This feature is available for Reputation+, Engage+, and Agent+ clients.
When to Use a Custom Classifier
Use a custom AI NLP classifier when you need to:
Tag comments based on tone, topic, or intent
Filter specific mentions (e.g., product issues, competitor references)
Automate moderation or prioritization rules at scale
💡 Example: Want to auto-tag comments about shipping delays, product sizing, or positive testimonials? A classifier handles this instantly.
Start Here: Choose the Right Workflow Type
To create an AI-powered classifier:
Go to Control Panel > Workflows
Select or create a tagging workflow
When prompted to define what to tag, choose AI + NLP
Why AI + NLP?
Use this when you need flexible, brand-specific tagging that adapts to evolving language and nuanced phrasing.
Overview: The 6-Step Classifier Builder
BrandBastion’s NLP classifier setup walks you through:
Settings – Define your model’s purpose
Filters – Narrow the input pool
Categories – Describe what you want to detect
Tasks – Link each category to a tag
Model Training – Fine-tune using real examples
Testing – Validate tagging accuracy before launch
Step 1: Kick Off Your Workflow (Settings)
Start by defining what you want to classify and why it matters.
Workflow title: Use something clear and purpose-specific (e.g., Pre-Purchase Interest Classifier)
Permissions: Toggle “Advanced user permissions” if access should be restricted
Model purpose: Choose one or more:
Moderation – Flag harmful or off-topic comments
Responding – Detect comments that need replies
Insights – Group messages into key themes for reporting
📌 Your selection here impacts how the model prioritizes tagging logic.
Step 2: Apply Filters
Filters help you narrow the scope of content going into your model.
Include filters: Limit input by channel, language, platform, message type, etc.
Exclude filters: Block specific words, phrases, or conditions that should not be tagged.
Step 3: Define categories
Define the types of messages you want the classifier to identify. Use clear descriptions to help distinguish between categories.
For each category, complete these fields:
What kind of messages should be tagged? Describe which messages belong in the category, including common phrases and context (minimum 100 characters).
Real examples that should be tagged (optional): Add real or hypothetical comments that match the category to help improve accuracy.
What should be excluded (optional): Describe similar messages that should not be tagged to help reduce misclassification. You can also provide examples.
💡 Tip: If categories overlap, refine their descriptions and exclusions to help the model distinguish between them.
Step 4: Assign Tasks and Tags
Match each category to a tag that will be applied when a match is found.
You can:
Use a custom tag
Assign multiple tasks per model
🔁 You can reuse the same tag across tasks — just make sure each task’s logic is distinct.
Step 5: Train the Model
This is where AI learning kicks in.
Review and confirm how sample comments should be tagged
Adjust the tagging decision and confidence level if needed
Add more examples to improve accuracy
👍 Mark confidently-tagged examples to train stronger predictions
⚠️ Label edge cases carefully to avoid false positives
Step 6: Test and Validate
Before going live, paste sample comments into the test window.
The system will show which comments would be tagged, and how
Use this to:
Spot tagging errors
Fine-tune filters, exclusions, or examples
Confirm readiness
✅ When satisfied, click Create and Enable Workflow
Post-Launch: What You Can Do with Tags
Once deployed, your model will start tagging messages, which you can use for:
Auto-hiding of toxic or off-topic comments
Alerts when key topics appear
Auto-archiving low-priority content
Sentiment overrides for specific categories
Auto-assigning messages to certain users