Rasa——对话机器人开源框架

Rasa是一套对话机器人的开源框架，主要用于构建面向任务（Task Oriented）的对话系统。

NLU自然语言理解

• 基于规则

• 神经网络方法

  DIET(Dual Intent Entity Transformer)

在Rasa中两种方法都可以使用，但神经网络方法是核心

一个rasa项目包含的文件：

• domain

  包含了聊天机器人需要知道的所有东西

  • Responses: these are the things the assistant can say to users
  • Intents: These are categories of things users say
  • Slots: These are variables remembered over the course of a conversation
  • Entities: these are pieces of information extracted from incoming text
  • Forms and Actions: These add application logic and extend what your assistant can do

• Stories

  training data to teach your assistant what is should do next

• Rules

  a way to describe short pieces of conversations that always go the same way

• config

  • language 指定机器人使用的语言
  • pipeline 定义用于检测意图和实体的步骤
  • policies 指定如何响应用户输入的对话管理技术

Do

• Use actual user conversations as stories
• Have small stories that aren’t full conversations
• Use rules for one-off interactions(checking account balance, checking if this is a bot)

Don’t

• Use rules for multi-turn interactions
• Use OR statements and checkpointing often
• Write out every possible conversation flow start to finish
• Delay user testing

Rasa style CDD:

• You only need to start with the most popular, important intents & a way to handle things outside them
• Continue to build from there if that’s what users need