All Projects → xiaoxiong74 → Rasa_chatbot

xiaoxiong74 / Rasa_chatbot

基于rasa_nlu,rasa_core,rasa_core_sdk构建的聊天机器人

Programming Languages

python
139335 projects - #7 most used programming language

Projects that are alternatives of or similar to Rasa chatbot

Xatkit
The simplest way to build all types of smart chatbots and digital assistants
Stars: ✭ 97 (+2.11%)
Mutual labels:  bot, chatbot-framework
Hangouts Chat Samples
Chat Bot Samples for Google Chat.
Stars: ✭ 284 (+198.95%)
Mutual labels:  bot, chatbot-framework
Flottbot
A chatbot framework written in Go. All configurations are made in YAML files, or inside scripts written in your favorite language.
Stars: ✭ 175 (+84.21%)
Mutual labels:  bot, chatbot-framework
Botfuel Dialog
Botfuel SDK to build highly conversational chatbots
Stars: ✭ 96 (+1.05%)
Mutual labels:  bot, chatbot-framework
Botman
A framework agnostic PHP library to build chat bots
Stars: ✭ 5,538 (+5729.47%)
Mutual labels:  bot, chatbot-framework
Rasa core
Rasa Core is now part of the Rasa repo: An open source machine learning framework to automate text-and voice-based conversations
Stars: ✭ 2,302 (+2323.16%)
Mutual labels:  bot, chatbot-framework
Rasatalk
A chatbot framework for Rasa NLU
Stars: ✭ 225 (+136.84%)
Mutual labels:  bot, chatbot-framework
Poshbot
Powershell-based bot framework
Stars: ✭ 410 (+331.58%)
Mutual labels:  bot, chatbot-framework
Stealth
An open source Ruby framework for text and voice chatbots. 🤖
Stars: ✭ 481 (+406.32%)
Mutual labels:  bot, chatbot-framework
Joe
A general-purpose bot library inspired by Hubot but written in Go. 🤖
Stars: ✭ 417 (+338.95%)
Mutual labels:  bot, chatbot-framework
Botpress
🤖 Dev tools to reliably understand text and automate conversations. Built-in NLU. Connect & deploy on any messaging channel (Slack, MS Teams, website, Telegram, etc).
Stars: ✭ 9,486 (+9885.26%)
Mutual labels:  bot, chatbot-framework
Bootbot
Facebook Messenger Bot Framework for Node.js
Stars: ✭ 886 (+832.63%)
Mutual labels:  bot, chatbot-framework
Fb Botmill
A Java framework for building bots on Facebook's Messenger Platform.
Stars: ✭ 67 (-29.47%)
Mutual labels:  bot, chatbot-framework
Go Tgbot
Golang telegram bot API wrapper, session-based router and middleware
Stars: ✭ 90 (-5.26%)
Mutual labels:  bot
Coinboss
⚡️💸💰 A Serverless Coinbase tradingbot.
Stars: ✭ 92 (-3.16%)
Mutual labels:  bot
Node Sdk
An official module for interacting with the top.gg API
Stars: ✭ 90 (-5.26%)
Mutual labels:  bot
Xenon
Create backups of your discord server
Stars: ✭ 90 (-5.26%)
Mutual labels:  bot
Wow Fish Bot
World of Warcraft ( WoW ) Fish BOT. Python. Simple. For me.
Stars: ✭ 93 (-2.11%)
Mutual labels:  bot
Skyra
All-in-one multipurpose Discord Bot designed to carry out most of your server's needs with great performance and stability.
Stars: ✭ 92 (-3.16%)
Mutual labels:  bot
Wasapbot
[abandoned ❗] Simple WhatsApp bot written in PHP, respond to private & group messages. Uses Chat-API
Stars: ✭ 89 (-6.32%)
Mutual labels:  bot

Rasa Core and Rasa NLU

Introduction

这个聊天机器人demo是用开源NLU框架rasa-nlu完成意图识别与实体识别,用rasa-core完成对话管理和与对话生成。

  • 本demo完成的对话主要有:
  • 1: 办理套餐、查询话费和流量(会话场景1)
  • 2:案件查询(会话场景2)
  • 3:Q&A问答+闲聊(合并在unknow_intent的场景里)
  • 本demo实现流程

Rasa-Chatbot

  • demo主要参考了
  • 主要包版本
python:        3.6.8
rasa-nlu:      0.14.4
rasa-core:     0.13.2
rasa-core-sdk: 0.12.1
tensorflow     1.12.0
keras          2.2.4
  • 主要文件描述
  • data/rasa_dataset_training.json :nlu训练数据
  • configs/_config.yml 类文件:模型流程定义(language、pipeline等)。nlu_model_config.yml中的pipeline可自定义,这里由于数据量较少,用了开源的方法和词向量(total_word_feature_extractor.dat)。如果你的rasa_dataset_training.json上数据足够多,可以尝试使用nlu_embedding_config.yml(本demo使用)配置来训练nlu model.
  • mobile_domain.yml :各组件、动作的定义集合,其实就是特征
  • endpoint.yml 服务地址、会话存储地址(url)
  • data/mobile_edit_story.md :定义各种对话场景,会话流训练数据
  • bot.py :各种训练nul与 dialogue的方法
  • actions.py :负责执行自定义 Action (通常都是具体的业务动作,在本项目中通信业务查询、案件查询、闲聊或Q&A)
  • data/total_word_feature_extractor.dat : 一个训练好的中文特征数据(使用nlu_moel_config.yml配置训练时会用到)
  • data/news_12g_baidubaike_20g_novel_90g_embedding_64.bin :训练好的word2vec模型(train_nlu_wordvector:wordvector_config.yml中用到),可下载更大的训练好的模型,下载地址:连接 密码:9aza

