The Data Science & AI Lab is developing a Q&A model to be deployed on websites and Microsoft Teams to answer various categories of questions from related data sources, e.g. undergraduate and graduate course catalogs, student residential life, admissions, COVID-19 procedures, programs and plans, etc. Users can chat with the bot and get the answers quickly instead of searching.
Scenario: There is much data in the form of FAQs, brochures, instructions, troubleshooting tickets, etc. The progress in AI brought new capacity and services to users.
- Static data on websites, PDFs, handouts, etc. are routined, often not easy and quick to find information.
- Human involvement if information isn’t available or not easy to find
- Budget shortage
Solution: Azure based Question Answering models trained with text data, could be deployed to chatbots on Web pages, MS Teams, Skype, Facebook, Etc. A Question Answering model does not replace associated websites or data sources, but works in parallel, providing an easy way to find information, and may refer to them.
Data: Pairs of Questions & Answers from
- Text data from Frequently Asked Questions (FAQ) on departmental web pages
- Undergraduate/Graduate Catalogs
- Knowledge bases (Confluence)
The data is hosted on Confluence, and read via its Web Service. Confluence pages in a Tree structure are traversed and read and saved, later used for the Questions & Answers extractor, and used to train a language model.
Deployment: The service could be deployed to Web, MS Teams, Facebook Fan Pages, and websites in general.
Result example: Snapshot examples for Microsoft Teams (also web)
If you would like to use it, please go to the sidebar on the left side of MS Teams, click Apps, search for “Virtual Assistant”, add the app, then start chatting with it. You can start with saying hi, then ask any questions related to ITS Confluence (https://confluence.uconn.edu/ikb/) as the training data was from it.