Twitter's API limits you to querying a user's most recent 3200 tweets. This is a pain in the ass. However, we can circumvent this limit using Selenium and doing some webscraping.

We can query a user's entire time on twitter, finding the IDs for each of their tweets. From there, we can use the tweepy API to query the complete metadata associated with each tweet. You can adjust which metadata are collected by changing the variable METADATA_LIST at the top of scrape.py. Personally, I was just collecting text to train a model, so I only cared about the full_text field in addition to whether the tweet was a retweet.

I've included a list of all available tweet attributes at the top of scrape.py so that you can adjust things as you wish.

NOTE: This scraper will notice if a user has less than 3200 tweets. In this case, it will do a "quickscrape" to grab all available tweets at once (significantly faster). It will store them in the exact same manner as a manual scrape.

• python3
• Modules (via pip - see requirements.txt):
  • selenium
  • tweepy
  • requests
  • requests_oauthlib
  • beautifulsoup4
• Chrome webdriver (you can use other drivers. Personally I use, brew install chromedriver because I work on a Mac)
• Twitter API developer credentials

I'll run the script two times on one of my advisors. By default, the scraper will start whenever the user created their twitter. I've chosen to look at a 1 year window, scraping at two week intervals. I then go from the beginning of 2019 until the present day, at 1 week intervals. The scraped tweets are stored in a JSON file that bears the Twitter user's handle.

$ ./scrape.py --help
usage: python3 scrape.py [options]
scrape.py - Twitter Scraping Tool
optional arguments:
 -h, --help show this help message and exit
 -u USERNAME, --username USERNAME
 Scrape this user\'s Tweets
 --since SINCE Get Tweets after this date (Example: 2010年01月01日).
 --until UNTIL Get Tweets before this date (Example: 2018年12月07日).
 --by BY Scrape this many days at a time
 --delay DELAY Time given to load a page before scraping it (seconds)
 --debug Debug mode. Shows Selenium at work + additional logging
$ ./scrape.py -u phillipcompeau --by 14 --since 2018年01月01日 --until 2019年01月01日 
[ scraping user @phillipcompeau... ]
[ 1156 existing tweets in phillipcompeau.json ]
[ searching for tweets... ]
[ found 254 new tweets ]
[ retrieving new tweets (estimated time: 18 seconds)... ]
- batch 1 of 3
- batch 2 of 3
- batch 3 of 3
[ finished scraping ]
[ stored tweets in phillipcompeau.json ]
$ ./scrape.py -u phillipcompeau --since 2019年01月01日 --by 7
[ scraping user @phillipcompeau... ]
[ 1410 existing tweets in phillipcompeau.json ]
[ searching for tweets... ]
[ found 541 new tweets ]
[ retrieving new tweets (estimated time: 36 seconds)... ]
- batch 1 of 6
- batch 2 of 6
- batch 3 of 6
- batch 4 of 6
- batch 5 of 6
- batch 6 of 6
[ finished scraping ]
[ stored tweets in phillipcompeau.json ]
$ ./scrape.py -u realwoofy
[ scraping user @realwoofy... ]
[ 149 existing tweets in realwoofy.json ]
[ searching for tweets... ]
[ user has fewer than 3200 tweets, conducting quickscrape... ]
[ found 3 new tweets ]
[ finished scraping ]
[ stored tweets in realwoofy.json ]

• run python3 scrape.py and add the arguments you desire. Try ./scrape.py --help for all options.
  • -u followed by the username [required]
  • --since followed by a date string, e.g., (2017年01月01日). Defaults to whenever the user created their Twitter
  • --until followed by a date string, e.g., (2018年01月01日). Defaults to the current day
  • --by followed by the number of days to scrape at once (default: 7)
    • If someone tweets dozens of times a day, it might be better to use a lower number
  • --delay followed by an integer. This will be the number of seconds to wait on each page load before reading the page
    • if your internet is slow, put this higher (default: 3 seconds)
  • --debug. This will disable headless mode on the WebDriver so that you can watch it scrape. This is useful for assessing why it's unable to find tweets.
• a browser window will pop up and begin scraping
• when the browser window closes, metadata collection begins for all new tweets
• when collection finishes, it will dump all the data to a .json file that corresponds to the twitter handle
  • don't worry about running two scrapes that have a time overlap; it will only retrieve new tweets!

• do you get a driver error when you try and execute the script?
  • make sure your browser is up to date and that you have a driver version that matches your browser version
  • you can also open scrape.py and change the driver to use Chrome() or Firefox()
• does the scraper seem like it's missing tweets that you know should be there?
  • try increasing the --delay parameter, it likely isn't waiting long enough for everything to load
  • try decreasing the --by parameter, it likely has too many tweets showing up on certain days

• sign up for a developer account here https://dev.twitter.com/
• get your key here: https://apps.twitter.com/
• once you have your key, open the file api_key.example.py and fill in your info
  • once you're done, remove example from the file name