Wit.ai: how it identifies intent and classifies objects from custom expressions

I have been studying wit.ai for several days. I found the key points of the wit.ai bot engine:

  • A story based - create a story to greet, order pizza, order a laptop, ask for a forecast
  • Role-playing entity - location: form, location: to. Here "from" and "to" are the role of "location"
  • Composite / embedded object - a car (model, color, model). Here the model, color, modelYear can be nested under the car object.
  • Search Strategies: trait, free text, keywords
  • Understanding the bot by creating some stories
  • Coincidence points are called trust.
  • Custom expression length 256 at maximum
  • Search from a predefined list of keywords, expressions to match
  • Nested context
  • User Entities, Predefined Entities
  • Entity-based actions: if only, if always conditions
  • For a given custom expression, wit looks for a match in the keyword list, free-text
  • For a given custom expression, wit searches for keyword positions in the listed expressions below the object
  • Branching for missing information in this user expression
  • ? # 1 : Widget X 2000? : 30 . # 2 : Widget X 2000? Bot: Best Buy. №2 : Widget X 2000? : 30 . : ? ----
  • , ,
  • wit.ai / ? , context.key .
  • Wit.ai? , , ,
  • Wit.ai?

, wit , wit.ai.

- / ML, wit.ai, , . , , .

.

+6
1

wit.ai:

duckling, , .

, , , .

, Rasa

+3

Source: https://habr.com/ru/post/1015747/


All Articles