google_takeout_parser

google_takeout_parser

Parses data out of your Google Takeout (History, Activity, Youtube, Locations, etc…). This:

• parses both the Historical HTML and new JSON format for Google Takeouts
• caches individual takeout results behind cachew
• merges multiple takeouts into unique events

---

• Installation
• Usage
  • CLI Usage
  • Library Usage
• Legacy HTML Parsing
• Contributing
• Testing

This doesn’t handle all cases, but I have yet to find a parser that does, so here is my attempt at parsing what I see as the most useful info from it. The Google Takeout is pretty particular, and the contents of the directory depend on what you select while exporting. Unhandled files will warn, though feel free to PR a parser or create an issue if this doesn’t parse some part you want.

The google takeout output changes all the time. Files change names, products are added/removed, fields on the data change, the types or accuracy of data changes. I have been using and updating this parser since 2020, and other users have successfully parsed data back from 2018, but the data in your account may hit an edge case that we never happened to. If you do hit an error, feel free to create an issue.

This can take a few minutes to parse depending on what you have in your Takeout (especially while using the old HTML format), so this uses cachew to cache the function result for each Takeout you may have. That means this’ll take a few minutes the first time parsing a takeout, but then only a few seconds every subsequent time.

Since google slowly removes your old data over time, I would recommend periodically backing up your data (personally I do it once every few months), to not lose any old events and get access to data from new ones. To do that, go to takeout.google.com; For Reference, once on that page, I hit Deselect All, then select:

• Chrome
• Google Play Store
• Keep
• My Activity
  • Select JSON as format
• Youtube and Youtube Music
  • Select JSON as format
  • In options, deselect music (library and uploads) and videos

Be sure to select JSON instead of HTML whenever possible. Code to parse the HTML format is included here, but it is treated as legacy code and comes with worse performance and a myriad of other issues. See legacy html parsing

The process for obtaining these isn’t that great – you have to manually go to takeout.google.com every few months, select what you want to export info for, and then it puts the zipped file into your google drive. You can tell it to run it at specific intervals, but I personally haven’t found that to be that reliable.

This currently parses:

• Activity (from dozens of Google Services) - My Activity/*.html|*.json)
• Chrome History - Chrome/BrowserHistory.json
• Google Play Installs - Google Play Store/Installs.json
• Keep (Notes) - Keep/*.json
• Location History (though, this may not available through any new takeouts anymore as its stored on device):
  • Semantic Location HistoryLocation History/Semantic Location History/*
  • Location History Location History/Location History.json, Location History/Records.json
• Youtube:
  • History - YouTube and YouTube Music/history/*.html|*.json
  • Comments:
    • Legacy HTML Comment/Live chats: YouTube and YouTube Music/my-comments/*.html and YouTube and YouTube Music/my-live-chat-messages/*.html
    • CSV/JSON (comment text is stored as a JSON blob, see below
      • Youtube/comments/comments.csv
      • Youtube/live chats/live chats.csv
  • Likes: YouTube and YouTube Music/playlists/likes.json

This was extracted out of my HPI modules, which was in turn modified from the google files in karlicoss/HPI

Installation

Requires python3.10+

To install with pip, run:

pip install google-takeout-parser

Usage

The directory structure of the google takeout changes depending on your Google accounts main language. If this doesn’t support your language, see contributing. This currently supports:

• EN: English
• DE: German (thanks to @parthux1)

CLI Usage

Can be accessed by either google_takeout_parser or python -m google_takeout_parser. Offers a basic interface to list/clear the cache directory, and/or parse/merge a takeout and interact with it in a REPL:

Usage: google_takeout_parser parse [OPTIONS] TAKEOUT_DIR

  Parse a takeout directory takeout

Options:
  -f, --filter [Activity|LikedYoutubeVideo|PlayStoreAppInstall|Location|ChromeHistory|YoutubeComment|PlaceVisit]
                                  Filter to only show events of this type
  -l, --locale [EN|DE]            Locale to use for matching filenames [default: EN]  [env var:
                                  GOOGLE_TAKEOUT_PARSER_LOCALE]
  -a, --action [repl|summary|json]
                                  What to do with the parsed result  [default: repl]
  --cache / --no-cache            [default: no-cache]
  -h, --help                      Show this message and exit.

