AI Content Moderation
Creating safer communities for all
Last updated
Creating safer communities for all
Last updated
At Amity, we leverage AI for automatic content moderation to ensure that you have a safe online environment. We offer two types of AI moderation:
Pre-Moderation: Content is automatically reviewed before it is posted. Our AI system scans the content upon upload, and generates a confidence value. If this confidence value is equal to, or higher than the configured threshold, the content will be blocked from being posted.
Post-Moderation: Post-moderation occurs after content has been posted. Our AI system considers two main factors, flagConfidence
and blockConfidence
. It scans the content posted and generates a confidence value. Based on this value, different actions are automatically taken - depending the configured threshold.
To enable the AI Content Moderation feature, please contact our support team. Once the support team raise a task ticket, our development team will enable the feature for you.
Our AI pre-moderation feature is currently only available for image moderation. It ensures that all uploaded images are scanned for inappropriate, offensive, and undesirable content before it is published. Our AI system scans and detects undesirable content in images across 4 categories:
Nudity
Suggestive content
Violence
Disturbing
To enable image moderation, login to Amity Console. Under Moderation > Image Moderation, toggle "Enable image moderation" to "ON".
Once you've enabled image moderation, you will need to set the confidence level for each moderation category. In the context of content moderation, confidence levels represent the degree of certainty the AI system has in identifying specific content categories within an uploaded image. It is crucial to set confidence levels for each moderation category to fine-tune the system's sensitivity in a way that meets your moderation needs.
By default, confidence levels are set to "0" for each category. A confidence level of "0" implies a low threshold, making the system more likely to block images, potentially resulting in false positives, even if the content is not inappropriate. For more accurate and reliable results, it is recommended to set confidence levels at a higher threshold. A higher confidence value indicates a stronger certainty in the content classification.
When enabled, our AI system will scan images uploaded in posts, comments, and messages and return a confidence value. If this confidence value is equal to, or higher than the configured threshold, the content will be blocked from being posted. The undesirable image will need to be removed by the user in order for their post, comment, or message to be posted.
Our AI post-moderation feature goes beyond the basics, offering an enhanced moderation experience. It scans a wider range of content type across more moderation categories, and provides more flexibility in configuring actions to be taken based on flag and block confidence levels.
Our AI post-moderation feature supports moderating text, image, and video content within messages and posts, and text and image in comments.
We will be supporting AI livestream post-moderation soon!
Our AI text post-moderation feature detects and moderates text on posts, comments, and messages that are:
Sexually explicit or adult in certain situations
Sexually suggestive or mature in certain situations
Offensive in certain situations
Our AI image and video post-moderation feature detects images and videos on posts and messages, and images in comments that contain imagery depicting:
Adult Toys
Air crash
Alcohol
Alcoholic Beverages
Bare-chested Male
Corpses
Drinking
Drug Paraphernalia
Drug Products
Drug Use
Drugs
Emaciated Bodies
Explicit Nudity
Explosions and blasts
Extremist
Female Swimwear Or Underwear
Gambling
Graphic Female Nudity
Graphic Male Nudity
Graphic Violence Or Gore
Hanging
Hate Symbols
Illustrated Explicit Nudity
Male Swimwear Or Underwear
Middle Finger
Nazi Party
Nudity
Partial Nudity
Physical Violence
Pills
Revealing Clothes
Rude Gestures
Self Injury
Sexual Activity
Sexual Situations
Smoking
Suggestive
Tobacco
Tobacco Products
Violence
Visually Disturbing
Weapon Violence
Weapons
White Supremacy
Moderation is performed using two main factors - flagConfidence
and blockConfidence
. When a post, comment, or message is successfully created, our AI system will automatically scan both the text and media, generating a confidence value as a result. If this confidence value falls below flagConfidence
, the post, comment, and message passes moderation. In cases where the confidence value falls between flagConfidence
and blockConfidence
, our moderation feature will flag the content for review.
You will be able to review the flagged content in Amity Console or listen to Amity's real-time flagged post, flagged comment, or flagged message event through your webhook on the server side - allowing you to implement appropriate actions to be taken based on your moderation policies. Finally, if the confidence value surpasses blockConfidence
, the system deletes the content altogether, ensuring a responsive approach to maintaining a safe online environment.
As a default configuration, all categories have an initial flagConfidence
value of 40, and an initial blockConfidence
value of 80.
To use the enhanced moderation APIs, specify the appropriate API endpoint on each HTTP request. Each data center has a unique endpoint, so it's essential to adjust it accordingly. By selecting the correct endpoint associated with the right location, you ensure faster response times and optimize the overall performance of your API requests.
Get moderation confidence level API can be used to retrieve confidence level for each moderation category. The API will return list of moderation category along with the corresponding confidence level.
Set moderation confidence level API can be used to set confidence level to any moderation category. The API will return list of moderation category along with the corresponding confidence level.
We recommend adjusting the confidence levels for each category based on your moderation policies and the needs of your community. By doing so, you can ensure that Enhance Moderation provides the optimal level of moderation for your platform.
Name | Data Type | Description |
---|---|---|
Region | API Endpoint |
---|---|
category
String
Name of each moderation's category
flagConfidence
Number
Value of moderation category’s flag confidence
blockConfidence
Number
Value of moderation category’s block confidence
moderationType
String
Value of moderation type for the category. There are 2 possible values, "text" or "media"
Europe
Singapore
United States
Moderation categories and confidence level JSON
Request has succeeded
Request has succeeded.