TikTok’s Research
API, which was made available to researchers in the US and Europe in
2023, offers three endpoints, which are wrapped in three
traktok
functions:
- You can search
videos with
tt_search_api
ortt_search
- You can get
basic user information with
tt_user_info_api
ortt_user_info
- You can obtain
all comments of a video with
tt_comments_api
ortt_comments
Authentication
To get access to the Research API, you need to:
- be eligible;
- create a developer account;
- and then apply for access to the research API: https://developers.tiktok.com/application/research-api
Once are approved and have your client key and client secret, you can authenticate with:
It is recommended that you run this function only once without arguments, so that your key and secret can be entered through the pop up mask and do not remain unencrypted in your R history or a script. The function then runs through authentication for you and saves the resulting token encrypted on your hard drive. Just run it again in case your credentials change.
Usage
Search Videos
TikTok uses a fine-grained, yet complicated query syntax. For convenience, I wrapped this in internally, so you can search with a key phrase directly:
tt_query_videos("#rstats", max_pages = 2L)
#>
ℹ Making initial request
✔ Making initial request [734ms]
#>
ℹ Parsing data
✔ Parsing data [21ms]
#> ── search id: NA ───────────────────────────────────────────
#> # A tibble: 0 × 11
#> # ℹ 11 variables: video_id <lgl>, author_name <chr>,
#> # view_count <int>, comment_count <int>,
#> # region_code <chr>, create_time <dttm>,
#> # effect_ids <list>, music_id <chr>,
#> # video_description <chr>, hashtag_names <list>,
#> # voice_to_text <chr>
This will match your keyword or phrase against keywords and hashtags and return up to 200 results (each pages has 100 results and 2 pages are requested) from today and yesterday. Every whitespace is treated as an AND operator. To extend the data range, you can set a start and end (which can be a maximum of 30 days apart, but there is no limit how far you can go back):
tt_query_videos("#rstats",
max_pages = 2L,
start_date = as.Date("2023-11-01"),
end_date = as.Date("2023-11-29"))
#>
ℹ Making initial request
✔ Making initial request [1.1s]
#>
ℹ Parsing data
✔ Parsing data [30ms]
#> ── search id: 7336340473469998126 ──────────────────────────
#> # A tibble: 19 × 11
#> video_id author_name view_count comment_count region_code
#> <chr> <chr> <int> <int> <chr>
#> 1 7306893… statistics… 21 4 DE
#> 2 7306307… learningca… 129 12 US
#> 3 7305014… picanumeros 57 1 ES
#> 4 7302970… smooth.lea… 5568 7 AU
#> 5 7302470… statistics… 45 1 DE
#> 6 7300977… statistics… 1403 0 DE
#> 7 7300931… rigochando 1541 5 MX
#> 8 7300922… elartedeld… 87 0 ES
#> 9 7299987… statistics… 81 1 DE
#> 10 7299657… rigochando 795 5 MX
#> 11 7299342… rigochando 375 1 MX
#> 12 7298966… rigochando 1183 2 MX
#> 13 7296911… biofreelan… 2537 5 MX
#> 14 7296911… biofreelan… 1363 0 MX
#> 15 7296911… biofreelan… 680 1 MX
#> 16 7296688… mrpecners 60 2 US
#> 17 7296518… l_a_kelly 10 5 GB
#> 18 7296498… mrpecners 19 0 US
#> 19 7296288… casaresfel… 266 0 AR
#> # ℹ 6 more variables: create_time <dttm>,
#> # effect_ids <list>, music_id <chr>,
#> # video_description <chr>, hashtag_names <list>,
#> # voice_to_text <chr>
As said, the query syntax that TikTok uses is a little complicated,
as you can use AND, OR and NOT boolean operators on a number of fields
("create_date"
, "username"
,
"region_code"
, "video_id"
,
"hashtag_name"
, "keyword"
,
"music_id"
, "effect_id"
, and
"video_length"
):
Operator | Results are returned if… |
---|---|
AND | …all specified conditions are met |
OR | …any of the specified conditions are met |
NOT | …the not conditions are not met |
To make this easier to use, traktok
