Things necessary for Dr. Ventura-style:

• Store knowledge about the domain to be used in creating a new thing 
	○ web scrape those summary wikia pages 
	○ desired format: JSON would be great. Easy to deal with in java, easy to deal with in web scraping, 
• create a new example of the thing 
	○ ??????? 
• test whether it is a good thing
	○ Prolog engine?? 
	○ have a set of rules it needs to follow to be valid? 
• translate the computer version of it into something humans understand 
• test whether the human-understandable version is also good 

Idea: have the human person help introduce novel content into the thing

• human comes up with broad topic
• thing comes up with questions to help it find related concepts, pop culture ideas, solve connection problems it can't figure out on its own

Thought: make every (or almost every) into JSON things, give buckets of elements that could work sort each element of the plot into types - info about the thing

• example: phineas and ferb make a thing, it ends up being used somewhere else (little duffers)
• still exists, but is taken offscreen (zapped to another dimension)
• Mom and Dad don't mind it / it's not in the way (molecular transporter, flying car of the future … today!, 

I spent some time trying to figure out how we're going to do the overall structure. Still not sure about that, but I have a rough plan for the knowledge base part and an even rougher/vaguer plan for the actual algorithm.

As far as storing the knowledge base about episodes, I think I'd like to store that stuff as JSON. This'd be easy to integrate with a web scraper, and would play nicely with whatever Java code you wanted to do. It'd also be conceptually easier for me to work with than a SQL database.

I like the part about Mexica where it has ways to solve a problem, and then based off of the existing stories it can go through those and find a way to solve it. I think I could do something like that with the existing episodes, and find patterns in the key elements like how Doof is going to take over (there's a subset of plans that involve gathering some kind of unconventional minions, another one that involves embarassing his brother Roger or otherwise removing him from power, etc) and how Phineas and Ferb's thing goes away (it gets destroyed somehow, it still exists but is out of the picture, the idea was inherently temporary, someone else with a good use for it ends up with it, etc). The AI could base off of those to find alternative solutions to the main problems of the episode.

As far as the web scraping, I think I'll make some of the JSON things manually, and then get a thing that mostly gets stuff automatically but requires me to interpret text sometimes so it doesn't take so long to get a good knowledge base. More of the cobbling things together approach where we get code to do as much as it can, but use human input for pieces of it when necessary

How do I evaluate coherence? How do I evaluate coolness of the thing? How do I put value on the intersections between the plotlines?

Evaluate:

• novelty
	○ can it come up with something outside of what's already there?
	○ new to me, who has seen every episode
	○ new in general - not just rehashing existing episodes
	○ new structure - like the caveman ones, or when they decide to break part of the script
• value
	○ for us to decide later
• intentionality
	○ goal is to have a coherent sequence of interrelated events following the style of a typical phineas and ferb episode
mind/storygeneration/kendra/mon-16-oct.txt · Last modified: 2017/10/16 22:56 by kendradg
Back to top
CC Attribution-Share Alike 4.0 International
chimeric.de = chi`s home Valid CSS Driven by DokuWiki do yourself a favour and use a real browser - get firefox!! Recent changes RSS feed Valid XHTML 1.0