I was (for the most part) able to accomplish this by using the latest version of the Youtube Data API and the R package httr. The basic approach I took was to send multiple GET requests to the appropriate URL and grab the data in batches of 100 (the maximum the API allows) - i.e.
base_url <- "https://www.googleapis.com/youtube/v3/commentThreads/"
api_opts <- list(
part = "snippet",
maxResults = 100,
textFormat = "plainText",
videoId = "4H9pTgQY_mo",
key = "my_google_developer_api_key",
fields = "items,nextPageToken",
orderBy = "published")
where key is your actual Google Developer key, of course.
The initial batch is retrieved like this:
init_results <- httr::content(httr::GET(base_url, query = api_opts))
##
R> names(init_results)
#[1] "nextPageToken" "items"
R> init_results$nextPageToken
#[1] "Cg0Q-YjT3bmSxQIgACgBEhQIABDI3ZWQkbzEAhjVneqH75u4AhgCIGQ="
R> class(init_results)
#[1] "list"
The second element - items - is the actual result set from the first batch: it's a list of length 100, since we specified maxResults = 100 in the GET request. The first element - nextPageToken - is what we use to make sure each request returns the appropriate sequence of results. For example, we can get the next 100 results like this:
api_opts$pageToken <- gsub("\\=","",init_results$nextPageToken)
next_results <- httr::content(
httr::GET(base_url, query = api_opts))
##
R> next_results$nextPageToken
#[1] "ChYQ-YjT3bmSxQIYyN2VkJG8xAIgACgCEhQIABDI3ZWQkbzEAhiSsMv-ivu0AhgCIMgB"
where the current request's pageToken is returned as the previous requests nextPageToken, and we are given a new nextPageToken for obtaining out next batch of results.
This is pretty straightforward, but it would obviously be very tedious to have to keep changing the value of nextPageToken by hand after each request we send. Instead I thought this would be a good use case for a simple R6 class:
yt_scraper <- setRefClass(
"yt_scraper",
fields = list(
base_url = "character",
api_opts = "list",
nextPageToken = "character",
data = "list",
unique_count = "numeric",
done = "logical",
core_df = "data.frame"),
methods = list(
scrape = function() {
opts <- api_opts
if (nextPageToken != "") {
opts$pageToken <- nextPageToken
}
res <- httr::content(
httr::GET(base_url, query = opts))
nextPageToken <<- gsub("\\=","",res$nextPageToken)
data <<- c(data, res$items)
unique_count <<- length(unique(data))
},
scrape_all = function() {
while (TRUE) {
old_count <- unique_count
scrape()
if (unique_count == old_count) {
done <<- TRUE
nextPageToken <<- ""
data <<- unique(data)
break
}
}
},
initialize = function() {
base_url <<- "https://www.googleapis.com/youtube/v3/commentThreads/"
api_opts <<- list(
part = "snippet",
maxResults = 100,
textFormat = "plainText",
videoId = "4H9pTgQY_mo",
key = "my_google_developer_api_key",
fields = "items,nextPageToken",
orderBy = "published")
nextPageToken <<- ""
data <<- list()
unique_count <<- 0
done <<- FALSE
core_df <<- data.frame()
},
reset = function() {
data <<- list()
nextPageToken <<- ""
unique_count <<- 0
done <<- FALSE
core_df <<- data.frame()
},
cache_core_data = function() {
if (nrow(core_df) < unique_count) {
sub_data <- lapply(data, function(x) {
data.frame(
Comment = x$snippet$topLevelComment$snippet$textDisplay,
User = x$snippet$topLevelComment$snippet$authorDisplayName,
ReplyCount = x$snippet$totalReplyCount,
LikeCount = x$snippet$topLevelComment$snippet$likeCount,
PublishTime = x$snippet$topLevelComment$snippet$publishedAt,
CommentId = x$snippet$topLevelComment$id,
stringsAsFactors=FALSE)
})
core_df <<- do.call("rbind", sub_data)
} else {
message("\n`core_df` is already up to date.\n")
}
}
)
)
which can be used like this:
rObj <- yt_scraper()
##
R> rObj$data
#list()
R> rObj$unique_count
#[1] 0
##
rObj$scrape_all()
##
R> rObj$unique_count
#[1] 1673
R> length(rObj$data)
#[1] 1673
R> ##
R> head(rObj$core_df)
Comment User ReplyCount LikeCount PublishTime
1 That Andorra player was really Ruud..<U+feff> Cistrolat 0 6 2015-03-22T14:07:31.213Z
2 This just in; Karma is a bitch.<U+feff> Swagdalf The Obey 0 1 2015-03-21T20:00:26.044Z
3 Legend! Haha B)<U+feff> martyn baltussen 0 1 2015-01-26T15:33:00.311Z
4 When did Van der sar ran up? He must have run real fast!<U+feff> Witsakorn Poomjan 0 0 2015-01-04T03:33:36.157Z
5 <U+003c>b<U+003e>LOL<U+003c>/b<U+003e> F Hanif 5 19 2014-12-30T13:46:44.028Z
6 Fucking Legend.<U+feff> Heisenberg 0 12 2014-12-27T11:59:39.845Z
CommentId
1 z123ybioxyqojdgka231tn5zbl20tdcvn
2 z13hilaiftvus1cc1233trvrwzfjg1enm
3 z13fidjhbsvih5hok04cfrkrnla2htjpxfk
4 z12js3zpvm2hipgtf23oytbxqkyhcro12
5 z12egtfq5ojifdapz04ceffqfrregdnrrbk
6 z12fth0gemnwdtlnj22zg3vymlrogthwd04
As I alluded to earlier, this gets you almost everything - 1673 out of about 1790 total comments. For some reason, it does not seem to catch users' nested replies, and I'm not quite sure how to specify this within the API framework.
I had previously set up a Google Developer account a while back for using the Google Analytics API, but if you haven't done that yet, it should be pretty straightforward. Here's an overview - you shouldn't need to set up OAuth or anything like that, just make a project and create a new Public API access key.