uses a tidyverse
style approach to building queries. For example, to get to the same
query that matches #rstats against keywords and hashtags, you need to
build the query like this:
query() |> # start by using query()
query_or(field_name = "hashtag_name", # add an OR condition on the hashtag field
operation = "IN", # the value should IN the list of hashtags
field_values = "rstats") |> # the hashtag field does not accept the #-symbol
query_or(field_name = "keyword", # add another OR condition
operation = "IN",
field_values = "#rstats")
#> S3<traktok_query>
#> └─or: <list>
#> ├─<list>
#> │ ├─field_name: "hashtag_name"
#> │ ├─operation: "IN"
#> │ └─field_values: <list>
#> │ └─"rstats"
#> └─<list>
#> ├─field_name: "keyword"
#> ├─operation: "IN"
#> └─field_values: <list>
#> └─"#rstats"
If #rstats is found in either the hashtag or keywords of a video,
that video is then returned. Besides checking for EQ
ual,
you can also use one of the other operations:
Operation | Results are returned if field_values are… |
---|---|
EQ | equal to the value in the field |
IN | equal to a value in the field |
GT | greater than the value in the field |
GTE | greater than or equal to the value in the field |
LT | lower than the value in the field |
LTE | lower than or equal to the value in the field |
This makes building queries relatively complex, but allows for fine-grained searches in the TikTok data:
search_df <- query() |>
query_and(field_name = "region_code",
operation = "IN",
field_values = c("JP", "US")) |>
query_or(field_name = "hashtag_name",
operation = "EQ", # rstats is the only hashtag
field_values = "rstats") |>
query_or(field_name = "keyword",
operation = "IN", # rstats is one of the keywords
field_values = "rstats") |>
query_not(operation = "EQ",
field_name = "video_length",
field_values = "SHORT") |>
tt_search_api(start_date = as.Date("2023-11-01"),
end_date = as.Date("2023-11-29"))
#>
ℹ Making initial request
✔ Making initial request [734ms]
#>
ℹ Parsing data
✔ Parsing data [17ms]
search_df
#> ── search id: 7336340473470030894 ──────────────────────────
#> # A tibble: 2 × 11
#> video_id author_name view_count comment_count region_code
#> <chr> <chr> <int> <int> <chr>
#> 1 72966888… mrpecners 60 2 US
#> 2 72964986… mrpecners 19 0 US
#> # ℹ 6 more variables: create_time <dttm>,
#> # effect_ids <list>, music_id <chr>,
#> # video_description <chr>, hashtag_names <list>,
#> # voice_to_text <chr>
This will return videos posted in the US or Japan, that have rstats
as the only hashtag or as one of the keywords and have a length of
"MID"
, "LONG"
, or "EXTRA_LONG"
.1
Get Basic User Information
There is not really much to getting basic user info, but this is how you can do it:
tt_user_info_api(username = c("tiktok", "https://www.tiktok.com/@statisticsglobe"))
#> # A tibble: 2 × 8
#> display_name follower_count following_count is_verified
#> <chr> <int> <int> <lgl>
#> 1 TikTok 78785286 30 TRUE
#> 2 Statistics Glo… 289 1 FALSE
#> # ℹ 4 more variables: likes_count <int>, video_count <int>,
#> # avatar_url <chr>, bio_description <chr>
Obtain all Comments of a Video
There is again, not much to talk about when it comes to the comments API. You need to supply a video ID, which you either have already:
tt_comments_api(video_id = "7302470379501604128")
#>
ℹ Making initial request
✔ Making initial request [7.2s]
#>
ℹ Parsing data
✔ Parsing data [18ms]
#> ── search id: ─────────────────────────────────────────────
#> # A tibble: 1 × 7
#> text video_id create_time id like_count
#> <chr> <chr> <int> <chr> <int>
#> 1 and why would we do… 7302470… 1700243424 7302… 0
#> # ℹ 2 more variables: parent_comment_id <chr>,
#> # reply_count <int>
Or you got it from a search:
tt_comments_api(video_id = search_df$video_id[1])
#>
✔ Making initial request [55.7s]
