McKean's Research Journal
Meeting on 3/15
Use Markov chain with a set of buckets that are determined by the goals/location/stage changes of the characters, and classify each event into those changes.
Hierarchical model might be a benefit later on, but it would be better to start with a flat set of buckets (around 40).
Events should exist in only one bucket to help the model learn appropriate transitions. If the events exist in multiple buckets, you might get more creative transitions.
Week of 2/12
Research Meeting: 1hr. Went over how to present the structure of the story graphically (for the website and otherwise)
Searched for frameworks to make representing structure easier (2hr)
reading/training on JQuery and JS (7 hr)
Built JQuery/JS files to pull data from the server
Week of 2/5
Research Meeting (1hr)
Built
HTML and
CSS structure for the website including dummy links for layout and a form for submission (7 hr)
Read up on
HTML and
CSS Grid stuff (3hr)
Week of 1/29
Week of 1/22
Week of 1/16
-Presentation and Lit Review: 4 hours
-Research Presentation Meeting: 1hr
Week of 1/7
-Presentation and Lit Review: 5 hours
Christmas Break
-Presentation and Literature Review: 17 hours
Week of 12/4
-Research Meeting (12/7) (1h) - Talked through how to use the helper, and how to design our helper website
TODO:
Week of 11/27
-Research Meeting (11/30) (1h) - Talked through ways to speed up Episode database generation.
TODO:
-Literature Review composed, drafted, edited, revised (30 hours).
Week of 11/13
-Meeting with Kendra (11/14) (1h) - worked out details of model classes. Worked on rudimentary generator
-Worked through problems of model generalization and problems of story generation (1h). Read articles and condensed info (1h).
Week of 11/6
-Research Meeting (11/9) (1h)
Takeaways: for Story Generation
Use back- and forward-chaining (like Talespin), but with checkpoints
Later: govern behavior of characters/events with probabilistic rules
Later: Keep list of pending goals (ordered and including author goals) that guides chaining/pruning. Use this to correct global structure.
TODO:
() Initially just store the events from the episodes using the existing data structure.
() Later, develop a system that will generalize these events, allowing them to have more versatility and chaining power.
- Lots of research reading. See summary page for details (5h)
- Writing proposal for literature review (3h)
Week of 10/30
-Research Meeting (11/2) (1h)
-Read research articles: “Narrative Models: Narratology…”, “Story Generator Algorithms, Gervas”, “Models and Approaches to Generation…” (4h)
Week of 10/23
(Mostly worked on ORCA grant application (4h) and reading research articles (4h))
Week of 10/16
10/19 - Research Meeting and work on web scraping (2h)
Takeaways:
Week of 10/9:
10/12 - Research Meeting (1h) and read research articles (2h)
Takeaways:
In order to truly have novelty, you need to pull from another source, which would make it open world, requiring contextual information to be able to reason about this new idea.
We need the flexibility of an open-world system, but maintain the enclosed, resolving, structure of a closed-world system.
Need to know the architecture of how it's going to work before starting to code.
Semi-automated system in which a subject from the open world is incorporated to the closed system with its context. The system will ask the user about contextual elements of the subject matter (i.e. “who is the bad guy?”, “What are the attributes of the subject matter?” “What is it related to?”)
10/10 - Met with Kendra and discussed graphs, research articles and web scraping. (1h)
Takeaways:
TODO:
Week of 9/25:
9/28 - Found research articles and linked them to the page (.5h) and Research meeting (1h)
TODO:
Takeaways:
Potentially make an algorithm that combines primary characters into a subplot and mixes their primary motives as an element of the subplot.
Use lucidchart to help organize and generalize the flow of episodes.
Creativity is about knowing which rules to break (or not). So only focus on the formulaic episodes in order to find that formula.
9/26 - Met with Kendra to discuss plot (1h)
Takeaways:
We need a framework that is loose enough to cover all episodes, but is as strict as possible (so that it actually provides utility).
At its core, an episode is a bunch of subplots, each with their own characters, that develop individually, then interweave, and then resolve and conclude.
Each character has general motives/patterns that are consistent throughout the series, but also has temporary/specific motives that are more confined to a particular episode.
Another element (which we are going to leave for later) includes running jokes between episodes, or smaller easter-eggs that aren't critical to the plot.
TODO's:
(x) Link research articles to the resources page. (Look up the ones that Dr. V gave us)
( ) Look at representative episodes and compare to the graphs.
( ) Try to systematically produce an episode by hand (-chicken festival; potatoes)
Back to top