Python如何爬取微信公众号文章和评论(基于 Fiddler 抓包分析)

2019-11-25 12:35:46

字体：大中小

来源：转载

供稿：网友

背景说明

感觉微信公众号算得是比较难爬的平台之一，不过一番折腾之后还是小有收获的。没有用Scrapy(估计爬太快也有反爬限制)，但后面会开始整理写一些实战出来。简单介绍下本次的开发环境：

python3
requests
psycopg2 (操作postgres数据库)

抓包分析

本次实战对抓取的公众号没有限制，但不同公众号每次抓取之前都要进行分析。打开Fiddler，将手机配置好相关代理，为避免干扰过多，这里给Fiddler加个过滤规则，只需要指定微信域名mp.weixin.qq.com就好：

Fiddler配置Filter规则

平时关注的公众号也比较多，本次实战以“36氪”公众号为例，继续往下看：

“36氪”公众号

公众号右上角 -> 全部消息

在公众号主页，右上角有三个实心圆点，点击进入消息界面，下滑找到并点击“全部消息”，往下请求加载几次历史文章，然后回到Fiddler界面，不出意外的话应该可以看到这几次请求，可以看到返回的数据是json格式的，同时文章数据是以json字符串的形式定义在general_msg_list字段中：

公众号文章列表抓包请求

分析文章列表接口

把请求URL和Cookie贴上来进行分析：

https://mp.weixin.qq.com/mp/profile_ext?action=getmsg&__biz=MzI2NDk5NzA0Mw==&f=json&offset=10&count=10&is_ok=1&scene=126&uin=777&key=777&pass_ticket=QhOypNwH5dAr5w6UgMjyBrTSOdMEUT86vWc73GANoziWFl8xJd1hIMbMZ82KgCpN&wxtoken=&appmsg_token=971_LwY7Z%252BFBoaEv5z8k_dFWfJkdySbNkMR4OmFxNw~~&x5=1&f=jsonCookie: pgv_pvid=2027337976; pgv_info=ssid=s3015512850; rewardsn=; wxtokenkey=777; wxuin=2089823341; devicetype=android-26; version=26070237; lang=zh_CN;pass_ticket=NDndxxaZ7p6Z9PYulWpLqMbI0i3ULFeCPIHBFu1sf5pX2IhkGfyxZ6b9JieSYRUy;wap_sid2=CO3YwOQHEogBQnN4VTNhNmxQWmc3UHI2U3kteWhUeVExZHFVMnN0QXlsbzVJRUJKc1pkdVFUU2Y5UzhSVEtOZmt1VVlYTkR4SEllQ2huejlTTThJWndMQzZfYUw2SldLVGVMQUthUjc3QWdVMUdoaGN0Nml2SU05cXR1dTN2RkhRUVd1V2Y3SFJ5d01BQUF+fjCB1pLcBTgNQJVO

下面把重要的参数说明一下，没提到的说明就不那么重要了：

__biz：相当于是当前公众号的id(唯一固定标志)
offset：文章数据接口请求偏移量标志(从0开始)，每次返回的json数据中会有下一次请求的offset，注意这里并不是按某些规则递增的
count：每次请求的数据量(亲测最多可以是10)
pass_ticket：可以理解是请求票据，而且隔一段时间后(大概几个小时)就会过期，这也是为什么微信公众号比较难按固定规则进行抓取的原因
appmsg_token：同样理解为非固定有过期策略的票据
Cookie：使用的时候可以把整段贴上去，但最少仅需要wap_sid2这部分

