Posted onAug 18, 2022 • Edited onFeb 6, 2023 • Originally published atserpapi.com

Scrape YouTube video page with Python

#webscraping #tutorial #python #programming

What will be scraped

Full Code

If you don't need explanation, have a look atfull code example in the online IDE.

importre,json,timefromseleniumimportwebdriverfromselenium.webdriver.chrome.serviceimportServicefromwebdriver_manager.chromeimportChromeDriverManagerfromparselimportSelectordefscroll_page(url):service=Service(ChromeDriverManager().install())options=webdriver.ChromeOptions()options.add_argument("--headless")options.add_argument('--lang=en')options.add_argument("user-agent=Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/104.0.0.0 Safari/537.36")driver=webdriver.Chrome(service=service,options=options)driver.get(url)old_height=driver.execute_script("""        function getHeight() {            return document.querySelector('ytd-app').scrollHeight;        }        return getHeight();    """)whileTrue:driver.execute_script("window.scrollTo(0, document.querySelector('ytd-app').scrollHeight)")time.sleep(2)new_height=driver.execute_script("""            function getHeight() {                return document.querySelector('ytd-app').scrollHeight;            }            return getHeight();        """)ifnew_height==old_height:breakold_height=new_heightselector=Selector(driver.page_source)driver.quit()returnselectordefscrape_all_data(selector):youtube_video_page=[]all_script_tags=selector.css('script').getall()title=selector.css(".title .ytd-video-primary-info-renderer::text").get()# https://regex101.com/r/gHeLwZ/1views=int(re.search(r"(.*)\s",selector.css(".view-count::text").get()).group().replace(",",""))# https://regex101.com/r/9OGwJp/1likes=int(re.search(r"(.*)\s",selector.css("#top-level-buttons-computed > ytd-toggle-button-renderer:first-child #text::attr(aria-label)").get()).group().replace(",",""))date=selector.css("#info-strings yt-formatted-string::text").get()duration=selector.css(".ytp-time-duration::text").get()# https://regex101.com/r/0JNma3/1keywords="".join(re.findall(r'"keywords":\[(.*)\],"channelId":".*"',str(all_script_tags))).replace('\"','').split(",")# https://regex101.com/r/9VhH1s/1thumbnail=re.findall(r'\[{"url":"(\S+)","width":\d*,"height":\d*},',str(all_script_tags))[0].split('",')[0]channel={# https://regex101.com/r/xFUzq5/1"id":"".join(re.findall(r'"channelId":"(.*)","isOwnerViewing"',str(all_script_tags))),"name":selector.css("#channel-name a::text").get(),"link":f'https://www.youtube.com{selector.css("#channel-name a::attr(href)").get()}',"subscribers":selector.css("#owner-sub-count::text").get(),"thumbnail":selector.css("#img::attr(src)").get(),}description=selector.css(".ytd-expandable-video-description-body-renderer span:nth-child(1)::text").get()hash_tags=[{"name":hash_tag.css("::text").get(),"link":f'https://www.youtube.com{hash_tag.css("::attr(href)").get()}'}forhash_taginselector.css(".ytd-expandable-video-description-body-renderer a")]# https://regex101.com/r/onRk9j/1category="".join(re.findall(r'"category":"(.*)","publishDate"',str(all_script_tags)))comments_amount=int(selector.css("#count .count-text span:nth-child(1)::text").get().replace(",",""))comments=[]forcommentinselector.css("#contents > ytd-comment-thread-renderer"):comments.append({"author":comment.css("#author-text span::text").get().strip(),"link":f'https://www.youtube.com{comment.css("#author-text::attr(href)").get()}',"date":comment.css(".published-time-text a::text").get(),"likes":comment.css("#vote-count-middle::text").get().strip(),"comment":comment.css("#content-text::text").get(),"avatar":comment.css("#author-thumbnail #img::attr(src)").get(),})suggested_videos=[]forvideoinselector.css("ytd-compact-video-renderer"):suggested_videos.append({"title":video.css("#video-title::text").get().strip(),"link":f'https://www.youtube.com{video.css("#thumbnail::attr(href)").get()}',"channel_name":video.css("#channel-name #text::text").get(),"date":video.css("#metadata-line span:nth-child(2)::text").get(),"views":video.css("#metadata-line span:nth-child(1)::text").get(),"duration":video.css("#overlays #text::text").get().strip(),"thumbnail":video.css("#thumbnail img::attr(src)").get(),})youtube_video_page.append({"title":title,"views":views,"likes":likes,"date":date,"duration":duration,"channel":channel,"keywords":keywords,"thumbnail":thumbnail,"description":description,"hash_tags":hash_tags,"category":category,"suggested_videos":suggested_videos,"comments_amount":comments_amount,"comments":comments,})print(json.dumps(youtube_video_page,indent=2,ensure_ascii=False))defmain():url="https://www.youtube.com/watch?v=fbh3OAw3VVQ"result=scroll_page(url)scrape_all_data(result)if__name__=="__main__":main()

Preparation

Install libraries:

pip install parsel selenium webdriver webdriver_manager

Reduce the chance of being blocked

Make sure you're usingrequest headers user-agent to act as a "real" user visit. Because defaultrequestsuser-agent ispython-requests and websites understand that it's most likely a script that sends a request.Check what's youruser-agent.

