tiktok_research_api_helper 0.1.0rc1

A small package to interface with the TikTok Research API

tiktok_research_api_helper

This package provides both a CLI application and python library for querying video information from the TikTok Research API.

This library requires TikTok Research API access. It does not provide any access by itself.

Requirements

Python3.11+ is required. Some newer features are directly used and earlier versions won't work (e.g. Walrus, type hinting chaining "|", etc., StrEnum)

Python code usage

Create secrets.yaml

You need to put your API credentials in yaml file which the client code will use for authentication. Expected fields (no quotes):

client_id: 123
client_secret: abc
client_key: abc

Using the interface:

Construct an API query

A query is a combination of a "type (and, or, not)" with multiple Conditions ("Cond")

Each condition is a combination of a "field" (Fields, F), "value" and a operation ("Operations", "Op").

from tiktok_research_api_helper.query import VideoQuery, Cond, Fields, Op

query = VideoQuery(
        and_=[
            Cond(Fields.hashtag_name, "garfield", Op.EQ),
            Cond(Fields.region_code, "US", Op.EQ),

            # Alternative version with multiple countries - Then the operation changes to "IN" instead of "EQ" (equals) as it's a list
            # the library handles list vs str natively
            # Cond(Fields.region_code, ["US", "UK"], Op.IN),
        ],
    )

TikTokApiClient provides a high-level interface to fetch all api results, and optionally store them in a database

from pathlib import Path
from datetime import datetime
from tiktok_research_api_helper.query import VideoQuery, Cond, Fields, Op
from tiktok_research_api_helper.api_client import ApiClientConfig, TikTokApiClient, VideoQueryConfig

client_config = ApiClientConfig(engine=None, # No database engine configured, so
                                             # client cannot store results
                                api_credentials_file=Path("./secrets.yaml"))
api_client = TikTokApiClient.from_config(client_config)

# Now setup our query with start and end dates.
query_config = VideoQueryConfig(query=query,
                                start_date=datetime.fromisoformat("2024-03-01"),
                                end_date=datetime.fromisoformat("2024-03-02"))

# api_results_iter yields each API reponse as a parsed TikTokApiClientFetchResult.
# Iteration stops when the API indicates the query results have been fully delivered. or if client_config.max_api_requests is reached.
for result in api_client.api_results_iter(query_config):
    # do something with the result
    print(result.videos)


# Alternatively fetch_all fetches all API results and returns a single TikTokApiClientFetchResult with all API results. NOTE: this blocks until all results are fetched which could be multiple days if query results exceed daily quota limit.
api_client.fetch_all(query_config)


# If you provide a SqlAlchemy engine in the ApiClientConfig you can use TikTokApiClient to store results as they are received
api_client.fetch_and_store_all(query_config) # or equivalent call: fetch_all(query_config, store_results_after_each_response=True)

You can also fetch user info and comments for videos that match the qurey:

query_config = VideoQueryConfig(query=query,
                                start_date=datetime.fromisoformat("2024-03-01"),
                                end_date=datetime.fromisoformat("2024-03-02"),
                                fetch_comments=True,
                                fetch_user_info=True)

# Reusing same client before.
results = api_client.fetch_all(query_config)
print('Videos: ", results.videos)
print('User info: ", results.user_info)
print('Comments: ", results.comments)

TikTokApiRequestClient and TikTokRequest provide a lower-level interface to API

Fetching Videos

from pathlib import Path
from tiktok_research_api_helper.api_client import TikTokApiRequestClient, TikTokVideoRequest

# reads from secrets.yaml in the same directory
request_client = TikTokApiRequestClient.from_credentials_file(Path("./secrets.yaml"))
from tiktok_research_api_helper.query import VideoQuery, Cond, Fields, Op

query = VideoQuery(or_=Cond(Fields.video_id, ["7345557461438385450", "123456"], Op.IN))

# sample query
video_req = TikTokVideoRequest(
    query=query,
    start_date="20240301",
    end_date="20240329",
)

# then fetch the first page of results for the query. NOTE: this does not automatically fetch subsequent pages.
result = request_client.fetch_videos(video_req)

# to request the next page of resuls, you must create a new request with the cursor and search_id values from previous result. NOTE: make sure to check results.data['has_more'] == true
new_video_req = TikTokVideoRequest(query=query,
                cursor=result.data['cursor'],
                search_id=result.data['search_id'],
            )
result = request_client.fetch_videos(new_video_req)

Fetching Comments

from tiktok_research_api_helper.api_client import TikTokCommentsRequest

video_id = "7345557461438385450"
comments_req = TikTokCommentsRequest(video_id=video_id)

result = request_client.fetch_comments(comments_req)
for comment in result.coments:
  print(comment)

Fetching user info

from tiktok_research_api_helper.api_client import TikTokUserInfoRequest

username = "example"
user_info_req = TikTokUserInfoRequest(username=username)

result = request_client.fetch_user_info(user_info_req)
print(username, " user info: ", result.user_info[username])

Basic CLI usage

1. This library requires TikTok Research API access. It does not provide any access by itself.
2. Create a new file secrets.yaml in the root folder you are running code from (you can specify a different file with --api-credentials-file). View the sample_secrets.yaml file for formatting. The client_id, client_secret and client_key are required. The library automatically manages the access token and refreshes it when needed.
3. View the ExampleInterface.ipynb for a quick example of interfacing with it for small queries.