是不是感觉有点麻烦，毕竟不是要搞大规模专业的爬虫，所以单就一个公众号这么分析下来，还是可以往下继续的，贴上截取的一段json数据，用于设计文章数据表：

{"ret": 0,"errmsg": "ok","msg_count": 10,"can_msg_continue": 1,"general_msg_list": "{/"list/":[{/"comm_msg_info/":{/"id/":1000005700,/"type/":49,/"datetime/":1535100943,/"fakeid/":/"3264997043/",/"status/":2,/"content/":/"/"},/"app_msg_ext_info/":{/"title/":/"金融危机又十年：钱荒之下，二手基金迎来高光时刻/",/"digest/":/"退出永远是基金的主旋律。/",/"content/":/"/",/"fileid/":100034824,/"content_url/":/"http://////mp.weixin.qq.com///s?__biz=MzI2NDk5NzA0Mw==&mid=2247518479&idx=1&sn=124ab52f7478c1069a6b4592cdf3c5f5&chksm=eaa6d8d3ddd151c5bb95a7ae118de6d080023246aa0a419e1d53bfe48a8d9a77e52b752d9b80&scene=27#wechat_redirect/",/"source_url/":/"/",/"cover/":/"http://////mmbiz.qpic.cn///mmbiz_jpg///QicyPhNHD5vYgdpprkibtnWCAN7l4ZaqibKvopNyCWWLQAwX7QpzWicnQSVfcBZmPrR5YuHS45JIUzVjb0dZTiaLPyA///0?wx_fmt=jpeg/",/"subtype/":9,/"is_multi/":0,/"multi_app_msg_item_list/":[],/"author/":/"石亚琼/",/"copyright_stat/":11,/"duration/":0,/"del_flag/":1,/"item_show_type/":0,/"audio_fileid/":0,/"play_url/":/"/",/"malicious_title_reason_id/":0,/"malicious_content_type/":0}}]}","next_offset": 20,"video_count": 1,"use_video_tab": 1,"real_type": 0}

可以简单抽取想要的数据，这里将文章表结构定义如下，顺便贴上建表的SQL语句：

文章数据表

-- ------------------------------ Table structure for tb_article-- ----------------------------DROP TABLE IF EXISTS "public"."tb_article";CREATE TABLE "public"."tb_article" ("id" serial4 PRIMARY KEY,"msg_id" int8 NOT NULL,"title" varchar(200) COLLATE "pg_catalog"."default" NOT NULL,"author" varchar(20) COLLATE "pg_catalog"."default","cover" varchar(500) COLLATE "pg_catalog"."default","digest" varchar(200) COLLATE "pg_catalog"."default","source_url" varchar(800) COLLATE "pg_catalog"."default","content_url" varchar(600) COLLATE "pg_catalog"."default" NOT NULL,"post_time" timestamp(6),"create_time" timestamp(6) NOT NULL);COMMENT ON COLUMN "public"."tb_article"."id" IS '自增主键';COMMENT ON COLUMN "public"."tb_article"."msg_id" IS '消息id (唯一)';COMMENT ON COLUMN "public"."tb_article"."title" IS '标题';COMMENT ON COLUMN "public"."tb_article"."author" IS '作者';COMMENT ON COLUMN "public"."tb_article"."cover" IS '封面图';COMMENT ON COLUMN "public"."tb_article"."digest" IS '关键字';COMMENT ON COLUMN "public"."tb_article"."source_url" IS '原文地址';COMMENT ON COLUMN "public"."tb_article"."content_url" IS '文章地址';COMMENT ON COLUMN "public"."tb_article"."post_time" IS '发布时间';COMMENT ON COLUMN "public"."tb_article"."create_time" IS '入库时间';COMMENT ON TABLE "public"."tb_article" IS '公众号文章表';-- ------------------------------ Indexes structure for table tb_article-- ----------------------------CREATE UNIQUE INDEX "unique_msg_id" ON "public"."tb_article" USING btree ("msg_id" "pg_catalog"."int8_ops" ASC NULLS LAST);