There's ahow to reduce the chance of being blocked while web scraping blog post that can get you familiar with basic and more advanced approaches.

Code Explanation

Import libraries:

importre,json,timefromseleniumimportwebdriverfromselenium.webdriver.chrome.serviceimportServicefromwebdriver_manager.chromeimportChromeDriverManagerfromparselimportSelector

Library	Purpose
`re`	to extract parts of the data via regular expression.
`json`	to convert extracted data to a JSON object.
`time`	to work with time in Python.
`webdriver`	to drive a browser natively, as a user would, either locally or on a remote machine using the Selenium server.
`Service`	to manage the starting and stopping of the ChromeDriver.
`Selector`	XML/HTML parser that have fullXPath and CSS selectors support.

Top-level code environment

This code uses the generally accepted rule of using the__name__ == "__main__" construct:

defmain():url="https://www.youtube.com/watch?v=fbh3OAw3VVQ"result=scroll_page(url)scrape_all_data(result)if__name__=="__main__":main()

This check will only be performed if the user has run this file. If the user imports this file into another, then the check will not work.

You can watch the videoPython Tutorial: ifname == 'main' for more details.

Scroll page

The function takes a YouTube video page URL and returns a full HTML structure.

First, let's understand how YouTube works. YouTube is a dynamic site. The YouTube video page stores a lot of the data you saw in thewhat will be scraped section. Data such ascomments andsuggested videos does not load immediately. If the user needs more data, they will simply scroll the page and YouTube will download a small package of data. Accordingly, to get all the data, you need to scroll to the end of the page.

In this case,selenium library is used, which allows you to simulate user actions in the browser. Forselenium to work, you need to useChromeDriver, which can be downloaded manually or using code. In our case, the second method is used. To control the start and stop ofChromeDriver, you need to useService which will install browser binaries under the hood:

service=Service(ChromeDriverManager().install())

You should also addoptions to work correctly:

options=webdriver.ChromeOptions()options.add_argument("--headless")options.add_argument('--lang=en')options.add_argument("user-agent=Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/104.0.0.0 Safari/537.36")

Chrome options	Explanation
`--headless`	to run Chrome in headless mode.
`--lang=en`	to set the browser language to English.
`user-agent`	to act as a "real" user request from the browser by passing it torequest headers.Check what's your`user-agent`.

Now we can startwebdriver and pass the url to theget() method.

driver=webdriver.Chrome(service=service,options=options)driver.get(url)

The page scrolling algorithm looks like this:

Find out the initial page height and write the result to theold_height variable.
Scroll the page using the script and wait 2 seconds for the data to load.
Find out the new page height and write the result to thenew_height variable.
If the variablesnew_height andold_height are equal, then we complete the algorithm, otherwise we write the value of the variablenew_height to the variableold_height and return to step 2.

Getting the page height and scroll is done by pasting the JavaScript code into theexecute_script() method.

old_height=driver.execute_script("""    function getHeight() {        return document.querySelector('ytd-app').scrollHeight;    }    return getHeight();""")whileTrue:driver.execute_script("window.scrollTo(0, document.querySelector('ytd-app').scrollHeight)")time.sleep(2)new_height=driver.execute_script("""        function getHeight() {            return document.querySelector('ytd-app').scrollHeight;        }        return getHeight();    """)ifnew_height==old_height:breakold_height=new_height

Let's use theSelector fromParsel library, in which we pass thehtml structure with all the data that was received after scrolling the page. This is necessary in order to successfully retrieve data in the next function. After all the operations done, stop the driver:

selector=Selector(driver.page_source)# extracting code from HTMLdriver.quit()

The function looks like this:

defscroll_page(url):service=Service(ChromeDriverManager().install())options=webdriver.ChromeOptions()options.add_argument("--headless")options.add_argument('--lang=en')options.add_argument("user-agent=Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/104.0.0.0 Safari/537.36")driver=webdriver.Chrome(service=service,options=options)driver.get(url)old_height=driver.execute_script("""        function getHeight() {            return document.querySelector('ytd-app').scrollHeight;        }        return getHeight();    """)whileTrue:driver.execute_script("window.scrollTo(0, document.querySelector('ytd-app').scrollHeight)")time.sleep(2)new_height=driver.execute_script("""            function getHeight() {                return document.querySelector('ytd-app').scrollHeight;            }            return getHeight();        """)ifnew_height==old_height:breakold_height=new_heightselector=Selector(driver.page_source)driver.quit()returnselector

In the gif below, I demonstrate how this function works:

Scrape all data

This function takes a full HTML structure and prints all results in json format.

