使用R爬取Glassdoor医院评论：特定字段提取技术求助

以下是针对你需求的R代码优化版本，可精准提取Pros、Cons、管理层建议、推荐意愿、CEO认可度、业务前景及分项小评分：

library(rvest)
library(tidyverse)
library(stringr)

# 目标页面URL
url <- "https://www.glassdoor.com/Reviews/Montefiore-Nyack-Hospital-Reviews-E2312619.htm?sort.sortType=RD&sort.ascending=false&filter.iso3Language=eng"
page <- read_html(url)

# 先抓取所有评论节点，避免内容错位
review_nodes <- page %>% html_nodes("li.empReview")

# 1. 评论标题
review_titles <- review_nodes %>% 
  html_node(".reviewLink") %>% 
  html_text(trim = TRUE)

# 2. 评论日期（清理前缀）
review_dates <- review_nodes %>% 
  html_node(".middle.common__EiReviewDetailsStyle__newGrey") %>% 
  html_text(trim = TRUE) %>% 
  str_remove("Reviewed ")

# 3. Pros
review_pros <- review_nodes %>% 
  html_node("[data-test='pros']") %>% 
  html_text(trim = TRUE)

# 4. Cons
review_cons <- review_nodes %>% 
  html_node("[data-test='cons']") %>% 
  html_text(trim = TRUE)

# 5. 给管理层的建议
review_advice <- review_nodes %>% 
  html_node("[data-test='advice-management']") %>% 
  html_text(trim = TRUE)

# 6. 总体评分
review_ratings <- review_nodes %>% 
  html_node(".ratingNumber.mr-xsm") %>% 
  html_text(trim = TRUE) %>% 
  as.numeric()

# 7. 推荐意愿（转换为数值：推荐=1，不推荐=-1，中立=0）
recommendations_text <- review_nodes %>% 
  html_node(".recommends") %>% 
  html_text(trim = TRUE)
recommendations_numeric <- case_when(
  str_detect(recommendations_text, "Recommend") ~ 1,
  str_detect(recommendations_text, "Not Recommend") ~ -1,
  TRUE ~ 0
)

# 8. CEO认可度（提取百分比）
ceo_approval <- review_nodes %>% 
  html_node("[data-test='ceo']") %>% 
  html_text(trim = TRUE) %>% 
  str_extract("\\d+%")

# 9. 业务前景
business_outlook <- review_nodes %>% 
  html_node("[data-test='business-outlook']") %>% 
  html_text(trim = TRUE)

# 10. 分项小评分（工作生活平衡、薪酬福利等5项）
sub_ratings <- review_nodes %>% 
  html_nodes(".subRatings .ratingNumber") %>% 
  html_text(trim = TRUE) %>% 
  as.numeric() %>% 
  matrix(ncol = 5, byrow = TRUE)
sub_ratings_df <- as.data.frame(sub_ratings) %>% 
  set_names(c("WorkLifeBalance", "CompBenefits", "JobSecurity", "Management", "Culture"))

# 合并所有数据为完整数据框
reviews <- tibble(
  Title = review_titles,
  Date = review_dates,
  OverallRating = review_ratings,
  Pros = review_pros,
  Cons = review_cons,
  AdviceToManagement = review_advice,
  Recommend = recommendations_numeric,
  CEOApproval = ceo_approval,
  BusinessOutlook = business_outlook
) %>% 
  bind_cols(sub_ratings_df)

# 查看结果
print(reviews)

关键说明 #

• 使用data-test属性选择器：Glassdoor为核心评论元素设置了该属性，比动态class更稳定，不易随页面更新失效
• 基于单评论节点提取：先抓取所有评论的父节点，再从每个节点内提取字段，彻底避免不同评论内容错位
• 文本清理：对日期、百分比等字段做了格式整理，直接得到可用数据
• 分项评分处理：将连续的小评分数值按固定列数拆分为数据框，匹配对应分项名称

内容的提问来源于stack exchange，提问作者AndrooGC