附请求文章接口并解析数据保存到数据库的相关代码：

class WxMps(object):"""微信公众号文章、评论抓取爬虫"""def __init__(self, _biz, _pass_ticket, _app_msg_token, _cookie, _offset=0):self.offset = _offsetself.biz = _biz # 公众号标志self.msg_token = _app_msg_token # 票据(非固定)self.pass_ticket = _pass_ticket # 票据(非固定)self.headers = {'Cookie': _cookie, # Cookie(非固定)'User-Agent': 'Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko) Version/4.0 Chrome/57.0.2987.132 '}wx_mps = 'wxmps' # 这里数据库、用户、密码一致(需替换成实际的)self.postgres = pgs.Pgs(host='localhost', port='5432', db_name=wx_mps, user=wx_mps, password=wx_mps)def start(self):"""请求获取公众号的文章接口"""offset = self.offsetwhile True:api = 'https://mp.weixin.qq.com/mp/profile_ext?action=getmsg&__biz={0}&f=json&offset={1}' /'&count=10&is_ok=1&scene=124&uin=777&key=777&pass_ticket={2}&wxtoken=&appmsg_token' /'={3}&x5=1&f=json'.format(self.biz, offset, self.pass_ticket, self.msg_token)resp = requests.get(api, headers=self.headers).json()ret, status = resp.get('ret'), resp.get('errmsg') # 状态信息if ret == 0 or status == 'ok':print('Crawl article: ' + api)offset = resp['next_offset'] # 下一次请求偏移量general_msg_list = resp['general_msg_list']msg_list = json.loads(general_msg_list)['list'] # 获取文章列表for msg in msg_list:comm_msg_info = msg['comm_msg_info'] # 该数据是本次推送多篇文章公共的msg_id = comm_msg_info['id'] # 文章idpost_time = datetime.fromtimestamp(comm_msg_info['datetime']) # 发布时间# msg_type = comm_msg_info['type'] # 文章类型# msg_data = json.dumps(comm_msg_info, ensure_ascii=False) # msg原数据app_msg_ext_info = msg.get('app_msg_ext_info') # article原数据if app_msg_ext_info:# 本次推送的首条文章self._parse_articles(app_msg_ext_info, msg_id, post_time)# 本次推送的其余文章multi_app_msg_item_list = app_msg_ext_info.get('multi_app_msg_item_list')if multi_app_msg_item_list:for item in multi_app_msg_item_list:msg_id = item['fileid'] # 文章idif msg_id == 0:msg_id = int(time.time() * 1000) # 设置唯一id,解决部分文章id=0出现唯一索引冲突的情况self._parse_articles(item, msg_id, post_time)print('next offset is %d' % offset)else:print('Before break , Current offset is %d' % offset)breakdef _parse_articles(self, info, msg_id, post_time):"""解析嵌套文章数据并保存入库"""title = info.get('title') # 标题cover = info.get('cover') # 封面图author = info.get('author') # 作者digest = info.get('digest') # 关键字source_url = info.get('source_url') # 原文地址content_url = info.get('content_url') # 微信地址# ext_data = json.dumps(info, ensure_ascii=False) # 原始数据self.postgres.handler(self._save_article(), (msg_id, title, author, cover, digest,source_url, content_url, post_time,datetime.now()), fetch=True)@staticmethoddef _save_article():sql = 'insert into tb_article(msg_id,title,author,cover,digest,source_url,content_url,post_time,create_time) ' /'values(%s,%s,%s,%s,%s,%s,%s,%s,%s)'return sql if __name__ == '__main__':biz = 'MzI2NDk5NzA0Mw==' # "36氪"pass_ticket = 'NDndxxaZ7p6Z9PYulWpLqMbI0i3ULFeCPIHBFu1sf5pX2IhkGfyxZ6b9JieSYRUy'app_msg_token = '971_Z0lVNQBcGsWColSubRO9H13ZjrPhjuljyxLtiQ~~'cookie = 'wap_sid2=CO3YwOQHEogBQnN4VTNhNmxQWmc3UHI2U3kteWhUeVExZHFVMnN0QXlsbzVJRUJKc1pkdVFUU2Y5UzhSVEtOZmt1VVlYTkR4SEllQ2huejlTTThJWndMQzZfYUw2SldLVGVMQUthUjc3QWdVMUdoaGN0Nml2SU05cXR1dTN2RkhRUVd1V2Y3SFJ5d01BQUF+fjCB1pLcBTgNQJVO'# 以上信息不同公众号每次抓取都需要借助抓包工具做修改wxMps = WxMps(biz, pass_ticket, app_msg_token, cookie)wxMps.start() # 开始爬取文章

