Comments Moderation

Overview

The Comment Moderation feature in WhiteBeard News Suite uses AI-powered tools to automatically review and moderate user comments on your content. This intelligent system helps protect your community by filtering inappropriate content while allowing legitimate discussions to proceed without delay.

Key Features

• Automated AI classification of sensitive content
• Custom LLM-based moderation with configurable prompts
• Multi-stage moderation workflow
• Manual review and approval system
• Extensible with custom metadata for community moderation

---

How Comment Moderation Works

When a user posts a comment on your site, it goes through an automated multi-stage moderation process:

Stage 1: AI Classification

As soon as a comment is posted, an AI Classifier analyzes the content to identify potentially sensitive categories, including:

• Hate speech and discriminatory language
• Sexual or explicit content
• Violence and graphic descriptions
• Harassment and bullying
• Spam and commercial content
• Self-harm or dangerous activities
• Personal information and doxxing

If the comment is classified into any of these sensitive categories, it is automatically set to pending moderation and will not appear publicly until reviewed by a moderator.

Stage 2: LLM-Based Review

Comments that pass the initial AI classification proceed to a more sophisticated review using a Large Language Model (LLM). This stage uses a custom prompt that you define to evaluate the comment based on your specific community guidelines and standards.

The LLM analyzes the comment context, tone, and content to determine if it meets your publication’s standards. Based on this evaluation:

• Approved comments are published immediately and become visible to all users
• Flagged comments are set to pending moderation for manual review

---

Configuring Comment Moderation

Setting Up Your Custom Moderation Prompt

To configure the LLM-based moderation with your own guidelines:

1. Navigate to Administration Panel: From the main dashboard, go to Administration → Publisher Configuration

2. Access AI Assistant Settings: On the Publisher Configuration page, click on the AI Assistant tab on the right side

3. Create a Moderation Prompt:

   • Click on the Create a prompt button
   • Write your custom moderation prompt that reflects your community guidelines and standards
   • Save the prompt

4. Assign the Prompt:

   • On the same page, locate the setting called “Comment moderation prompt”
   • Select the prompt you just created from the dropdown menu
   • Save your configuration

Crafting an Effective Moderation Prompt

Your custom prompt should clearly define what is acceptable and unacceptable in your community. Consider including:

• Your publication’s values and editorial standards
• Specific examples of acceptable vs. unacceptable comments
• Context about your audience and community expectations
• Guidelines for borderline cases
• Instructions on how the AI should evaluate tone and intent

Example prompt:

Evaluate this comment based on the following guidelines:
- Accept constructive criticism and respectful disagreement
- Reject personal attacks, insults, or ad hominem arguments
- Reject comments containing profanity or vulgar language
- Accept comments that disagree with the article if they're respectful
- Reject spam, promotional content, or off-topic comments
- Consider context and intent, not just individual words

---

Managing Moderated Comments

Reviewing Pending Comments

Comments flagged for moderation can be reviewed by administrators:

1. Navigate to the Comments section in your dashboard
2. Filter by Pending Moderation status
3. Review each comment in context with the original article
4. Take one of the following actions:
   • Approve the comment to make it publicly visible
   • Decline the comment to reject it permanently

Comment Statuses

Comments in the system can have the following statuses:

• Approved: Visible to all users on the website
• Pending Moderation: Flagged by AI and awaiting manual review
• Declined: Rejected by a moderator and not visible to users

---

Advanced Features

Community Moderation with Custom Metadata

For advanced moderation workflows, you can leverage the Comments API to add custom metadata to comments. This enables community-driven moderation features:

Use Cases for Metadata

• User Voting: Allow community members to upvote or downvote comments
• Report Counts: Track how many users have flagged a comment
• Trust Scores: Assign reputation scores to commenters based on history
• Custom Rules: Implement automatic actions based on metadata thresholds

Implementation Example

Using the Comments API, you can:

1. Add metadata fields to store community votes
2. Implement custom logic to aggregate votes or reports
3. Automatically approve or escalate comments based on your rules
4. Build trust-based systems that give established community members more weight

For example, you could configure a rule where:

• Comments from users with high trust scores are auto-approved
• Comments with multiple community flags are sent to moderation
• Comments with positive vote ratios bypass additional review

Refer to the Comments API.

---

Best Practices

Moderation Tips

• Review your moderation prompt regularly to ensure it aligns with evolving community standards
• Monitor false positives and adjust your custom prompt if legitimate comments are being flagged too frequently
• Respond to declined comments with clear explanations when appropriate to educate your community
• Be consistent in applying your moderation standards
• Document your guidelines publicly so users understand your expectations

Transparency

Consider publishing your community guidelines so users know what’s expected:

• What types of comments are encouraged
• What behavior will result in moderation or bans
• How the automated moderation process works
• How to appeal moderation decisions

---

Troubleshooting

Comments being flagged too often

If legitimate comments are frequently ending up in moderation:

• Review and refine your custom LLM prompt
• Check if the AI classifier sensitivity needs adjustment
• Consider whitelisting trusted community members

Comments getting through that shouldn’t

If inappropriate comments are being approved:

• Strengthen your custom moderation prompt with more specific examples
• Add additional sensitive categories to the AI classifier
• Implement community moderation metadata for additional filtering

Need help?

If you need assistance with comment moderation configuration or have questions about advanced features, contact your WhiteBeard News Suite support team.