If you use a language this doesn’t support, see contributing.

To clear the cachew cache: google_takeout_parser cache_dir clear

A few examples of parsing takeouts:

$ google_takeout_parser --quiet parse ~/data/Unpacked_Takout --cache
Interact with the export using res

In [1]: res[-2]
Out[1]: PlayStoreAppInstall(title='Hangouts', device_name='motorola moto g(7) play', dt=datetime.datetime(2020, 8, 2, 15, 51, 50, 180000, tzinfo=datetime.timezone.utc))

In [2]: len(res)
Out[2]: 236654

$ google_takeout_parser --quiet merge ./Takeout-Old ./Takeout-New --action summary --no-cache

Counter(
    {
        "Activity": 366292,
        "Location": 147581,
        "YoutubeComment": 131,
        "PlayStoreAppInstall": 122,
        "LikedYoutubeVideo": 100,
        "ChromeHistory": 4,
    }
)

Can also dump the info to JSON; e.g. to filter YouTube-related stuff from your Activity using jq:

google_takeout_parser --quiet parse -a json -f Activity --no-cache ./Takeout-New |
	# select stuff like Youtube, m.youtube.com, youtube.com using jq
	jq '.[] | select(.header | ascii_downcase | test("youtube"))' |
	# grab the titleUrl, ignoring nulls
	jq 'select(.titleUrl) | .titleUrl' -r

Also contains a small utility command to help move/extract the google takeout:

$ google_takeout_parser move --from ~/Downloads/takeout*.zip --to-dir ~/data/google_takeout --extract
Extracting /home/username/Downloads/takeout-20211023T070558Z-001.zip to /tmp/tmp07ua_0id
Moving /tmp/tmp07ua_0id/Takeout to /home/username/data/google_takeout/Takeout-1634993897
$ ls -1 ~/data/google_takeout/Takeout-1634993897
archive_browser.html
Chrome
'Google Play Store'
'Location History'
'My Activity'
'YouTube and YouTube Music'

Library Usage

Assuming you maintain an unpacked view, e.g. like:

 $ tree -L 1 ./Takeout-1599315526
./Takeout-1599315526
├── Google Play Store
├── Location History
├── My Activity
└── YouTube and YouTube Music

To parse one takeout:

from google_takeout_parser.path_dispatch import TakeoutParser

tp = TakeoutParser("/full/path/to/Takeout-1599315526")
# to check if files are all handled
tp.dispatch_map()
# to parse without caching the results in ~/.cache/google_takeout_parser
uncached = list(tp.parse())
# to parse with cachew cache https://github.com/karlicoss/cachew
cached = list(tp.parse(cache=True))

To parse a locale this doesn’t support yet, you can create a dictionary which maps the names of the files to functions, see locales/en.py for an example. That can be passed as handlers to TakeoutParser

To cache and merge takeouts (maintains a single dependency on the paths you pass – so if you change the input paths, it does a full recompute)

from google_takeout_parser.merge import cached_merge_takeouts

results = list(
    cached_merge_takeouts(
        ["/full/path/to/Takeout-1599315526", "/full/path/to/Takeout-1634971143"]
    )
)

If you don’t want to cache the results but want to merge results from multiple takeouts, can do something custom by directly using the merge_events function:

from google_takeout_parser.merge import merge_events, TakeoutParser

itrs = []  # list of iterators of google events
for path in ["path/to/Takeout-1599315526" "path/to/Takeout-1616796262"]:
    # ignore errors, error_policy can be 'yield', 'raise' or 'drop'
    tk = TakeoutParser(path, error_policy="drop")
    itrs.append(tk.parse(cache=False))
res = list(merge_events(*itrs))

