NLP Classifier

Quick Start

To get started:

• Choose a trained model from the Model to use dropdown
• Set the Threshold value (0-1) to control prediction confidence
• Send text via msg.payload.text
• Receive classification results in msg.payload

Configuration

Model to use (required)

Select a pre-trained model from the dropdown menu. Models must be trained beforehand using the text classifier trainer block.

Threshold (required)

Confidence threshold (0 to 1) for filtering classification results. Default: 0.5

Note: Lower values return more predictions with potentially lower confidence. Higher values are more selective.

Common Input Format (All Algorithms)

msg.payload.text (string | array)

The input text to classify. Can be a single string or an array of strings.

Example: "This product exceeded my expectations! Highly recommend it to everyone."

Output by Model Type

Low Infra - Good Accuracy - v1 / Low Infra - Good Accuracy - v2 (Transformer-based)

msg.payload contains an output array of predicted class labels.

Example: {"output": ["positive_sentiment"]} or {"output": ["tech", "product_review"]}

Low Infra - Low Accuracy (Deprecate) (FastText-based)

msg.payload is an object with class labels and confidence scores (no output wrapper).

Example: {"predicted_class": ["positive_sentiment"], "confidence": [0.91]}

Example

Input (msg.payload)

{
  "text": "This product exceeded my expectations. Highly recommend it."
}

Output (msg.payload)

{
  "output": ["positive_sentiment"]
}

Errors

When the block fails, it raises an error. Use a Catch block in your flow to handle failures and inspect the error payload.

Common mistakes

• Empty text: msg.payload.text is required and must be a non-empty string.
• Threshold out of range: Threshold must be between 0 and 1.
• No predictions above threshold: Lower the threshold if you want more results.

Best Practices

• Use clean, well-formatted text for better classification accuracy
• Start with threshold 0.5 and adjust based on your precision/recall requirements
• Lower threshold (0.3-0.4) for better coverage, higher (0.6-0.8) for more confident predictions
• Ensure training data covers all expected text categories with balanced examples
• Use domain-specific models when working with specialized text (technical, medical, legal, etc.)
• Regularly retrain models as language patterns and classification categories evolve
• Always validate classification results in production applications