It should be noted that there is data that is not displayed on the YouTube video page. To extract them, you need to useregular expression and search for the<script> tags:

all_script_tags=selector.css('script').getall()# https://regex101.com/r/0JNma3/1keywords="".join(re.findall(r'"keywords":\[(.*)\],"channelId":".*"',str(all_script_tags))).replace('\"','').split(",")# https://regex101.com/r/9VhH1s/1thumbnail=re.findall(r'\[{"url":"(\S+)","width":\d*,"height":\d*},',str(all_script_tags))[0].split('",')[0]channel={# https://regex101.com/r/xFUzq5/1"id":"".join(re.findall(r'"channelId":"(.*)","isOwnerViewing"',str(all_script_tags))),}# https://regex101.com/r/onRk9j/1category="".join(re.findall(r'"category":"(.*)","publishDate"',str(all_script_tags)))

Code	Explanation
`css()`	to access elements by the passed selector.
`getall()`	to actually extract text data from all matching objects.
`"".join()`	to concatenate a list into a string.
`replace()`	to replace all occurrences of the old substring with the new one without extra elements.
`split()`	to return a list of words in a string, separating the string with a delimiter string.
`findall()`	to find for a pattern in a string and return the list of all match objects.

In this way, you can also extract visible data on the page. But the best way is to extract the data from the received HTML structure from thescroll_page function.

The complete function to scrape all data would look like this:

defscrape_all_data(selector):youtube_video_page=[]all_script_tags=selector.css('script').getall()title=selector.css(".title .ytd-video-primary-info-renderer::text").get()# https://regex101.com/r/gHeLwZ/1views=int(re.search(r"(.*)\s",selector.css(".view-count::text").get()).group().replace(",",""))# https://regex101.com/r/9OGwJp/1likes=int(re.search(r"(.*)\s",selector.css("#top-level-buttons-computed > ytd-toggle-button-renderer:first-child #text::attr(aria-label)").get()).group().replace(",",""))date=selector.css("#info-strings yt-formatted-string::text").get()duration=selector.css(".ytp-time-duration::text").get()# https://regex101.com/r/0JNma3/1keywords="".join(re.findall(r'"keywords":\[(.*)\],"channelId":".*"',str(all_script_tags))).replace('\"','').split(",")# https://regex101.com/r/9VhH1s/1thumbnail=re.findall(r'\[{"url":"(\S+)","width":\d*,"height":\d*},',str(all_script_tags))[0].split('",')[0]channel={# https://regex101.com/r/xFUzq5/1"id":"".join(re.findall(r'"channelId":"(.*)","isOwnerViewing"',str(all_script_tags))),"name":selector.css("#channel-name a::text").get(),"link":f'https://www.youtube.com{selector.css("#channel-name a::attr(href)").get()}',"subscribers":selector.css("#owner-sub-count::text").get(),"thumbnail":selector.css("#img::attr(src)").get(),}description=selector.css(".ytd-expandable-video-description-body-renderer span:nth-child(1)::text").get()hashtags=[{"name":hash_tag.css("::text").get(),"link":f'https://www.youtube.com{hash_tag.css("::attr(href)").get()}'}forhash_taginselector.css(".ytd-expandable-video-description-body-renderer a")ifhash_tag.css("::text").get()[0]=='#']# https://regex101.com/r/onRk9j/1category="".join(re.findall(r'"category":"(.*)","publishDate"',str(all_script_tags)))comments_amount=int(selector.css("#count .count-text span:nth-child(1)::text").get().replace(",",""))comments=[]forcommentinselector.css("#contents > ytd-comment-thread-renderer"):comments.append({"author":comment.css("#author-text span::text").get().strip(),"link":f'https://www.youtube.com{comment.css("#author-text::attr(href)").get()}',"date":comment.css(".published-time-text a::text").get(),"likes":comment.css("#vote-count-middle::text").get().strip(),"comment":comment.css("#content-text::text").get(),"avatar":comment.css("#author-thumbnail #img::attr(src)").get(),})suggested_videos=[]forvideoinselector.css("ytd-compact-video-renderer"):suggested_videos.append({"title":video.css("#video-title::text").get().strip(),"link":f'https://www.youtube.com{video.css("#thumbnail::attr(href)").get()}',"channel_name":video.css("#channel-name #text::text").get(),"date":video.css("#metadata-line span:nth-child(2)::text").get(),"views":video.css("#metadata-line span:nth-child(1)::text").get(),"duration":video.css("#overlays #text::text").get().strip(),"thumbnail":video.css("#thumbnail img::attr(src)").get(),})youtube_video_page.append({"title":title,"views":views,"likes":likes,"date":date,"duration":duration,"channel":channel,"keywords":keywords,"thumbnail":thumbnail,"description":description,"hashtags":hashtags,"category":category,"suggested_videos":suggested_videos,"comments_amount":comments_amount,"comments":comments,})print(json.dumps(youtube_video_page,indent=2,ensure_ascii=False))

