- 已编辑
晒泥版来啦：
一个爬网页的练习：看看 R 邮件列表中最热门的讨论是什么
yihui
爬好了，如下：
- r-devel:
- r-help:
- 源代码： https://github.com/jienagu/tidyverse_examples/blob/master/web_scraping_r_help.R
- csv 文件： https://github.com/jienagu/tidyverse_examples/blob/master/web_scraping_r_help.csv
- 前十名：
> everything2[1:10,] link Link_url count 1 [R] Problems in Recommending R https://stat.ethz.ch/pipermail/r-help/2009-February/379744.html 126 2 [R] R in the NY Times https://stat.ethz.ch/pipermail/r-help/2009-January/377079.html 118 3 [R] Google's R Style Guide https://stat.ethz.ch/pipermail/r-help/2009-August/402727.html 100 4 [R] How to comment in R https://stat.ethz.ch/pipermail/r-help/2009-February/380938.html 78 5 [R] Inefficiency of SAS Programming https://stat.ethz.ch/pipermail/r-help/2009-February/382798.html 66 6 [R] installing R on Ubuntu https://stat.ethz.ch/pipermail/r-help/2009-February/380559.html 64 7 [R] Popularity of R, SAS, SPSS, Stata... https://stat.ethz.ch/pipermail/r-help/2010-June/243043.html 64 8 [R] A comment about R: https://stat.ethz.ch/pipermail/r-help/2006-January/085205.html 63 9 [R] productivity tools in R? https://stat.ethz.ch/pipermail/r-help/2009-July/396050.html 56 10 [R] two questions for R beginners https://stat.ethz.ch/pipermail/r-help/2010-March/230116.html 55
- 已编辑
基于 Jiena 的代码，来个依赖最小的版本
# 安装必要的依赖
packages <- c("rvest", "knitr")
lapply(packages, function(pkg) {
if (system.file(package = pkg) == "") install.packages(pkg)
})
# 确保 Windows 下的中文环境也能获取正确的日期格式化结果
Sys.setlocale("LC_TIME", "C")
# 格式化日期序列
all_months <- format(
seq(
from = as.Date("1997-04-01"),
to = Sys.Date(), by = "1 month"
),
"%Y-%B"
)
# 清理帖子主题
clean_discuss_topic <- function(x) {
# 去掉中括号及其内容
x <- gsub("(\\[.*?\\])", "", x)
# 去掉末尾换行符 \n
x <- gsub("(\\\n)$", "", x)
# 两个以上的空格替换为一个空格
x <- gsub("( {2,})", " ", x)
x
}
library(magrittr)
x <- "2019-February"
base_url <- "https://stat.ethz.ch/pipermail/r-devel"
# 下面的部分可以打包成一个函数
# 输入是日期 x 输出是一个 markdown 表格
# 抓取当月的数据
scrap_webpage <- xml2::read_html(paste(base_url, x, "subject.html", sep = "/"))
# Extract the URLs 提取链接尾部
tail_url <- scrap_webpage %>%
rvest::html_nodes("a") %>%
rvest::html_attr("href")
# Extract the theme 提取链接对应的讨论主题
discuss_topic <- scrap_webpage %>%
rvest::html_nodes("a") %>%
rvest::html_text()
# url 和 讨论主题合并为数据框
discuss_df <- data.frame(discuss_topic = discuss_topic, tail_url = tail_url)
# 清理无效的帖子记录
discuss_df <- discuss_df[grepl(pattern = "\\.html$", x = discuss_df$tail_url), ]
# 清理帖子主题内容
discuss_df$discuss_topic <- clean_discuss_topic(discuss_df$discuss_topic)
# 去重 # 只保留第一条发帖记录
discuss_uni_df <- discuss_df[!duplicated(discuss_df$discuss_topic), ]
# 分组计数
discuss_count_df <- as.data.frame(table(discuss_df$discuss_topic), stringsAsFactors = FALSE)
# 对 discuss_count_df 的列重命名
colnames(discuss_count_df) <- c("discuss_topic", "count")
# 按讨论主题合并数据框
discuss <- merge(discuss_uni_df, discuss_count_df, by = "discuss_topic")
# 添加完整的讨论帖的 url
discuss <- transform(discuss, full_url = paste(base_url, x, tail_url, sep = "/"))
# 选取讨论主题、主题链接和楼层高度
discuss <- discuss[, c("discuss_topic", "full_url", "count")]
