Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Introduction

Motivation and Summary

Networking: This work addresses a novel distributed chat bot network methodology architecture

Mobility: Chat bots are implemented by Mobile Agents (Mobile Software)

Interaction: Chat bots performing dialogues by using a script and knowledge data base

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Objectives

1. From human-machine to human-network communication. Next generation Chat bots improve social networking;

2. Filling the reality gap. Hypothesis: A group of cooperating Chat bots can improve quality of dialogue flows significantly;

3. Scalability. Distributed Chat bot networks implemented with self-organising agents can extend knowledge bases and fit to large-scale environments like Smart City ⇒ Crowd Control.

4. Sensors. Chat bots are not limited to conversation only; they are sensor data sources for analytics, too!

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

State of the Art - Where we are

• Today human-machine dialogues performed and moderated by chat bots are ubiquitous.

Chat bots are now integrated in popular messaging programs and also appear as stand-alone services like Amazon Alexa, Microsoft’s Cortana, and Apples Siri. Central part of dialogue robots is Natural Language Processing (NLP) and Knowledge Logic.

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

History

History of Dialogue Robots

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

The Beginning

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Next Generation Bots

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

• Commonly, centralised and server-based chat bot software is used to implement rule-based and intelligent dialogue robots.

  • Human networking is not supported.

• Many Chat bots are still simple finite-state machines with limited knowledge bases, language understanding, and capability to construct meaningful dialogue flows!

  • Missing user personalisation
  • No user feedback propagation and adaptation

• Interaction Interfaces:

  • ⊕ Human-Bot Interaction
  • ⊕ Machine-Bot Interaction
  • ⊖ Bot-Bot Interaction
  • ⊖ Human-Network (Human-Bot-Human) Interaction

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Taxonomy of Chat Bot and Classes

Human-Machine Interface

QA Bots

Question-Answer Bots for a limited topic field, basically a natural language interface to FAQ scripts → Information Sources

Conversational Bots

More general conversational bots with free form text input and output → Entertainment
Examples: Eliza!, jabberwacky, Cleverbot, Alexa (really a CB?), Osiris (really a CB?)

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Intelligent and Learning Bots

Either QA or conversational bots (or a mix) that grow and adapt with increasing dialogues.
Intelligent chat bots aim to improve dialogues and conversational quality over time and user experience.

Personalised Bots

Either QA or conversational bots (or a mix) that adapt to specific persons (remembering history, skills, likes) to maximize the utlity of the conversation for the user.

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Dialogue Processor

Static Bots

The dialogue processor uses a simple question-answer table (finite-state machine)

Dynamic Bots

The dialogue processor uses an extensible and modifyable dialogue data base

Rule-based Bots

Conversational rules provide the base for reactive chat bots → Symbolic agents with outside-in approach

Intelligent Bots

More natural dynamic and reactive bots based on sub-symbolic learning agents with inside-out approach

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Concept and Architecture

Towards Smart City Environments and Self-☆ Event Management

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Chat Bot as Mobile Agents

• A Chat bot is implemented and represented by a mobile software agent

The agent poses a state-based reactive agent behaviour that is programmed in JavaScript and processed by the JavaScript Agent Machine (JAM)

• The agent performs:
  • Dialogue moderation, processing, and synthesis;
  • Data base management;
  • Knowledge discovery and Networking; and
  • Learning and Planning on a higher level.

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Chat Bot Agent and Platform

Connected Bots (JAM: JavaScript Agent Machine platform, NLP: Natural Language Processing)

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Dialogue Data Base

• The dialogue data base consists of active (functional) dialogue snippet rows

Dialogue Snippet

• There are three classes of dialogues:
  (1) Reactive (2) Pro-active (3) Survey

• A snippet is a functional and state-based data structure capable of

  • executing DB queries and NLP,
  • requesting sensors, contacting external services,
  • modifying the snippet (adding data ..).

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Snippet Format

type dialog database = dialog snippet []
type dialog snippet = {
  tag       : string,
  phrases?  : string [], topics? : string [], 
              keywords? : string [],
  stimulus  : { activation? : function,  
                message? : string|function, 
                question? : string|function, 
                choices? : [], range? : []},
  process?  : function, 
  action    : { pre? : function, 
                response? : function, },
  answer?   : [],     data?     : [],
}

• The snippet functions can perform NLP supplied by access to the compromise framework
• Sensors (location, external WEB APIs) can be accessed, too.

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Chat Bot Agent Model

Agent behaviour model with prominent activities

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Chat Bot Communication

• Networking is performed via:

  • WEB APIs (e.g., weather or map services, wiki queries, ..);
  • Bot-Bot communication to extend information range;
  • Bot-Other Bot communication (e.g., with CleverBot) to extend conversational capabilities.

• Bot-bot communication uses messaging:

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Demonstration

DB: Only 20 dialogue snippets, size ≈ 10kB!

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Results

• A small dialogue data base (< 50 entries, < 20kB) is sufficient to handle common dialogue flows;

  • Missing dialogue matches are compensated by remote services and by requesting other Chat bot agents (human-bot-human interaction)

• Chat bot mobility and communication is lightweight and fast:

  • Migration: Code+Data (DB) < 100kB, latency < 1s
  • Communication: Messages between Chat bot agents require < 1kB (a dialogue snippet), latency < 100ms

• Scaling with respect to communication partners is almost linear

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Conclusion

The fusion of Chat bots technologies with multi-agent systems enables the orchestration and connection of dynamic large-scale chat bot networks

• Capable to interact with users dynamically either directly or indirectly via bot messaging.

• Both specialisation by hierarchical bot networks as well as cooperation is supported (knowledge extension).

• Spatially separated users can be connected and profit from extending the information range via chat bots.

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

The slim agent processing platform can be easily integrated in existing software or WEB pages, especially supporting mobile networks and devices.

• Low resource requirements!

• Dialogue data bases owned by a Chat bot agent consists of active dialogue snippets (can be exchanged)

• The agent bot communication and interaction enables distributed knowledge and dialogue data bases.

One main issue of the simplified NLP and script-snippet data base approach is the quality of dialogues and the acceptance of users that must be improved in future work.

Stefan Bosse: Distributed Serverless Chat Bot Networks using mobile Agents

Thank You!

Thank you for your attention. All questions are welcome!

• Further information can be found here:
  • http://ag-0.de
  • http://sblab.de