Skip to main content

Documentation Index

Fetch the complete documentation index at: https://koreai.mintlify.app/llms.txt

Use this file to discover all available pages before exploring further.

Back to NLP Topics Traits identify specific characteristics in user utterances—before intent recognition—and use them to drive intent detection, dialog transitions, and entity population. Example: The utterance “My flight is late and I will miss my connection” expresses the traits flight delay and emergency. The emergency trait routes the conversation to a human agent without requiring a direct intent match.

​
What Traits Enable

  • Indirect entity extraction — infer entity values from context (e.g., “seat in front with extra legroom” implies First Class).
  • Rule-based intent recognition — trait rule matches are treated as definitive matches.
  • Keyword/phrase identification — match traits to specific words and synonyms.
  • Inference from sentence context — detect traits from overall meaning, not just keywords.

​
Trait Definition

Go to Natural Language > Training > Traits.

​
Trait Type

A Trait Type is a collection of related traits (e.g., Flight Fare containing Economy, Business, First Class).
OptionDescription
ML-BasedTrained with words, phrases, or utterances. One trait per type is detected.
Pattern-BasedTrained with patterns. Multiple traits can be detected; ordering determines priority.
Network type (ML-Based):
  • Standard — n-gram model (configurable up to 5-gram).
  • Few-shot Model — embedding-based. Options: BGE M3 and Pretrained MPNet.
N-gram configuration:
  • n-gram: set max sequence length (1-5; default 1).
  • skip-gram: set sequence length (2-4) and max skip distance (1-3).
  • Trait names must be unique within a Trait Type.
  • Traits with the same name can exist across multiple types, but are difficult to distinguish in rules/detection results.
  • Only spaces and underscores are allowed as special characters in trait names.
  • For multilingual assistants, add language-specific traits.
  • When a trait name is modified, manually update all rules that reference it — the platform does not do this automatically.

​
Trait Association Rules

​
Dialog Execution

Link traits to intents using rules. A trait rule match is treated as a definitive intent match. Add rules from:
  • Traits section — click Add New Rule.
  • Intent Node — NLP Properties > Rules section.
  • Natural Language > Training > Intents — click the Rules tab for an intent.
Each rule can have one or more conditions using AND. Multiple rules per intent are OR-ed — the intent matches if any one rule matches.

​
Knowledge Graph Intents

Associate traits with KG terms:
  1. Go to Knowledge AI > FAQs > Manage KG.
  2. Click the Settings (gear icon) on a node.
  3. Assign the trait. One trait per node.

​
Trait Detection

  • One trait per Trait Type is detected (for ML-based types).
  • Multiple traits can be detected for pattern-based types.
  • Detected traits are added to the context object: context.traits (array of detected trait names).
Use context.traits in:
  • Intent identification rules.
  • Dialog transition conditions (contains operator).
  • Entity population logic.
  • App definitions.
Trait detection data is included in Batch Testing reports and the Find Intent API.

​
Dialog Transition Using Traits

Use context.traits in Connection Rules (under dialog Connection > Properties): 
context.traits contains "economy"

​
Use Case Example

Scenario: Book a flight based on cost preference.
  1. Create Trait Type Flight Fare with trait Economy trained on utterances like “low cost”, “budget”, “cheapest”.
  2. Add a rule for the Book Flight intent: trigger when Economy trait is detected.
  3. Add connection conditions in the dialog: branch based on context.traits contains "economy".
Result: “I am looking for a low-cost option to London” triggers the Book Flight intent and routes to the economy booking flow.