The events this returns is a combination of all types in the models.py, to filter to a particular type you can provide that to skip parsing other files:

from google_takeout_parser.models import Location
from google_takeout_parser.path_dispatch import TakeoutParser

# filter_type can be a list to filter multiple types
locations = list(TakeoutParser("path/to/Takeout").parse(filter_type=Location))
len(locations)
99913

I personally exclusively use this through the HPI google takeout file, as a configuration layer to locate where my takeouts are on disk, and since that ‘automatically’ unzips the takeouts (I store them as the zips), i.e., doesn’t require me to maintain an unpacked view

Youtube Comment JSON

The CSV youtube comment files’ content are stored as a JSON blob, which look like:

{
  "takeoutSegments": [
    {
      "text": "I symlink /bin/sh to dash, its a very minimal/posix compliant implementation. see "
    },
    {
      "text": "https://wiki.archlinux.org/index.php/Dash",
      "link": {
        "linkUrl": "https://wiki.archlinux.org/index.php/Dash"
      }
    }
  ]
}

This exposes some functions to help parse those, into text, markdown, or just extract the links:

from google_takeout_parser.path_dispatch import TakeoutParser
from google_takeout_parser.models import CSVYoutubeComment
from google_takeout_parser.parse_csv import (
    extract_comment_links,
    reconstruct_comment_content,
)


path = "./Takeout-1599315526"
comments = list(
    TakeoutParser(path, error_policy="raise").parse(
        cache=False, filter_type=CSVYoutubeComment
    )
)
for e in comments:
    print(extract_comment_links(e.contentJSON))
    print(reconstruct_comment_content(e.contentJSON, "text"))
    print(reconstruct_comment_content(e.contentJSON, "markdown"))

Legacy HTML Parsing

I would heavily recommend against using the HTML format for My Activity. It is not always possible to properly parse the metadata, is more prone to errors parsing dates due to local timezones, and takes much longer to parse than the JSON format.

On certain machines, the giant HTML files may even take so much memory that the process is eventually killed for using too much memory. For a workaround, see split_html.

Contributing

Just to give a brief overview, to add new functionality (parsing some new folder that this doesn’t currently support), you’d need to:

• Add a model for it in models.py subclassing BaseEvent and adding it to the Union at the bottom of the file. That should have a key property function which describes each event uniquely (this is used to remove duplicate items when merging takeouts)
• Write a function which takes the Path to the file you’re trying to parse and converts it to the model you created (See examples in parse_json.py). Ideally extract a single raw item from the takeout file add a test for it so its obvious when/if the format changes.
• Add a regex match for the file path to the handler map in google_takeout_parser/locales/en.py.

Dont feel required to add support for all locales, its somewhat annoying to swap languages on google, request a takeout, wait for it to process and then swap back.

Though, if your takeout is in some language this doesn’t support, you can create an issue with the file structure (run find Takeout and/or tree Takeout), or contribute a locale file by creating a path -> function mapping (see locales), and adding it to the global LOCALES variables in locales/all.py and locales/main.py

This is a pretty difficult to maintain, as it requires a lot of manual testing from people who have access to these takeouts, and who actively use the language that the takeout is in. My google accounts main language is English, so I upkeep that locale whenever I notice changes, but its not trivial to port those changes to other locales without swapping my language, making an export, waiting, and then switching back. I keep track of mismatched changes in this board

Ideally, when first creating a locale file, you would select everything when doing a takeout (not just the My Activity/Chrome/Location History like I suggested above), so paths that are not parsed can be ignored properly.

Testing

git clone 'https://github.com/purarue/google_takeout_parser'
cd ./google_takeout_parser
pip install '.[testing]'
mypy ./google_takeout_parser
flake8 ./google_takeout_parser
pytest