分析文章评论接口

获取评论的思路大致是一样的，只是会更加麻烦一点。首先在手机端点开一篇有评论的文章，然后查看Fiddler抓取的请求：

公众号文章评论

公众号文章评论接口抓包请求

提取其中的URL和Cookie再次分析：

https://mp.weixin.qq.com/mp/appmsg_comment?action=getcomment&scene=0&__biz=MzI2NDk5NzA0Mw==&appmsgid=2247518723&idx=1&comment_id=433253969406607362&offset=0&limit=100&uin=777&key=777&pass_ticket=NDndxxaZ7p6Z9PYulWpLqMbI0i3ULFeCPIHBFu1sf5pX2IhkGfyxZ6b9JieSYRUy&wxtoken=777&devicetype=android-26&clientversion=26070237&appmsg_token=971_dLK7htA1j8LbMUk8pvJKRlC_o218HEgwDbS9uARPOyQ34_vfXv3iDstqYnq2gAyze1dBKm4ZMTlKeyfx&x5=1&f=jsonCookie: pgv_pvid=2027337976; pgv_info=ssid=s3015512850; rewardsn=; wxuin=2089823341; devicetype=android-26; version=26070237; lang=zh_CN; pass_ticket=NDndxxaZ7p6Z9PYulWpLqMbI0i3ULFeCPIHBFu1sf5pX2IhkGfyxZ6b9JieSYRUy; wap_sid2=CO3YwOQHEogBdENPSVdaS3pHOWc1V2QzY1NvZG9PYk1DMndPS3NfbGlHM0Vfal8zLU9kcUdkWTQxdUYwckFBT3RZM1VYUXFaWkFad3NVaWFXZ28zbEFIQ2pTa1lqZktfb01vcGdPLTQ0aGdJQ2xOSXoxTVFvNUg3SVpBMV9GRU1lbnotci1MWWl5d01BQUF+fjCj45PcBTgNQAE=; wxtokenkey=777

接着分析参数：

__biz：同上
pass_ticket：同上
Cookie：同上
offset和limit：代表偏移量和请求数量，由于公众号评论最多展示100条，所以这两个参数也不用改它
comment_id：获取文章评论数据的标记id，固定但需要从当前文章结构(Html)解析提取
appmsgid：票据id，非固定每次需要从当前文章结构(Html)解析提取
appmsg_token：票据token，非固定每次需要从当前文章结构(Html)解析提取

可以看到最后三个参数要解析html获取(当初真的找了好久才想到看文章网页结构)。从文章请求接口可以获得文章地址，对应上面的content_url字段，但请求该地址前仍需要对url做相关处理，不然上面三个参数会有缺失，也就获取不到后面评论内容：

def _parse_article_detail(self, content_url, article_id):"""从文章页提取相关参数用于获取评论,article_id是已保存的文章id"""try:api = content_url.replace('amp;', '').replace('#wechat_redirect', '').replace('http', 'https')html = requests.get(api, headers=self.headers).textexcept:print('获取评论失败' + content_url)else:# group(0) is current linestr_comment = re.search(r'var comment_id = "(.*)" /|/| "(.*)" /* 1;', html)str_msg = re.search(r"var appmsgid = '' /|/| '(.*)'/|/|", html)str_token = re.search(r'window.appmsg_token = "(.*)";', html)if str_comment and str_msg and str_token:comment_id = str_comment.group(1) # 评论id(固定)app_msg_id = str_msg.group(1) # 票据id(非固定)appmsg_token = str_token.group(1) # 票据token(非固定)

再回来看该接口返回的json数据，分析结构后然后定义数据表(含SQL)：

文章评论数据表

-- ------------------------------ Table structure for tb_article_comment-- ----------------------------DROP TABLE IF EXISTS "public"."tb_article_comment";CREATE TABLE "public"."tb_article_comment" ("id" serial4 PRIMARY KEY,"article_id" int4 NOT NULL,"comment_id" varchar(50) COLLATE "pg_catalog"."default","nick_name" varchar(50) COLLATE "pg_catalog"."default" NOT NULL,"logo_url" varchar(300) COLLATE "pg_catalog"."default","content_id" varchar(50) COLLATE "pg_catalog"."default" NOT NULL,"content" varchar(3000) COLLATE "pg_catalog"."default" NOT NULL,"like_num" int2,"comment_time" timestamp(6),"create_time" timestamp(6) NOT NULL);COMMENT ON COLUMN "public"."tb_article_comment"."id" IS '自增主键';COMMENT ON COLUMN "public"."tb_article_comment"."article_id" IS '文章外键id';COMMENT ON COLUMN "public"."tb_article_comment"."comment_id" IS '评论接口id';COMMENT ON COLUMN "public"."tb_article_comment"."nick_name" IS '用户昵称';COMMENT ON COLUMN "public"."tb_article_comment"."logo_url" IS '头像地址';COMMENT ON COLUMN "public"."tb_article_comment"."content_id" IS '评论id (唯一)';COMMENT ON COLUMN "public"."tb_article_comment"."content" IS '评论内容';COMMENT ON COLUMN "public"."tb_article_comment"."like_num" IS '点赞数';COMMENT ON COLUMN "public"."tb_article_comment"."comment_time" IS '评论时间';COMMENT ON COLUMN "public"."tb_article_comment"."create_time" IS '入库时间';COMMENT ON TABLE "public"."tb_article_comment" IS '公众号文章评论表';-- ------------------------------ Indexes structure for table tb_article_comment-- ----------------------------CREATE UNIQUE INDEX "unique_content_id" ON "public"."tb_article_comment" USING btree ("content_id" COLLATE "pg_catalog"."default" "pg_catalog"."text_ops" ASC NULLS LAST);