Code	Explanation
`youtube_video_page`	a temporary`list` where extracted data will beappended at the end of the function.
`css()`	to access elements by the passed selector.
`::text` or`::attr(<attribute>)`	to extract textual or attribute data from the node.
`get()`	to actually extract the textual data.
`search()`	to search for a pattern in a string and return the corresponding match object.
`group()`	to extract the found element from the match object.
`youtube_video_page.append({})`	to`append` extracted data to a`list` as a dictionary.
`strip()`	to return a copy of the string with the leading and trailing characters removed.

Output

[{"title":"Hogwarts Legacy 8 Minutes Exclusive Gameplay (Unreal Engine 4K 60FPS HDR)","views":215244,"likes":2632,"date":"Aug 5, 2022","duration":"8:20","channel":{"id":"UCAvRtPIFFbnfLHswfRS6zhQ","name":"GameV","link":"https://www.youtube.com/c/GameV","subscribers":"55.5K subscribers","thumbnail":"https://yt3.ggpht.com/ytc/AMLnZu-8W0taSvZWdVQXoqge6XSLmhk39jns9e7S-zjLQA=s48-c-k-c0x00ffffff-no-rj"},"keywords":["hogwarts legacy","hogwarts legacy gameplay","hogwarts legacy gameplay trailer",...otherkeywords],"thumbnail":"https://i.ytimg.com/vi/fbh3OAw3VVQ/hqdefault.jpg?sqp=-oaymwEbCKgBEF5IVfKriqkDDggBFQAAiEIYAXABwAEG\\\\u0026rs=AOn4CLCucgyCMVC9CvMdz2vULM4FQZI3GQ","description":"Hogwarts Legacy 8 Minutes Exclusive Gameplay (Unreal Engine 4K 60FPS HDR)\n\nHogwarts Legacy is an upcoming action role-playing video game developed by Avalanche Software and published by Warner Bros. Interactive Entertainment under its Portkey Games label, using Unreal Engine. It is set in the Wizarding World universe, based on the Harry Potter novel and film series and Fantastic Beasts. The game is set to be released in late 2022 for Microsoft Windows, Nintendo Switch, PlayStation 4, PlayStation 5, Xbox One, and Xbox Series X/S\n\n","hashtags":[{"name":"#hogwartslegacy","link":"https://www.youtube.com/hashtag/hogwartslegacy"},...otherhashtags],"category":"Gaming","suggested_videos":[{"title":"The Story of Hogwarts Legacy Is Crazy..","link":"https://www.youtube.com/watch?v=_WW2VauR968","channel_name":"WolfheartFPS","date":"2 days ago","views":"6K views","duration":"16:44","thumbnail":"https://i.ytimg.com/vi/_WW2VauR968/hqdefault.jpg?sqp=-oaymwEbCKgBEF5IVfKriqkDDggBFQAAiEIYAXABwAEG&rs=AOn4CLDTi3OShNy4DTW21bC6IQfAPc-DCw"},...othervideos],"comments_amount":359,"comments":[{"author":"Aaron","link":"https://www.youtube.com/channel/UCCsQ_qXy5oM0pbGI1Az0aGA","date":"1 day ago (edited)","likes":"101","comment":"Particle effects on contact look crazy. There's clearly a lot of detail in this game. I just hope the gameplay isn't stale. One thing I noticed is the enemies seem kind of dumb just standing around during combat. Maybe there is a difficulty setting that will change that.","avatar":"https://yt3.ggpht.com/ytc/AMLnZu8phDhzI_6ncZ7bh7sWBcti3np2xgdrTSNqqoBw_w=s48-c-k-c0x00ffffff-no-rj"},...othercomments]}]