Skip to main content

😊 Sentiment Moderation

The Sentiment Moderation feature leverages AI to detect and act on messages with negative sentiment about the project or community. This automated system helps maintain a positive atmosphere and ensures that potentially harmful messages are addressed swiftly.

Where to configure it

In the webapp: Open your workspace, select the Telegram group under Channels in the sidebar, then select AI Features, then the AI Moderator tab. Sentiment moderation is the Hostility rule (Built-in) under Messages. Its switch saves instantly; open the rule for its full settings, which save with the Save bar in that dialog. While the AI Moderator itself is on, hostility is enforced by its judge, and the action, duration and veto settings only take effect when the moderator is off.

In Telegram (bot menu): the /settings menu route below still works.

Overview​

Each message is labelled positive, neutral or negative. When enough of a member's recent messages are negative, the AI double-checks them and takes the action you chose. Members can veto the action through a vote.

Configuration in Telegram (bot menu)​

To access Sentiment Moderation settings:

  1. Use the /settings command in your chat
  2. Navigate to 🤖 AI → 😊 Sentiment Moderation
  3. Use the Toggle Sentiment option to turn the feature 🟢 Enabled or 🔴 Disabled

Sentiment Moderation Settings​

Threshold​

The share of the member's recent messages that must be labelled negative before action is considered. Fren One looks at the member's last few messages (the Consecutive messages count in the webapp), counts how many were labelled negative, and when that share reaches the threshold it asks the AI to confirm before acting.

Available options:

  • 0.7 - Very sensitive: 70% of the recent messages must be negative
  • 0.75 - Moderately sensitive
  • 0.8 - Balanced: 80% must be negative
  • 0.85 - Less sensitive
  • 0.9 - Least sensitive: 90% must be negative
  • Other - Custom value

The threshold is about how many recent messages are negative, not how strongly negative each one is.

Understanding the threshold:

  • A higher threshold requires more of the recent messages to be negative, so it is less sensitive
  • A lower threshold acts on a smaller share of negative messages, so it catches more members
  • The webapp shows the same value as a percentage: Negative threshold (%)

💡 Be careful with low threshold settings: the bot will act when only a small share of a member's recent messages are negative.

Important consideration: A user experiencing an error or other difficulty (for example, "I can't complete my registration" or "Registration doesn't work") is technically a negative sentiment, and a very low threshold setting may result in action being taken against that user.

Threshold Calibration

You may need to take some time to tweak this setting to fit the general culture and sentiment of your community, or make specific changes to the threshold when you are running a campaign that might result in more reports of errors or problems.

Messages​

The number of recent messages analyzed per user. The webapp calls it Consecutive messages.

How it works: The bot looks at the member's last X messages (counting only messages since their last restriction) and works out what share of them were labelled negative. Nothing is evaluated until the member has sent at least X messages. This ensures the bot looks at patterns rather than single messages.

Choosing the right number:

  • Fewer messages (3-5): Faster response to negative behavior, but more prone to false positives
  • More messages (10-20): More accurate assessment of consistent negativity, but slower to detect problematic users
Example

With Messages set to 5 and the threshold at 0.8, the bot looks at the member's last 5 messages. If at least 4 of them (80%) were labelled negative, it asks the AI to confirm, and acts if the AI agrees.

Veto​

Allow group members to vote and veto actions taken against users deemed negative.

Enable/Disable:

  • 🟢 Enabled: Community can vote to reverse bot actions
  • 🔴 Disabled: Bot actions are final

When enabled, configure:

Votes Needed​

The minimum number of votes required to veto an action against a user.

Common configurations:

  • 3-5 votes: Small communities
  • 5-10 votes: Medium communities
  • 10-20+ votes: Large communities

How veto works:

  1. Bot takes action against user (mute/kick/ban/warn)
  2. Bot announces the action with veto option
  3. Community members can vote to "save" the user
  4. If enough votes are reached, the action is reversed
Community Participation

The veto system creates community buy-in and prevents false positives while still maintaining automated protection.

Example: If Votes Needed is set to 5, community members must collectively cast 5 "save" votes to overturn the bot's decision and restore the user.

Action​

Specifies the action to take against members whose recent messages are consistently negative.

Available actions:

  • Mute: Temporarily restrict the user from sending messages
  • Kick: Temporarily remove the user from the group (can rejoin)
  • Ban: Permanently remove the user from the group
  • Warn: Issue a warning to the user

Duration (for Mute action)​

If Mute is selected, specify the duration:

Common durations:

  • 1 hour: Brief timeout
  • 6 hours: Extended timeout
  • 1 day: Day-long restriction
  • 7 days: Week-long restriction
  • Custom: Set your own duration
Action Severity

Start with less severe actions (Warn or short Mute) and monitor results before escalating to Kick or Ban. The goal is to maintain positivity, not alienate legitimate community members.

Troubleshooting

Too many false positives?

  • Increase threshold (try 0.85 or 0.9)
  • Increase number of messages analyzed
  • Enable veto system
  • Review what's being flagged and adjust

Missing obvious toxic users?

  • Decrease threshold (try 0.75 or 0.7)
  • Decrease number of messages analyzed
  • Consider combining with other moderation tools

Veto system not working?

  • Verify veto is enabled
  • Check votes needed isn't too high for community size
  • Ensure community understands how to veto

Actions not being taken?

  • Verify sentiment moderation is enabled
  • Check bot has necessary permissions
  • Ensure threshold and messages are configured

Technical issues being flagged?

  • Increase threshold temporarily during issues
  • Communicate with community about temporary changes
  • Use manual moderation during technical problems
  • Consider disabling during major incident responses

Sentiment detection seems inaccurate?

  • AI sentiment may need more context (increase messages)
  • Non-English content may be less accurate
  • Sarcasm and irony can be misinterpreted
  • Consider language and cultural factors

On this page