#>
ℹ Parsing data
✔ Parsing data [18ms]
#> ── search id: ─────────────────────────────────────────────
#> # A tibble: 2 × 7
#> like_count parent_comment_id reply_count text video_id
#> <int> <chr> <int> <chr> <chr>
#> 1 1 7296688856609475882 1 So co… 7296688…
#> 2 0 7296690681388204831 0 Thank… 7296688…
#> # ℹ 2 more variables: create_time <int>, id <chr>
Or you let the function extract if from a URL to a video:
tt_comments_api(video_id = "https://www.tiktok.com/@nicksinghtech/video/7195762648716152107?q=%23rstats")
#>
ℹ Making initial request
✔ Making initial request [2m 16.7s]
#>
ℹ Parsing data
✔ Parsing data [19ms]
#> ── search id: ─────────────────────────────────────────────
#> # A tibble: 96 × 7
#> create_time id like_count parent_comment_id
#> <int> <chr> <int> <chr>
#> 1 1675394834 719576596990925… 314 7195762648716152…
#> 2 1675457114 719603344301613… 232 7195762648716152…
#> 3 1675458796 719604066348022… 177 7195762648716152…
#> 4 1675395061 719576692726561… 166 7195765969909252…
#> 5 1675624739 719675339339355… 71 7195762648716152…
#> 6 1675465779 719607064381200… 71 7195762648716152…
#> 7 1675494738 719619490971140… 27 7195762648716152…
#> 8 1675691471 719703995331384… 17 7196040663480222…
#> 9 1675656122 719688817955866… 16 7195762648716152…
#> 10 1675440749 719596313215706… 16 7195762648716152…
#> # ℹ 86 more rows
#> # ℹ 3 more variables: reply_count <int>, text <chr>,
#> # video_id <chr>
And that is essentially it. Note, that if you find the functionality of the Research API lacking, there is nothing that keeps you from using the unofficial API functions.
Dealing with rate limits and continuing old searches
At the moment of writing this vignette, the TikTok rate limits the Research API as follows:
Currently, the daily limit is set at 1000 requests per day, allowing you to obtain up to 100,000 records per day across our APIs. (Video and Comments API can return 100 records per request). The daily quota gets reset at 12 AM UTC. [Source]
Depending on what you would like to do, this might not be enough for you. In this case, you can actually save a search and pick it back up after the reset. To facilitate this, search result objects contain two extra pieces of information in the attributes:
search_df <- query() |>
query_and(field_name = "region_code",
operation = "IN",
field_values = c("JP", "US")) |>
tt_search_api(start_date = as.Date("2023-11-01"),
end_date = as.Date("2023-11-29"),
max_pages = 1)
#>
ℹ Making initial request
✔ Making initial request [1.9s]
#>
ℹ Parsing data
✔ Parsing data [20ms]
attr(search_df, "search_id")
#> [1] "7336340473470063662"
attr(search_df, "cursor")
#> [1] 100
When you want to continue this search, whether because of rate limit
or because you decided you want more results, you can do so by providing
search_id
and cursor
to
tt_search_api
. If your search was cut short by the rate
limit or another issue, you can retrieve the results already received
with search_df <- last_query()
. search_df
will in both cases contain the relevant search_id
and
cursor
in the attributes:
search_df2 <- query() |>
query_and(field_name = "region_code",
operation = "IN",
field_values = c("JP", "US")) |>
tt_search_api(start_date = as.Date("2023-11-01"),
end_date = as.Date("2023-11-29"),
# this part is new
start_cursor = attr(search_df, "cursor"),
search_id = attr(search_df, "search_id"),
####
max_pages = 1)
#>
ℹ Making initial request
✔ Making initial request [5.1s]
#>
ℹ Parsing data
✔ Parsing data [21ms]
attr(search_df2, "search_id")
#> [1] "7336340473470063662"
attr(search_df2, "cursor")
#> [1] 200
Note that the cursor is not equal to how many videos you got before,
as the API also counts videos that are “deleted/marked as private by
users etc.” [See max_count
in Query
Videos].