Discussion Forum

Discussion bot for K12

I am a newbie to Grakn, although I’m stepping through the tutorials for Python clients as I write this post. I’m looking for sounding boards to provide feedback on an idea I have for an automated K12 discussion moderator. The design is a bit different from most knowledgebases because the knowledgebase purposely starts out with less than 100% of the facts. I wouldn’t want a student to be able to get an answer to the key question too soon and short circuit the discussion. The knowledgebase progressively gets more facts as time passes and students gain insights. If you want more information, I’ve started organizing my thoughts in a Google Slides deck at http://bit.ly/38Ksd5P

Hi - you can definitely use Grakn as a knowledge base for Q&A, assuming you can match your user input to the correct Graql query! We have some users of Grakn who are attempting to convert English and other language formats to Graql, so they can extract information. Perhaps some of the sessions at our conference this week (grakncosmos.com) can help a little bit.

You could also encode in rules what the next question to ask is, given all the knowledge that has been acquired. You could also model the “ideal” full KB as one set of data in the database, and have a running growing set of data based on collected data. The rules could try to infer what to ask next based on the delta - this sounds like a relatively big task though!

@joshua,
Thanks for your thoughts and I’ll plan on attending GraknCosmos in 2021.

I was thinking of a KB that was a growing set of data that could be added from the “ideal KB”, as well as answers to questions posed to students and statements made by the teacher. The questions asked could come from a “progression list” that is how most discussions progress about a particular topic. I’m also thinking of monitoring what areas of the KB are getting queried so the questions can steer students towards ignored parts of the KB.

I’m don’t think I will have to use NL to convert the user queries into Graql. That is too much complexity for my passion project. Instead, I’m hopeful that relatively narrow discussion topics will make the question set manageable. For example, there is a finite number of important discussion questions or prompts for the causes of World War 1.