万里长征快到头了，最后贴上这部分代码，由于要先获取文章地址，所以和上面获取文章数据的代码是一起的：

import jsonimport reimport timefrom datetime import datetimeimport requestsfrom utils import pgsclass WxMps(object):"""微信公众号文章、评论抓取爬虫"""def __init__(self, _biz, _pass_ticket, _app_msg_token, _cookie, _offset=0):self.offset = _offsetself.biz = _biz # 公众号标志self.msg_token = _app_msg_token # 票据(非固定)self.pass_ticket = _pass_ticket # 票据(非固定)self.headers = {'Cookie': _cookie, # Cookie(非固定)'User-Agent': 'Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko) Version/4.0 Chrome/57.0.2987.132 '}wx_mps = 'wxmps' # 这里数据库、用户、密码一致(需替换成实际的)self.postgres = pgs.Pgs(host='localhost', port='5432', db_name=wx_mps, user=wx_mps, password=wx_mps)def start(self):"""请求获取公众号的文章接口"""offset = self.offsetwhile True:api = 'https://mp.weixin.qq.com/mp/profile_ext?action=getmsg&__biz={0}&f=json&offset={1}' /'&count=10&is_ok=1&scene=124&uin=777&key=777&pass_ticket={2}&wxtoken=&appmsg_token' /'={3}&x5=1&f=json'.format(self.biz, offset, self.pass_ticket, self.msg_token)resp = requests.get(api, headers=self.headers).json()ret, status = resp.get('ret'), resp.get('errmsg') # 状态信息if ret == 0 or status == 'ok':print('Crawl article: ' + api)offset = resp['next_offset'] # 下一次请求偏移量general_msg_list = resp['general_msg_list']msg_list = json.loads(general_msg_list)['list'] # 获取文章列表for msg in msg_list:comm_msg_info = msg['comm_msg_info'] # 该数据是本次推送多篇文章公共的msg_id = comm_msg_info['id'] # 文章idpost_time = datetime.fromtimestamp(comm_msg_info['datetime']) # 发布时间# msg_type = comm_msg_info['type'] # 文章类型# msg_data = json.dumps(comm_msg_info, ensure_ascii=False) # msg原数据app_msg_ext_info = msg.get('app_msg_ext_info') # article原数据if app_msg_ext_info:# 本次推送的首条文章self._parse_articles(app_msg_ext_info, msg_id, post_time)# 本次推送的其余文章multi_app_msg_item_list = app_msg_ext_info.get('multi_app_msg_item_list')if multi_app_msg_item_list:for item in multi_app_msg_item_list:msg_id = item['fileid'] # 文章idif msg_id == 0:msg_id = int(time.time() * 1000) # 设置唯一id,解决部分文章id=0出现唯一索引冲突的情况self._parse_articles(item, msg_id, post_time)print('next offset is %d' % offset)else:print('Before break , Current offset is %d' % offset)breakdef _parse_articles(self, info, msg_id, post_time):"""解析嵌套文章数据并保存入库"""title = info.get('title') # 标题cover = info.get('cover') # 封面图author = info.get('author') # 作者digest = info.get('digest') # 关键字source_url = info.get('source_url') # 原文地址content_url = info.get('content_url') # 微信地址# ext_data = json.dumps(info, ensure_ascii=False) # 原始数据content_url = content_url.replace('amp;', '').replace('#wechat_redirect', '').replace('http', 'https')article_id = self.postgres.handler(self._save_article(), (msg_id, title, author, cover, digest,source_url, content_url, post_time,datetime.now()), fetch=True)if article_id:self._parse_article_detail(content_url, article_id)def _parse_article_detail(self, content_url, article_id):"""从文章页提取相关参数用于获取评论,article_id是已保存的文章id"""try:html = requests.get(content_url, headers=self.headers).textexcept:print('获取评论失败' + content_url)else:# group(0) is current linestr_comment = re.search(r'var comment_id = "(.