# 按楼层高度排序，转化为 Markdown 表格形式输出
discuss[order(discuss$count, decreasing = TRUE), ] %>%
knitr::kable(format = "markdown", row.names = FALSE) %>%
cat(file = paste0(x, "-disuss.md"), sep = "\n")总依赖如下，现在差不多可以往 Travis 上搞定时任务了，只要把输出的 markdown 文件推回 Github 即可
tools::package_dependencies(packages,recursive=T) %>% unlist %>% unique
[1] "xml2" "httr" "magrittr" "selectr" "curl" "jsonlite"
[7] "mime" "openssl" "R6" "methods" "stringr" "Rcpp"
[13] "tools" "askpass" "utils" "glue" "stringi" "sys"
[19] "stats" "evaluate" "highr" "markdown" "yaml" "xfun" 我还不知道论坛里怎么贴 Markdwon 表格，即文件 2019-February-disuss.md 的内容，所以请移步 <https://github.com/XiangyunHuang/RGraphics/issues/5>
如果将所有的帖子都扒拉下来，根据帖子主题之间的关系，有没有可能将它们分类，用 shiny 做一个可视化前端，根据时间和楼层数显示每类帖子下面热门的讨论
如果把每个帖子的发帖人和回帖人也提取，那么还可以看特定的人的情况
根据帖子的 ID 长度来看是六位数，不足百万，SO 上面已经是八位数了
- 已编辑
yihui 晒泥版升级了， 只在打开时载入一次，滑块仅用来筛选时间区间。
另外，画了个简单的柱状图，显示每个月的回帖总量。
本论坛应该是不支持表格语法吧。
--- <https://d.cosx.org/d/420350-pagedown-pdf/9>
yanglei
没事，问的好。建议先试 r-devel.R，数量小一些。 以下是侦察过程：
- 建立好年月数列，看看
trans_month是不是这个格式，比如“1997-April”：#Loading the rvest package library('rvest') library(purrr) library(tidyverse) ## Create year month list current_month=as.Date(cut(Sys.Date(), "month")) date_seq=seq(from=as.Date("1997-04-01"),to=current_month , by="1 month") trans_month=format(as.Date(date_seq), "%Y-%B") ### Make a function to fetch the summarized data frame for every month unnecessary=c("thread.html#start", "author.html#start", "date.html#start", "http://www.stat.math.ethz.ch/mailman/listinfo/r-devel")- 先以 1997 年 4 月为例子在本地跑：
这样一步一步跑哪里出错一清二楚。最后看看您的#Specifying the url for desired website to be scraped url_base <- 'https://stat.ethz.ch/pipermail/r-devel/%s/subject.html' test_web=read_html(sprintf(url_base, '1997-April' ) ) # Extract the URLs url_ <- test_web %>% rvest::html_nodes("a") %>% rvest::html_attr("href") # Extract the link text link_ <- test_web %>% rvest::html_nodes("a") %>% rvest::html_text() test_df=data.frame(link = link_, url = url_) test_df2=test_df[which(!is.na(test_df$url)),] test_df3=test_df2[which(! test_df2$url %in% unnecessary),] test_df3$link=gsub("[\r\n]", " ", test_df3$link) test_df3$url=paste0("https://stat.ethz.ch/pipermail/r-devel/", "1997-April","/",test_df3$url) test_df4=test_df3 %>% group_by(link) %>% dplyr::mutate( Link_url = dplyr::first(url) ) %>% group_by(link,Link_url ) %>% summarize( count=n() ) test_df4=data.frame(test_df4)test_df4是不是如下：> head(test_df4) link Link_url count 1 R-alpha: contributed packages -- Yes, use library/<package>/.. ! https://stat.ethz.ch/pipermail/r-devel/1997-April/017072.html 2 2 R-alpha: ==NULL https://stat.ethz.ch/pipermail/r-devel/1997-April/017131.html 1 3 R-alpha: as.numeric https://stat.ethz.ch/pipermail/r-devel/1997-April/017023.html 3 4 R-alpha: Be nice to the English https://stat.ethz.ch/pipermail/r-devel/1997-April/017051.html 1 5 R-alpha: binom.test https://stat.ethz.ch/pipermail/r-devel/1997-April/017075.html 1 6 R-alpha: Bug & Patch in dbeta.c (0.50 - PreR 7) https://stat.ethz.ch/pipermail/r-devel/1997-April/017048.html 3