# coding=utf-8 import asyncio import json import re from functools import partial, wraps from json import JSONDecodeError from glom import glom from playwright.async_api import Browser from abs_spider import AbstractAiSeoSpider from domain.ai_seo import AiAnswer, AiSearchResult from utils import create_logger, css_to_dict from utils.image_utils import crop_image_left logger = create_logger(__name__) class DeepseekSpider(AbstractAiSeoSpider): def __init__(self, browser: Browser, prompt: str, keyword: str): super().__init__(browser, prompt, keyword) self.__listen_response = self.handle_listen_response_error(self.__listen_response) def get_home_url(self) -> str: return 'https://chat.deepseek.com/' def get_platform_id(self) -> int: return 1 def get_platform_name(self) -> str: return 'DeepSeek' async def _do_spider(self) -> AiAnswer: self._init_data() self.search_result_count = 0 await self.browser_page.goto(self.get_home_url(), timeout=600000) await asyncio.sleep(3) # 开启联网搜索 search_btn = self.browser_page.locator("span:text('联网搜索')").locator('..') if await search_btn.is_visible(): await search_btn.click() if self.think: # 开启深度思考 think_btn = self.browser_page.locator("span:text('深度思考 (R1)')").locator('..') if await think_btn.is_visible(): styles = css_to_dict(await think_btn.get_attribute('style')) if styles.get('--ds-button-color') == '#fff': await think_btn.click() await asyncio.sleep(1) chat_input_element = self.browser_page.locator("//textarea[@id='chat-input']") await chat_input_element.click() # 输入提问词 await self.browser_page.keyboard.type(self.prompt) await asyncio.sleep(1) await self.browser_page.keyboard.press('Enter') # 监听请求 self.browser_page.on('response', partial(self.__listen_response)) await self.completed_event.wait() # 报错检查 if self.fail_status: raise self.fail_exception # 打开搜索栏 search_btn_text = f'已搜索到 {self.search_result_count} 个网页' search_btn = self.browser_page.locator(f"div:text('{search_btn_text}')") # search_btn = self.browser_page.locator('div:has-text("搜索到")') if await search_btn.count() > 0: await search_btn.click() await asyncio.sleep(2) if self.think: # 思考元素 think_element = self.browser_page.locator("text=已深度思考（") think_element_count = await think_element.count() if think_element_count > 0: await think_element.nth(-1).click() await asyncio.sleep(2) # 获取回答元素 answer = self.browser_page.locator("//div[@class='ds-markdown ds-markdown--block']").nth(-1) box = await answer.bounding_box() # 设置视口大小 await self.browser_page.set_viewport_size({ 'width': 1920, 'height': int(box['height']) + 500 }) # 截图 screenshot_path = self._get_screenshot_path() await self.browser_page.screenshot(path=screenshot_path) # 切割图片 crop_image_left(screenshot_path, 250) self.ai_answer.screenshot_file = screenshot_path return self.ai_answer def handle_listen_response_error(self, func): """ 装饰器 用于处理请求回调中的异常 :param func: :return: """ @wraps(func) async def wrapper(*args, **kwargs): try: return await func(*args, **kwargs) except Exception as e: logger.error(f"DeepSeek响应异常: {e}", exc_info=True) # 标记失败状态 记录异常 self.fail_status = True self.fail_exception = e self.completed_event.set() return wrapper async def __listen_response(self, response): if '/api/v0/chat/completion' not in response.url: return # 读取流式数据 response_text = '' search_result_lists = list() start_content = False stream = await response.body() body = stream.decode('utf-8') datas = body.split("\n\n") for data_str in datas: if not data_str: continue data_str = data_str.replace('data: ', '') try: data = json.loads(data_str) if glom(data, 'v.0.v', default='') == 'TIMEOUT': self.fail_status = True logger.error("DeepSeek服务器繁忙") except JSONDecodeError as e: continue # 获取ai搜索结果 if data.get('p', '') == 'response/search_results' or isinstance(data.get('v', ''), list): logger.debug(f"获取到联网搜索结果") search_result_list = data.get('v', []) # 保存搜索结果 ai_search_result_list = [] for search_result in search_result_list: url = search_result.get('url', '') title = search_result.get('title', '') body = search_result.get('snippet', '') publish_time = search_result.get('published_at', '') host_name = search_result.get('site_name', '未知') ai_result = AiSearchResult(url=url, title=title, body=body, publish_time=publish_time, host_name=host_name) if ai_result.title and ai_result.url: ai_search_result_list.append(ai_result) logger.debug(f"ai参考资料: [{host_name}]{title}({url})") if ai_search_result_list: self.ai_answer.search_result = ai_search_result_list self.search_result_count = len(self.ai_answer.search_result) continue # 是否开始返回回复数据 if data.get('p', '') == 'response/content': start_content = True if start_content: # 获取ai回复 value = data.get('v', None) if isinstance(value, dict): continue if value is None: target = 'choices.0.delta.content' value = glom(data, target, default="") response_text = response_text + str(value) #匹配citation:中的数字 citation = list() citations = re.findall(r'citation:(\d+)', response_text) if citations: citation = list(set(citations)) # 保存搜索结果 ai_search_result_list = [] for index,search_result in enumerate(search_result_lists): url = search_result.get('url', '') title = search_result.get('title', '') body = search_result.get('snippet', '') publish_time = search_result.get('published_at', '') host_name = search_result.get('site_name', '未知') if str(index+1) in citation: is_referenced = "1" else: is_referenced = "0" ai_result = AiSearchResult(url=url, title=title, body=body, publish_time=publish_time, host_name=host_name, is_referenced=is_referenced) if ai_result.title and ai_result.url: ai_search_result_list.append(ai_result) logger.debug(f"ai参考资料: [{host_name}]{title}({url})") if ai_search_result_list: self.ai_answer.search_result = ai_search_result_list self.search_result_count = len(self.ai_answer.search_result) logger.debug(response_text) self.ai_answer.answer = response_text self.completed_event.set()