Command

train nlu model 训练NLU模型(可选择其他的,如train-nlu-wordvector)

python bot.py train-nlu

test nlu model 测试NLU模型,主要是看意图是否识别准确,是否抽取到实体

python -m rasa_nlu.server --path models/nlu   启动NUL模型服务

curl -XPOST 192.168.109.232:5000/parse -d '{"q":"我要查昨天下午的抢劫案", "project": "default", "model": "current"}'   

train dialogue 训练会话流程(可选择其他的,如train-nlu-transformer)

python bot.py train-dialogue-keras

test dialogue -client端测试对话流程(开启core client服务)

python -m rasa_core_sdk.endpoint --actions actions &

python -m rasa_core.run --nlu default/current --core models/dialogue_keras --endpoints endpoints.yml     

dialogue 交互式训练生成新的story(相当于自己构造对话场景数据。新的story可以append到之前训练使用的story中重新训练,重复此过程)

python -m rasa_core.train interactive -o models/dialogue_keras -d mobile_domain.yml -s data/mobile_edit_story.md --endpoints endpoints.yml  重头开始训练story,零启动
python -m rasa_core.train interactive --core models/dialogue_keras  --nlu default/current --endpoints endpoints.yml  通过已有story模型训练(构造更多的story,一般用这种方法)

provide dialogue service -Service端:提供对话服务接口(channel(如web)接入时开启此服务)

python -m rasa_core_sdk.endpoint --actions actions &

python -m rasa_core.run --nlu default/current --core models/dialogue_keras --credentials credentials.yml --endpoints endpoints.yml  开启core服务(Service) 

compare policy

python -m rasa_core.train compare -c keras_policy.yml embed_policy.yml -d mobile_domain.yml -s data/mobile_edit_story.md -o comparison_models/ --runs 3 --percentages 0 25 50 70

evaluate policy

python -m rasa_core.evaluate compare -s data/mobile_edit_story.md --core comparison_models/ -o comparison_results/

Some tips

批量生产nlu训练数据

训练数据的构造是非常费时的一件事,本demo data/rasa_dataset_training.json 是通过一些规则自动生成的,节省很多人力。

UI界面接入

UI界面接入可参考 https://github.com/howl-anderson/WeatherBot_UI 直接更改相应的端口或ip即可使用。

  • 启动方法:
  • 1、启动NLU服务
  • 2、启动dialogue service
  • 3、启动web服务

多看官方文档 rasa_nlurasa_core

其中也有些坑,使用期间有任何问题,欢迎随时issue!

Q&A

ner_duckling 无法使用

从rasa_nlu=0.14.0 开始就不使用ner_duckling,详见changelog,仅保留ner_duckling_http。因自己启动ner_duckling_http 报错,故自己把ner_duckling的模块又重新添加到了rasa_nlu中。添加方法如下:

  • 导入方法: from rasa_nlu.extractors.duckling_extractor import DucklingExtractor
  • 添加组件: 在组件列表component_classes 中加入 DucklingExtractor

train_dialogue_transformer训练报维度不匹配错误

在policy/attention_keras 中要求输入的特征是偶数个,即mobile_domain.yml的特征数据量,若报错删除一个或增加一个特征即可

train_nlu_wordvector报编码错误

因为rasa_nlu_gao中的word2vec模型使用的txt文本模型,我这里用的bin二进制模型,所以如果使用bin的二进制模型需要更改 rasa_nlu_gao中的源码。修改方法:

  • 1、定位到site-packages/rasa_nlu_gao/featurizers/intent_featurizer_wordvector.py
  • 2、定位到两处模型加载的地方 model = gensim.models.KeyedVectors.load_word2vec_format 将里面的binary 改为True即可

Some magical functions

rasa-nlu-gao新增了N多个个自定义组件,具体用法和说明请参考该作者的 rasa对话系统踩坑记,个人觉得对新入坑聊天机器人的童鞋很有帮助,感谢作者的贡献。简单使用方法如下:

首先需要下载rasa-nlu-gao

pip install rasa-nlu-gao

训练模型

python bot.py train-nlu-gao

测试使用模型

python -m rasa_nlu_gao.server -c config_embedding_bilstm.yml --path models/nlu_gao/
Note that the project description data, including the texts, logos, images, and/or trademarks, for each open source project belongs to its rightful owner. If you wish to add or remove any projects, please contact us at [email protected].