bulbul's Introduction

Welcome to BulBul

A lightweight & powerful skill bot with multi end*point support.

Commands

pip install *r requirements.dev.txt * Install dependents.
sh ./scripts/docs*live.sh * Star a live document.
sh ./scripts/lint.sh * Run lint.

Concepts

Here are the basic concepts, all pre-defined models saved in file bulbul/builtins.py.

NLU
- PreProcessor: Consume a Query object and return a Query object with processed text. Here is same examples:
  - remove while space.
  - convert emoji to token.
  - custom PreProcessor
- IntentTagger: Consume a Query object and might return a list of NLUResult, each NLUResult contains a Intent.
- JointTagger: Consume a Query object and might return a list of NLUResult, each NLUResult contains a Intent and a Slot.
- SlotTagger: Consume a NLUResult object (which contain Intent object) and return a NLUResult (might contain a Slot obejct).
Skill: Which consume a NLUResult object and return a optional SkillResponse
Plugin
- RuleBasedPlugin Which defines nlu rules and skill handler function in a single class.
Adapter: Adapter will consume requests from different backend and convert it to a unified request for BulBul, and convert bulbuls response to json format required by different backends.

Todo

bulbul's People

Contributors

Stargazers

Watchers

bulbul's Issues

structure of the bulbul

I think this project should support tranditional and deeplearning methods. Only new things can attract more developers. There are some component I need to describe.

Pre: Means that data processor before nlu. This process is recognized as necessary.
NLU：This process is the core component of the project. In this Module, We should have the same interface for different methods, eg: Rulebased, machine-learning-based, deeplearning-based. They should implement all of the interface, so, different methods/tools can work as the same.
Plugins: I don't have clear idea about it. But, I think the plugins is the pipeline mode.
NLG: this process, We should output the result according to the input. So, like NLU, this process should have the saome interface that support different methods. Such as Answer Selector, QuestionMatchBasedSelector, End2End, KnowledgeGraphBased, and so on.
Chat, This is finnal design of the project. We should complete all of the above idea, so this process can be fine to integrate various chat channel.

Recommend Projects