class: center, middle, inverse, title-slide # Automatic Sampling and Analysis of YouTube Data ## Excursus: Retrieving Video Subtitles ### Julian Kohne<br />Johannes Breuer<br /><b><i>M. Rohangis Mohseni</b></i> ### 2022-02-22 --- layout: true <div class="my-footer"> <div style="float: left;"><span>Julian Kohne, Johannes Breuer, <b><i>M. Rohangis Mohseni</b></i></span></div> <div style="float: right;"><span>GESIS, online, 2022-02-22</span></div> <div style="text-align: center;"><span>Excursus: Retrieving Video Subtitles</span></div> </div> <style type="text/css"> pre { font-size: 10px } </style> --- ## Retrieving *YouTube* Video Subtitles - Instead of transcribing a video, you can retrieve its subtitles via the *YouTube* API - What research would you conduct with video subtitles? --- ## Types of *YouTube* Subtitles - Videos with automatically created subtitles (*ASR*) - Always in English, even if video language is not English - Can be downloaded, but text quality can be bad (especially if translated) -- - Videos without any subtitles - Not sure if even possible because there always seems to be an *ASR* -- - Videos with more than one set of subtitles - Examples: *ASR* and regular subtitle, more than one language, more than one subtitle for the same language - Can be downloaded, but subtitle for analysis must be selected --- ## Disclaimer Due to a change to the *YouTube* API, the `tuber` function for retrieving video subtitles only work for videos that were created with the same account as the app used for the API access (see this [closed `tuber` issue on GitHub](https://github.com/soodoku/tuber/issues/78)). We will still discuss this function, but recommend that you use the [`youtubecaption` package](https://github.com/jooyoungseo/youtubecaption) for collecting subtitles for videos that you have not created yourself. --- ## Retrieving Video Subtitles with `tuber` First, we need to get the list of subtitles for a video. ```r library(tuber) caption_list <- list_caption_tracks(video_id = "nI_OfkQOG6Q") ``` *Note*: The `tuber` function `list_caption_tracks()` has an API quota cost ~ 50. --- ## Retrieving Video Subtitles with `tuber` Next, we need to get the ID of the subtitles we want to collect. ```r ID <- caption_list[1,"id"] ``` *Note*: You can adapt the number to select the subtitle that you want. The 1 in this example corresponds to the *ASR* captions (ASR = automatic sub). --- ## Retrieving Video Subtitles with `tuber` After that, we need to retrieve the subtitles and convert them from raw to char. ```r text <- rawToChar(get_captions(id = ID, format = "sbv")) ``` Now we can save the subtitles to a subtitle file. ```r write(text, file = "Captions.sbv", sep="\n") ``` --- ## Converting Subtitles - Subtitles come in a special format called SBV - The format contains time stamps etc. that we do not need for text analysis - We can read the format with the package [`subtools`](https://github.com/fkeck/subtools) --- ## Converting Subtitles ```r remotes::install_github("fkeck/subtools") library(subtools) subs <- read_subtitles("Captions.sbv", format = "subviewer") ``` With `subtools`, we can also retrieve the text from the subtitles. ```r subtext <- get_raw_text(subs) ``` Now the text is ready for further analysis (see the previous sessions for examples). --- ## Retrieving Video Subtitles with `youtubecaption` - Alternatively, you can retrieve captions with the package [`youtubecaption`](https://github.com/jooyoungseo/youtubecaption) - **Pros**: - No credentials necessary, therefore no quota reduction (package uses web scraping) - Subtitles are automatically converted into a dataframe, including texts and timestamps, so no manual conversion is needed - **Cons**: - If there is more than one subtitle version per language, there is no way to select a specific one in `R` (by default, the manually generated ones are selected; if you want the ASR subtitles instead, you would need to adapt the [underlying `Python` script](https://pypi.org/project/youtube-transcript-api/)) - You need to install [*Anaconda*](https://www.anaconda.com/products/individual) --- ## Demo: Collecting & processing subtitles with `tuber` & `subtools` You can find the code for collecting subtitles for *YouTube* videos in the `YouTubeSubtitles.R` file in the `scripts` folder. <img src="./Images/youtubesubtitles.jpg" width="75%" style="display: block; margin: auto;" /> --- ## Our Recommendation for Collecting Subtitels If you want to collect subtitles for videos that have been created/uploaded with an account that you have or can easily get access to (e.g., from your institution or a project partner), you can use the functions from the `tuber` package as shown in the demo script. If you want to collect subtitles from other videos, we suggest that you use the `list_caption_tracks()` function from the `tuber` package to check which subtitles are available and the `get_caption` function from the `youtubecaption` package for collecting them (given that you can properly set up *Anaconda* for using this package). --- class: center, middle # Any (further) questions?