text
stringlengths 0
202
|
|---|
**Introduction**
|
This README is a descriptive companion to the downloadable CSV of TikTok's Community Guidelines Enforcement Report.
|
The aim of this document is to explain the overall structure of the data, recommended analysis methods, and specific value definitions. This information is provided to aid manual or automated analysis.
|
TikTok regularly reports and releases this data because there is interest from governments, media, and the public on quantifiable facts about how TikTok performs content moderation.
|
**Structure of data**
|
The data file is in a flat format, whereby there is only one data point per row.
|
This format allows us to flexibly add more metrics, granularities, and time periods, while retaining a consistent structure.
|
The data file has four key sections as noted by the column names / headings.
|
**Metrics**
|
"Metric" is the full name of the unit of measurement, which includes:
|
- The entity - the countable object of interest, such as 'Video' or 'Account';
|
- The aspect - the high level enforcement characteristic, such as 'Removal' or 'Ban';
|
- And whether the metric is a total, rate/share (%), or other value.
|
For analysis purposes, it is most ideal to filter by the "Metric" field.
|
There is no 'All' value for this field, as results must be specific to a metric.
|
**Time period**
|
Time period defines an observed start and end time for metric calculations.
|
For the Community Guidelines Enforcement Report we provide data on a quarterly basis.
|
Unless otherwise noted, we report on closed cases for the specified period, NOT on open cases.
|
This applies to removal cases and appeal cases.
|
This allows us to have the full information about case outcomes and help ensure stability of results over time.
|
- "Period type" - Defines the period categorization, such as 'Quarterly';
|
- "Period" - Defines the specific period value, such as 'Jul-Sep 2022'.
|
These fields are dimensions suitable for filtering.
|
For analysis purposes, it is sufficient to filter the "Period" field.
|
There is no 'All' value for this field, as results must be specific to a time period.
|
**Granularities**
|
We produce multiple granularities to allow for deeper insight into enforcement efforts:
|
- "Policy type" - Defines the policy categorization, such as 'Ban reason' or 'Policy';
|
- "Issue" - The specific policy label under the policy categorization, such as 'Privacy & Security';
|
- "Task type" - Defines the moderation system categorization, such as 'All' or 'task_type';
|
- "Task" - The specific moderation system label under the task categorization, such as 'Automation' or 'Human moderation';
|
- "Location" - Defines the geographical coverage categorization, such as 'All' or 'Market';
|
- "Market" - The specific geographical coverage label under the location categorization, such as 'Australia' or 'Portugal'.
|
There are two major types of policy, "Policy" and "Sub-policy".
|
"Sub-policy" provides a more detailed, granular reporting of "Policy".
|
The reporting of "Location" typically refers to the originating country or region where the content was published, or where an account was last active, unless otherwise noted.
|
These fields are dimensions suitable for filtering.
|
For analysis purposes, it is sufficient to filter to the "Issue policy", "Task", and "Market" fields.
|
These fields can contain 'All' values - these are special and important values to denote total level (i.e. non-granular) analysis.
|
**Result**
|
The "Result" field is numerical fact data.
|
This data is already pre-aggregated to the associated dimensions.
|
The data may refer to a count, or a percentage, depending on the metric.
|
Whenever a metric is calculated which includes 1000 unique users or fewer, these results are deliberately omitted to help preserve privacy.
|
No dataset card yet
- Downloads last month
- 7