Querying

You can query the API for videos that include and/or exclude hashtags and/or keywords with the following flags:

hashtags

• --include-any-hashtags
• --include-all-hashtags
• --exclude-any-hashtags
• --exclude-all-hashtags

keywords

• --include-any-keywords
• --include-all-keywords
• --exclude-any-keywords
• --exclude-all-keywords

These flags take a comma separated list of values (eg --include-all-hashtags butter,cheese, --only-from-usernames amuro,roux)

flags with any in the name will query the API for videos that have one or more of the provided values. for example --include-any-hashtags butter,cheese would match videos with hashtags #butter, #cheese, and/or #butter #cheese (ie both). The same applies for keyword variants of these flags

flags with all in the name will query the API for videos that have all the provided values, and would not match videos which only a subset of the provided values. for example --include-all-hashtags butter,cheese would match videos with hashtags #butter #cheese, but would not match videos with only #butter but not #cheese and vice versa. The same applies for keyword variants of these flags

usernames

You can also limit results by username. Either querying for videos only from specific usernames or excluding videos from specific usernames. NOTE: these flags are mutually exclusive:

• --only-from-usernames
• --exclude-from-usernames

regions

You can also limit the videos by the region in which the use registered their account with --region (this flag can be provided multiple times include multiple regions). See tiktok API documentation for more info about this field https://developers.tiktok.com/doc/research-api-specs-query-videos/

fetching user info

--fetch-user-info For each video the API returns user info is fetched for the video's creator. (for more about what TikTok research API provides and how it is structured see https://developers.tiktok.com/doc/research-api-specs-query-user-info)

Fetching comments

--fetch-comments For each video the API returns comments (up to the first 1000 due to API limitations) are fetched. (for more about what the API provides for comments and how the responses are structured see https://developers.tiktok.com/doc/research-api-specs-query-video-comments) NOTE: fetching comments can significantly increase API quota consumption beacuse potentially every video will used 10 extra API requests.

Limiting number of API requests (to preserve precious API quota)

by default tool has no limit on API requests. When the API indicates quota has been exceed the tool sleeps and retries until quota resets at UTC midnight. If you wish to limit the number of requests, say to preserve precious little API quota, you can use the --max-api-requests flag which take a positive int. Once that many requests have been made crawling will stop even if the API indicates more results are available.

Print query without sending requests to API

If you would like to preview the query that would be sent to the API (without actually sending a request to the API) you can use the command print-query like so:

$ tiktok-lib print-query --include-all-keywords cheese,butter --exclude-any-hashtags pasta,tomato --region US --region FR --exclude-from-usernames carb_hater,only-vegetables

This prints the JSON query to stdout.

Advanced queries

If the provided querying functionality does not meet your needs or you want to provide your own query use the --query-file-json flag. This takes a path to a JSON file that will be used as the query for requests to the API. NOTE: the provided file is NOT checked for validity. See tiktok documentation for more info about crafting queries https://developers.tiktok.com/doc/research-api-get-started/

You can use the print-query command to create a starting point. For example if you wanted to match videos about shoes with more specific search criteria you could create a base onto which you would build with something like:

$ tiktok-lib print-query --include-any-keywords shoe,shoes,sneakers,pumps,heels,boots > shoes-query.json

then edit shoes-query.json as desired, and use it with

$ tiktok-lib run --query-json-file shoes-query.json ...

Large scale and database usage

1. For larger queries, first install SQLite.
2. Run a test query with tiktok-lib test
3. Edit the query.yaml file to include the query you want to run.
4. View the available commands in the run command with tiktok-lib run --help

You can also store files in a postgresql database. Use the --db-url flag to specify the connection string.

Limitations

• Currently only video data ("Query Videos"), user info, and video comments is supported directly.

Internals

• Long running queries are automatically split into smaller 7 days chunks. This is to avoid the 30 day limit on the TikTok API.

• The library automatically manages the access token and refreshes it when needed.

• TikTok research API quota is 1000 requests per day (https://developers.tiktok.com/doc/research-api-faq). When the API indicates that limit has been reached this library will retry (see --rate-limit-wait-strategy flag for available strategies) until quota limit resets and continue collection.

• TikTokApiClient provides a high-level interface for querying TikTok Research API

  • Handles API pagination (ie requesting results from API until API indicates query results have been completely delivered), access token fetch/refresh, and retry on request failures.
  • Client provides an iterator (api_results_iter) which yields each parsed API response, or fetch_all which returns all parsed results in one object.
  • store_fetch_result stores crawl and videos data to the database
  • fetch_and_store_all does all the above (fetching all results from API and storing them in database as responses are received).
  • fetch_comments and fetch_user_info cache responses (using video ID, and username respectively) to reduce API requests at the cost of additional memory usage. This done via rudimentary dict storage with the video ID or username mapping to the response.

