Content Moderation Glossary

A list of approved words, phrases, domains, or users that are exempt from moderation rules. Overrides blocklist entries.

A set of protocols and tools that allows different software applications to communicate. SafeComms provides a REST API for content moderation.

A unique authentication token used to identify and authorize requests to the SafeComms API. Keep your API keys secret to prevent unauthorized access.

Process allowing users to contest moderation decisions. Required by DSA (EU) and best practice under GDPR.

Use of algorithms and machine learning to detect and filter harmful content without human intervention. Faster and more scalable than manual review.

A user who intentionally posts harmful, abusive, or rule-violating content. May use evasion techniques to bypass filters.

Permanent or temporary suspension of a user account, preventing them from accessing the platform. Typically applied for severe or repeated violations.

A list of prohibited words, phrases, domains, or users. Content matching blocklist entries is automatically rejected or flagged.

Coordinated mass harassment, review bombing, or reporting campaigns. Often organized off-platform.

A numerical value (0-1 or 0-100) indicating how certain the moderation system is that content violates a rule. Higher scores = higher confidence.

Notice displayed before showing potentially disturbing content (violence, adult themes). Allows user choice.

US law requiring parental consent before collecting data from children under 13. Affects platforms that allow child users.

Illegal content depicting minors in sexual situations. Must be reported to authorities (NCMEC in US). Zero tolerance for storage or distribution.

Publishing private personal information about someone without consent, often with malicious intent (addresses, phone numbers, family details).

A legal contract between data controller and processor defining data handling responsibilities. Required under GDPR when using third-party services.

A specific URL in an API that performs a function. Example: `/v1/moderate/text` is the SafeComms text moderation endpoint.

Techniques used to bypass content filters, such as character substitution (@ for a), zero-width spaces, or deliberate misspellings.

When harmful content is incorrectly classified as safe. More dangerous than false positives but harder to detect.

When legitimate content is incorrectly flagged as violating rules. Reducing false positives improves user experience.

EU privacy law governing data collection, processing, and user rights. Applies to any platform serving EU users, regardless of location.

Content that attacks or demeans individuals or groups based on protected characteristics (race, religion, gender, sexual orientation, etc.).

Combination of automated and manual review. AI handles high-confidence cases; humans review edge cases and appeals.

Blocking access from a specific IP address. Less effective than account bans due to VPNs, but useful for bot traffic.

The time delay between sending a request and receiving a response. SafeComms averages 120ms latency for text moderation.

Alternative spelling using numbers and symbols (e.g., "h3ll0" for "hello"). Often used for filter evasion.

AI systems that learn patterns from data. Used in content moderation to detect toxicity, spam, and harmful content with high accuracy.

Human review of content. Slower and more expensive than automated moderation, but better at context and nuance.

A configuration defining which rules to apply, sensitivity thresholds, and actions to take. Allows different moderation for different content types.

AI technology that understands and analyzes human language. Powers contextual moderation beyond simple keyword matching.

US organization that receives CSAM reports via CyberTipline. Platforms must report suspected CSAM within 24 hours.

Content inappropriate for workplace viewing, typically sexual or violent imagery. May be allowed with age gates or content warnings.

The data sent in an API request or webhook. For SafeComms, the payload includes the content to moderate and optional metadata.

Fraudulent content attempting to steal credentials or financial information, often disguised as legitimate links or services.

Data that can identify an individual: email addresses, phone numbers, social security numbers, addresses, credit card details.

System that detects and blocks offensive language. Can be word-based (blocklists) or ML-based (contextual understanding).

Restricting the number of API requests or content submissions from a user/IP within a time window. Prevents abuse and spam.

A web service architecture using HTTP methods (GET, POST, etc.). SafeComms uses REST for easy integration with any language.

A specific moderation policy, such as "block profanity" or "flag hate speech." Multiple rules can be combined in a profile.

A library that simplifies API integration for a specific programming language. SafeComms offers SDKs for Node.js, Python, PHP, and more.

US law protecting platforms from liability for user-generated content. Allows "good faith" moderation without legal risk.

Determining the emotional tone of text (positive, negative, neutral). Used to detect hostile or aggressive language.

Hiding a user's content from others without notifying them. User thinks their posts are visible, but nobody else can see them.

Unwanted repetitive content, often commercial or phishing links. Includes comment spam, fake reviews, and bot-generated posts.

The minimum confidence score required to flag content. Lower thresholds = more sensitive (more false positives); higher = less sensitive (more false negatives).

Maximum time to wait for an API response before giving up. SafeComms recommends 5-10 second timeouts for moderation requests.

Rude, disrespectful, or hostile language likely to make users leave a conversation. Includes insults, profanity, and personal attacks.

Adjusting moderation thresholds and rules to balance false positives and false negatives for your specific use case.

Any content created by users rather than the platform: posts, comments, reviews, messages, images, videos.

Content depicting physical harm, injury, death, or graphic violence. Often prohibited or requires content warnings.

An HTTP callback that sends real-time notifications to your server when events occur, such as when content is flagged.

Newly emerging harmful content patterns that existing filters don't catch. Requires continuous model updates.