Chat Moderation in WisePub

WisePub offers powerful chat moderation features to help you maintain a professional and safe environment during your webinars and live events. This guide covers both manual and AI-powered moderation options.

Enabling Chat Moderation

Basic Moderation Setup

1. Access Chat Settings
   • While in your chat room, click on the Settings button
   • Navigate to Chat Options
   • Toggle Moderation to enable it
2. How Basic Moderation Works
   • All messages from non-moderators and non-admins are sent to a moderation queue
   • Messages must be manually reviewed before appearing in the chat
   • Access the moderation queue by clicking on the queue indicator
   • Approve or reject messages as needed

AI Moderation (Recommended)

AI moderation allows you to run webinars with minimal manual oversight by automatically filtering inappropriate content.

Setting Up AI Moderation

1. Configure AI Settings in Admin Panel

1. Navigate to Admin Panel
   • Go to your WisePub admin dashboard
   • Click on Settings
2. Create/Configure AI Settings
   • Look for AI Settings (or create a new setting if it doesn’t exist)
   • Set the following parameters:
     • Name: AI
     • Environment: Production
     • Type: AI Settings

2. Choose the Optimal AI Model

Recommended Configuration:

• Provider: Groq (free API key included with WisePub)
• Model: Llama 3.1 8B instance
• Why this model: Fastest processing for near-instant message approval

3. Customize Your Moderation Prompt

The AI moderation prompt determines what content gets filtered. Here’s what the default prompt covers:

Default Filtering Criteria

• Profanity and Swearing: Automatic filtering of inappropriate language
• Misinformation: False or misleading information
• Spam: Repetitive or promotional content
• Hate Speech and Discrimination: Offensive content targeting individuals or groups
• Negative Reviews/Complaints: Criticism that may disrupt the event flow
• Toxicity and Personal Attacks: Hostile or defamatory messages
• Personal Information: Phone numbers and email addresses shared by users

Important Prompt Requirements

• Always end your custom prompt with “pass and review”
• Include response examples for the AI
• Specify clear criteria for what should be moderated

Here is the current default prompt:

You are a content moderator for a chat application. Analyze each message to determine if it requires human review or can pass through automatically.

Evaluation Criteria
Flag for REVIEW if the message contains:

Personal Attacks & Defamation
Direct insults, character attacks, or defamatory statements about individuals
Threats or intimidation directed at specific people

Personal information
Any phone number or email shared by a user

Toxicity & Negativity
Abusive language, excessive profanity in aggressive contexts
Content intended to harm, demoralize, or create hostile environments
Trolling or deliberately provocative behavior

Product/Service Complaints
Negative reviews, complaints, or criticism about products/services being offered
Disputes about transactions, quality, or business practices

Hate Speech & Discrimination
Content targeting individuals/groups based on race, religion, gender, sexuality, nationality, or other protected characteristics
Harassment campaigns or coordinated attacks

Spam & Commercial Violations
Unsolicited advertising, promotional content, or link farming
Repetitive or bot-like messaging patterns

Misinformation & Harm
False information that could cause real-world harm
Content promoting dangerous activities or illegal behavior

Response Format
Required: Start your response with either "PASS" or "REVIEW"
Required: Follow with exactly 5 words explaining your decision
Decision Guidelines

PASS: Content appears safe for the community and doesn't violate guidelines
REVIEW: Content contains potential violations or requires human judgment
When in doubt: Choose REVIEW - human moderators provide final judgment

Examples

REVIEW: Contains personal attack language
PASS: Normal conversation about hobbies
REVIEW: Spam link detected in message

4. Save and Test Your Configuration

1. Save Settings: Click save after configuring your AI moderation prompt
2. Test the System:
   • Log out and access your room as a guest
   • Send various test messages to verify the moderation is working
   • Check that appropriate content passes through while inappropriate content is blocked

System Performance

High Throughput Support:

• Supports up to 1,000 moderation requests per minute
• Suitable for very popular chat rooms with high message volume
• Near-instant processing for seamless user experience

Best Practices

For Event Hosts

• Pre-Event Setup: Configure and test AI moderation before your live event
• Monitor Initially: Keep an eye on the moderation queue during your first few events to fine-tune settings
• Custom Prompts: Tailor your AI prompt to match your event’s specific requirements and audience

For Moderators

• Backup Oversight: Even with AI moderation, having human moderators available is recommended for complex situations
• Quick Response: Use the moderation queue to quickly review any edge cases the AI may flag for human review

Content Guidelines

• Clear Expectations: Inform participants about your chat guidelines at the beginning of events
• Consistent Enforcement: Ensure your AI prompt aligns with your stated community standards

Troubleshooting

Common Issues

• Messages Not Being Moderated: Verify AI settings are saved and the correct model is selected
• Too Many False Positives: Adjust your moderation prompt to be less restrictive
• Slow Processing: Ensure you’re using the recommended Llama 8B 1.8B instance

Getting Help

If you encounter issues with chat moderation, check your AI settings configuration and ensure all required fields are properly filled out. The system works best when the moderation prompt is clear and specific about what content should be filtered.