• Database

  • All "Crawls" (really each request to the API) are stored in a seperate table crawl and the data itself in video.
  • Mapping of video <-> crawl is stored in videos_to_crawls.
  • Hashtags are stored in a separate table (with an internal ID, NOT from the API) hashtag, and the mapping of hashtags <-> videos is stored in videos_to_hashtags
  • Effect IDs are stored simimarly to Hashtags with an associtation table videos_to_effect_ids. The naming in effect table can be little confusing because id is the internal database ID, and effect_id is the value from the API (which is a string, but this author has only ever seen ints (as strings) from API).
  • Data is written to DB after every TikTokRequest, by default containing up to 100 instances.
  • If a query tag (via the --query-tag flag) is provided, crawls and videos are associated to the query tag in crawls_to_query_tags and videos_to_query_tags tables respectively.

Roadmpap

• Fix warning when retrying - Only show if the retry is unsuccessful
• Add code docs
• Allow for continuing a query directly from the last run.
• Support for other data types (e.g. "User reposted videos", "User followers", etc)

Why ...

• Having a query as python code inside a file?
  • To make facilitate not having to write extensive json queries in the CLI.
• Not tiktok-research-client?
  • At the time of creation, the library was not available.

Development

Installation

Install with pip:

git clone <this repo>
cd tiktok-library
pip install .

OR you can install hatch (see https://hatch.pypa.io/latest/install/) and run code/tests from that:

git clone <this repo>
cd tiktok-library
hatch --env test run run # run unit tests
hatch run tiktok-lib run --db-url ... # query API

Testing

To run unit tests locally (requires pytest installed): python3 -m pytest OR use hatch hatch run test:run

To run postgresql integration test (requires docker installed, may have to run as sudo):

docker compose build && docker compose run postgres-integration-test && docker compose down

OR run with hatch (this runs above docker commands as sudo):

hatch run test:postgres-integration-test-docker-as-sudo

Automatic formatting and linting with ruff

To check if ruff would change code (but not actually make changes):

hatch fmt --check

To apply changes from ruff:

hatch fmt

NOTE: formatting fixes will not be applied if linter finds errors, if this happens you can run the formatter only with hatch fmt --formatter (good for when there is a linter issue the formatter can fix).

Running jupyter notebook via hatch

hatch run jupyter:notebook

Docker image

ghcr.io/cybersecurityfordemocracy/tiktok_research_api_helper is a very minimal wrapper to run tiktok-lib in a docker container. Currently the image is built by installing the specified version of this package from pypi (for simplicity and transparency).

To build the image you can use the following command:

bash -c 'VERSION=<PIP VERSION SPECIFIER HERE>; docker build -t ghcr.io/cybersecurityfordemocracy/tiktok_research_api_helper:${VERSION} --build-arg VERSION=${VERSION} .'

This will make sure that the image is tagged with the same version specifier as is used to install from pypi.

Alembic database schema migrations

Alembic is a tool/framework for database schema migrations. For instructions on how to use and and create revisions see https://alembic.sqlalchemy.org/en/latest/tutorial.html

There is hatch env for this which can be invoked like so:

$ hatch --env alembic run alembic ...

The alembic config in this repo alembic.ini is a basic config barely modified from generic default. To use it you will need to define sqlalchemy.url. If you want to operate on different database URLs you can use a technique documented in https://alembic.sqlalchemy.org/en/latest/cookbook.html#run-multiple-alembic-environments-from-one-ini-file briefly you would add something like the following to alembic.ini:

[test]
sqlalchemy.url = driver://user:pass@localhost/test_database_name

[prod]
sqlalchemy.url = driver://user:pass@localhost/prod_database_name

Then you can specify which database to use via the config section name:

$ hatch --env alembic run alembic --name test <alembic command>

NOTE: if a database has not had alembic run against it, but nonetheless has up-to-date schema, alembic commands will fail saying it is not on the latest version. You can "stamp" the database as being at HEAD (refering here to alembic versions, NOT git commits) with the following:

$ hatch --env alembic run alembic --name test stamp head

Publishing a new version

When we are ready to publish a new version of this pacakge we do the following:

1. Update the version spec in pyproject.toml (kudos for make doing this process with an alpha, or release candidate first)
2. Create a release via github using the version with a preceeding v as the tag name (ie v0.0.3-rc4). If this is an alpha, release candidate, etc make sure to check "pre-release" when creating the release (see https://docs.github.com/en/repositories/releasing-projects-on-github/managing-releases-in-a-repository)
3. Open a shell in the root of this repo for the remaing steps.
4. run git checkout <version tag> (this will put you in a detached HEAD state at the commit where you tagged the version in the previous step)
5. build the wheel and sdist packages: hatch clean && hatch build
6. make sure the wheel and sdist files have the intended version. for example, for version v0.0.3-rc4 the wheel file will be dist/tiktok_research_api_helper-0.0.3rc4-py3-none-any.whl
7. publish the packages: hatch publish (see https://hatch.pypa.io/1.12/publish/ for more info on how to configure)