Entities

60 minutes

  • Understanding Intents vs. Entities

    • Understanding why entities usually denote named nouns or the subject of intention
    • Understanding why intents are typically centered around verbs and descriptions
    • Understanding why if two intents differ only in a noun, consolidating into one intent with actions distinguished by entities is advisable
    • Exploring how entities are beneficial for slot filling to fulfill APIs or conveying information in messages

  • Annotating entities: Where and how

    • Beginning the workflow with intents and employing annotate mode before jumping into entities
    • Utilizing the discovery by elimination approach
    • Drawing parallels between "find similar variations" and "show similar to stash" in the intent flow
    • The improtance of focusing on one entity type at a time
    • Exploring the options to suggest entities or create a new one
    • Creating annotations as synonyms or key values

  • Managing Entities on the Entities pane

    • Merging and consolidating key values and synonyms to form a final entity
    • Paradigms of "annotating everywhere" or "only where necessary"
    • Significance of negative annotation like "May I have an appointment with Dr. May in May, please?"
    • The importance of comprehensive and consistent annotation
    • For external NLUs adopting "only where needed" annotation paradigms (DialogFlow/RASA)
    • Locating annotations with "annotate everywhere" and using intent groupings for displaying annotations
    • Utilizing allowed and blocked settings to facilitate "only where needed" annotation
    • Note: There's a limit of 200 annotations per cycle; iterate until completion for large datasets

  • Demonstrating Entity Testing

    • Examining entity results from a test run
    • Accessing phrases.csv and understanding labeled and detected entities per utterance
    • Reviewing Pass, IntentPass, and Entity Fail; their significance, especially in the context of entities within intents for many external NLUs
    • Understanding why HumanFirst NLU employs entirely separate intent and entity processing, ensuring consistent and useful results for testing annotation consistency
    • Comparing with the target external NLU to evaluate entity detection accuracy