.get_attribute('href') if next_url: yield scrapy.Request( url=response.urljoin(next_url), callback=self.parse, meta={ 'playwright': True, 'playwright_page': page, 'playwright_page_methods': [ PageMethod('wait_for_selector', 'div.abstract'), ] } ) await page.close() # MongoDB数据管道 class MongoDBPipeline: """MongoDB数据存储管道""" def __init__(self, mongo_uri, mongo_db): self.mongo_uri = mongo_uri self.mongo_db = mongo_db @classmethod def from_crawler(cls, crawler): return cls( mongo_uri=crawler.settings.get('MONGO_URI'), mongo_db=crawler.settings.get('MONGO_DATABASE', 'academic') ) def open_spider(self, spider): self.client = MongoClient(self.mongo_uri) self.db = self.client[self.mongo_db] def close_spider(self, spider): self.client.close() def process_item(self, item, spider): self.db.papers.update_one( {'title': item['title']}, {'$set': dict(item)}, upsert=True ) return item # 主程序入口 async def main(): """主程序入口""" # 配置参数 config = { 'redis_url': 'redis://localhost:6379/0', 'mongo_url': 'mongodb://localhost:27017/', 'database': 'academic_papers', 'headless': True, 'concurrent_limit': 5, 'delay': 1, 'timeout': 30000, 'proxy': None # 可配置代理: 'http://user:pass@proxy:port' } # 起始URL列表 start_urls = [ 'https://arxiv.org/list/cs.AI/recent', 'https://dl.acm.org/', 'https://scholar.google.com/', 'https://www.semanticscholar.org/', 'https://pubmed.ncbi.nlm.nih.gov/' ] # 创建爬虫实例 crawler = AcademicPaperCrawler(config) try: # 运行爬虫 await crawler.run(start_urls, max_pages=50) # 导出数据 await export_data_to_csv(config) except KeyboardInterrupt: print("\n用户中断，正在保存进度...") except Exception as e: print(f"爬虫运行失败: {str(e)}") finally: print("爬虫程序结束") async def export_data_to_csv(config): """导出数据到CSV文件""" import pandas as pd client = MongoClient(config['mongo_url']) db = client[config['database']] collection = db.papers # 查询所有数据 cursor = collection.find({}) papers = list(cursor) if papers: # 转换为DataFrame df = pd.DataFrame(papers) # 移除MongoDB的_id字段 if '_id' in df.columns: df.drop('_id', axis=1, inplace=True) # 保存到CSV timestamp = datetime.now().strftime('%Y%m%d_%H%M%S') filename = f'academic_papers_{timestamp}.csv' df.to_csv(filename, index=False, encoding='utf-8-sig') print(f"数据已导出到: {filename}") print(f"总共导出 {len(df)} 条记录") client.close() if __name__ == '__main__': # 运行异步主程序 asyncio.run(main())系统优化与高级特性
1. 反反爬虫策略
python
class AntiAntiCrawler: """反反爬虫策略管理器""" def __init__(self): self.user_agents = UserAgent() self.proxy_pool = [ 'http://proxy1:port', 'http://proxy2:port', # ... 代理池 ] def rotate_user_agent(self): """轮换User-Agent""" return self.user_agents.random def get_random_proxy(self): """获取随机代理""" return random.choice(self.proxy_pool) if self.proxy_pool else None def human_like_mouse_movement(self, page: Page): """模拟人类鼠标移动""" # 实现随机鼠标移动模式 pass def random_delay(self, min_seconds=1, max_seconds=3): """随机延迟""" time.sleep(random.uniform(min_seconds, max_seconds))
2. 分布式部署方案
python
# docker-compose.yml 配置 version: '3.8' services: redis: image: redis:alpine ports: - "6379:6379" mongodb: image: mongo:latest ports: - "27017:27017" volumes: - mongo_data:/data/db crawler: build: . depends_on: - redis - mongodb environment: - REDIS_URL=redis://redis:6379/0 - MONGO_URL=mongodb://mongodb:27017/ deploy: replicas: 3 # 启动3个爬虫实例