*)" /|/| "(.*)" /* 1;', html)str_msg = re.search(r"var appmsgid = '' /|/| '(.*)'/|/|", html)str_token = re.search(r'window.appmsg_token = "(.*)";', html)if str_comment and str_msg and str_token:comment_id = str_comment.group(1) # 评论id(固定)app_msg_id = str_msg.group(1) # 票据id(非固定)appmsg_token = str_token.group(1) # 票据token(非固定)# 缺一不可if appmsg_token and app_msg_id and comment_id:print('Crawl article comments: ' + content_url)self._crawl_comments(app_msg_id, comment_id, appmsg_token, article_id)def _crawl_comments(self, app_msg_id, comment_id, appmsg_token, article_id):"""抓取文章的评论"""api = 'https://mp.weixin.qq.com/mp/appmsg_comment?action=getcomment&scene=0&__biz={0}' /'&appmsgid={1}&idx=1&comment_id={2}&offset=0&limit=100&uin=777&key=777' /'&pass_ticket={3}&wxtoken=777&devicetype=android-26&clientversion=26060739' /'&appmsg_token={4}&x5=1&f=json'.format(self.biz, app_msg_id, comment_id,self.pass_ticket, appmsg_token)resp = requests.get(api, headers=self.headers).json()ret, status = resp['base_resp']['ret'], resp['base_resp']['errmsg']if ret == 0 or status == 'ok':elected_comment = resp['elected_comment']for comment in elected_comment:nick_name = comment.get('nick_name') # 昵称logo_url = comment.get('logo_url') # 头像comment_time = datetime.fromtimestamp(comment.get('create_time')) # 评论时间content = comment.get('content') # 评论内容content_id = comment.get('content_id') # idlike_num = comment.get('like_num') # 点赞数# reply_list = comment.get('reply')['reply_list'] # 回复数据self.postgres.handler(self._save_article_comment(), (article_id, comment_id, nick_name, logo_url,content_id, content, like_num, comment_time,datetime.now()))@staticmethoddef _save_article():sql = 'insert into tb_article(msg_id,title,author,cover,digest,source_url,content_url,post_time,create_time) ' /'values(%s,%s,%s,%s,%s,%s,%s,%s,%s) returning id'return sql@staticmethoddef _save_article_comment():sql = 'insert into tb_article_comment(article_id,comment_id,nick_name,logo_url,content_id,content,like_num,' /'comment_time,create_time) values(%s,%s,%s,%s,%s,%s,%s,%s,%s)'return sqlif __name__ == '__main__':biz = 'MzI2NDk5NzA0Mw==' # "36氪"pass_ticket = 'NDndxxaZ7p6Z9PYulWpLqMbI0i3ULFeCPIHBFu1sf5pX2IhkGfyxZ6b9JieSYRUy'app_msg_token = '971_Z0lVNQBcGsWColSubRO9H13ZjrPhjuljyxLtiQ~~'cookie = 'wap_sid2=CO3YwOQHEogBQnN4VTNhNmxQWmc3UHI2U3kteWhUeVExZHFVMnN0QXlsbzVJRUJKc1pkdVFUU2Y5UzhSVEtOZmt1VVlYTkR4SEllQ2huejlTTThJWndMQzZfYUw2SldLVGVMQUthUjc3QWdVMUdoaGN0Nml2SU05cXR1dTN2RkhRUVd1V2Y3SFJ5d01BQUF+fjCB1pLcBTgNQJVO'# 以上信息不同公众号每次抓取都需要借助抓包工具做修改wxMps = WxMps(biz, pass_ticket, app_msg_token, cookie)wxMps.start() # 开始爬取文章及评论

文末小结

最后展示下数据库里的数据，单线程爬的慢而且又没这方面的数据需求，所以也只是随便试了下手：