Class Descriptions – Hate & Bullying
How to Use This Page – Overview
This page gives a more detailed overview of Hive's visual moderation classes related to hate, white supremacy, terrorism, and bullying. If you need more details on these classes after reading our main Visual Moderation page, look here. We'll enumerate, as clearly as possible, which types of subject matter are covered by each class in our models.
Because all platforms have different moderation requirements and risk sensitivities, we recommend that you consult these descriptions carefully as you decide which classes to build into your moderation logic. At the end of the day, it's up to you to decide which classes are important to monitor based on your content policies.
NOTE:
To determine which class(es) cover specific types of visual content, it may be helpful to search this page (Ctrl/Cmd + F) with terms for that subject matter rather than looking for it in specific class descriptions.
General Notes
-
Hive's visual classifier is multi-headed. Each model head defines a group of categorizations we call classes. Each model head includes at least one positive class (e.g., yes_hot_dog) and a negative class (e.g., no_hot_dog). Scores returned for each class correlate with the model's certainty the image meets our ground truth definition for the category. This page attempts to explain these ground truth definitions as clearly as possible.
-
The model makes classifications for each model head independently. In other words, if an image scores highly in multiple classes, the image meets our definitions for each class. Confidence scores from each model head are generated separately and are not correlated in and of themselves. It's easiest to think of this as asking multiple, narrower models (that may or may not overlap in scope) to each make a prediction on an image.
-
As a corollary, a high confidence score in negative classes (e.g., no_nazi, no_kkk) does not mean the image is clean in general. This is simply the logical opposite of the positive classification; the subject matter captured by the positive class in that model head is not present. For example, an image that scores 0.99 in no_kkk can still score highly in yes_nazi any other class trained to flag other subject matter. For this reason, we will describe these negative classes as a non-exhaustive list of subject matter that is not captured by the positive classes in that model head but that might be helpful to distinguish what is and is not flagged by that model head (e.g., borderline content, content captured by other classes).
-
For all model heads described below, the model classifies animations, drawings, diagrams, paintings, graphics, illustrations, etc. in the same way as photographs or photorealistic images.
Nazi Head
This model head can be used to flag Nazi symbols, uniforms, and depictions of Hitler. These classes do not capture the same for other white supremacy or hate groups.
Detailed Class Descriptions
yes_nazi: the image depicts Nazi symbolism, such as:
- Swastikas in the style used to represent the Nazi party (i.e., clockwise form), including graffiti, tattoos, flags, clothing/uniforms etc.
- SS Insignias similarly depicted
- Nazi uniforms, personnel, and equipment if clearly identifiable as such
no_nazi: the image does not depict Nazi symbolism. To be clear, the following would not be flagged by yes_nazi:
- Images including the words "Nazi," "SS," "Hitler," and the like if no accompanying visual symbolism is present
- Symbols that are used outside of Nazi contexts, such as the left facing Swastika symbol in other Eurasian cultures that pre-date Nazism
- Actors or other people with Hitler-style mustaches if not dressed as/depicting Hitler
Terrorism Head
This model head can be used to catch flags and symbols associated with terrorist organizations such as Al-Qaeda and ISIS. It does not flag terrorist personnel, photographs of known terrorists, etc.
Detailed Class Descriptions
yes_terrorist: the image depicts symbols or flags associated with a terrorist organization, including:
- ISIS Flags, whether in digital form or depicted on actual flags, clothing, tattoos, etc.
- Al-Qaeda flags and symbols in the same formats
no_terrorist: the image does not depict terrorist flags or symbols. Specifically, the following would not flagged by yes_terrorist:
- Text containing the names of terrorist organizations or terrorist concepts
- Photographs of terrorist leaders such as Osama Bin Laden
- Members of terrorist organizations (e.g., armed personnel) if no flag/symbol is displayed
- Altered or parody versions of these flags and symbols if not visually similar or identical.
KKK Head
This model head can be used to flag images depicting symbols, clothing, etc. related to the Ku Klux Klan and similar white supremacist groups. It does not similar symbols related to Nazism, the Confederacy, etc. on their own. Animations, drawings, and illustrations follow the same definitions as photographs for these classes.
Detailed Class Descriptions
yes_kkk: the image depicts symbols associated with KKK, including:
- The KKK logo/insignia, either alone or depicted on flags, clothing, tattoos, documents, etc.
- KKK robes, hooded or otherwise
- Pointed hoods in the KKK style (i.e., with face covering), whether or not accompanied by a robe
- Burning crosses
no_kkk: the image does not depict the above symbols, clothing etc. To be clear, the following would not be captured by yes_kkk:
- Text containing the words "KKK" or "Ku Klux Klan"
- Photographs of KKK leaders if no KKK clothing/symbols are shown
- Photographs of KKK activity if no KKK clothing/symbols are shown
Confederate Flag Head
This model head can be used to identify images of the Confederate flag, whether on its own or depicted on clothing, objects, etc. This classification only applies to the common representation – the "stars and bars" – NOT the historical flag of the Confederacy, which is visually distinct. Graphics, illustrations, and animations follow the same definitions as photographs for these classes.
Detailed Class Descriptions
yes_confederate_flag: the image depicts the Confederate flag or an obvious variation. This includes:
- The Confederate flag ("stars and bars" style) depicted in a graphic, on a flag or clothing, tattoos, or objects such as automobiles.
- Edited versions or graphics where it clear that the Confederate flag is a component
- Obscured or half flags, such as flags not being flown, or in graphics/edited versions
- The old state flag of Mississippi (which includes the "stars and bars" in the upper left)
no_confederate_flag: the image does not depict a Confederate flag or a clear variant. To be clear, the following would not be flagged by yes_confederate_flag:
- The historical, original flag of the Confederacy (circularly arranged stars on a field of thick red and white stripes)
- The new state flag of Mississippi
- Flags of countries or states that are visually similar to the Confederate flag (e.g., Norwegian flag, union jack, etc.)
- Historical objects associated with the Confederacy or Confederate army
- Other flags or symbols associated with white supremacy (e.g., Nazi or KKK symbols)
Middle Finger Head
This model head can be used to flag images where someone is displaying the middle finger intentionally (e.g., as an insult)
Detailed Class Descriptions
yes_middle_finger: the image depicts a person giving the middle finger intentionally.
no_middle_finger: the middle finger sign is not shown, or the middle finger is shown but is part of a natural or unintentional hand position.
Updated 6 months ago