Agent based intelligent KM (Knowledge Management) Chatbot (Part 1)

Traditional approach of KM mainly base on a “KM Web Site” which use traditional approach to search and browse information, and with content of “KM café” and “Hubs” from different operation units, this approach has limitation of “information collection”, “maintenance”, and “knowledge sharing”.

This proposal is to build an “Agent based intelligent KM (Knowledge Management) Chatbot”, with a simple Q&A web UI which is similar to search engine or traditional chatbot.

 

Objective

To supplement existing KM (Knowledge Management) Portal by an “Agent based intelligent KM Chatbot”, with a simple UI of Q&A, and with below major characteristics:

l   Front end agent

l   Back-end tools which connect to ChatGPT through OpenAPI (or similar subscription service such as百度)

l   Usage of LLM (Large Language Model)

 

Current business

Exiting approach of KM Portal mainly base on a Web Site which:

l   Use traditional approach to organize, browse and search information, and

l   Provide content of “KM café” and “Hubs” from different business units

This approach has below limitation:

l   Require extensive effort to “collect”, “organize”, and “update” information, and

l   Not efficient in “Knowledge sharing”.

 

Suggested Business Process

1.     Build a knowledge database of particular business domain:

l   Based on Transfer Learning of existing language model

l   Supplement by Q&A templates, with information fill-in by user with related domain knowledge

Based on above two approaches, build interface with LLM model to create knowledge database of the “Intelligent Chatbot”

(**This is the core development of the whole project**)

1.1 Build Database : based on Transfer Learning of existing Language Model

a)     Instead of using template (in format of question and answer) to collect information, automate process of information collection by importing  documents such as “Application User Manual of IT system”, “DI” (Departmental Instruction) etc, to the system.

 

b)     Back-end system will interface with LMM model of ChatGPT (or百度) to extract MetaData (ie: conduct Semantic Analysis, to extract meaningful phases) , build relationship between MetaData and content of documents submitted.

 c)     By using concept of “Transfer Learning”, train and create a knowledge database which is related to particular business domain (eg: an particular IT system, operation of a Business Unit).

 

1.2   Build database : supplement by Q&A templates

Dedicated user from particular business domain (ie: senior user with particular domain knowledge) submit content of “question” and “answer” (in a form of “template”) to back-end system by batch. [Below is a Q&A template which use hospital as an example]

2.     Daily usage

Provide simple web UI for user to query information using regular english syntax and return “answer” or return “list of documents with different score”. [Please refer to below which use a hospital as an example]

 

 

3.   Feedback mechanism for continuous improvement of database

Provide control mechanism for administrator or dedicated knowledge users to fine tune result of answer, by re-submit revised “answer” of a previous “question”, this provide a mechanism of continuous improvement of the knowledge database.

 

4.   Development tools and others

l   The proposed system will be developed by using ASP.NET (for Web Site) , C# or Python (for interfacing with ChatGPT or 百度).

l   Support English first and then support Chinese Language.

Block diagram of “Intelligent KM Chatbot”