Discourse Graph Discotech
Join us for a two-day DiscoTech - an interdisciplinary hackathon - focused on making research more collaborative and composable.
The Problem:
Making sense of the state of the art in scientific literature (let alone contributing to it) is a cumbersome and faulty exercise. Most of the world’s domain knowledge is scattered across the web as PDFs, living who knows where and containing who knows what. How can we better “read the literature” and connect the dots between ideas?
Today, we mostly accomplish this by building citation trees and mapping public domain articles by topical similarity. But imagine the progress we’d make if we could easily traverse a map of smaller units than the papers themselves - a map of the details we actually want to connect. Instead of a map of article citations, we envision a world where researchers can traverse the claims and supporting evidence made across those articles.
Discourse graphs are the ontology we are describing here. Ultimately, a discourse graph is a series of links between these much smaller units of the research process - questions, claims, evidence, and sources.
With the help of embedding models, large language models, and natural language interfaces, we are setting out to build tools that create these graphs and help researchers engage with them…so that we can all get more out of the literature!
The Details:
On November 8th, 9th, and 10th, we’ll meet at the Fractal Tech Hub and begin scheming. We’ll divide the event in two parts:
November 8th @6 pm: Friday night salon - “What Are Discourse Graphs?”
Learn what discourse graphs are, what the existing problems are, and what topical spaces they can help address
Meet fellow attendees, group brainstorm, and eat good food.
November 9th (+10th) @10am - whenever!: Saturday hack/researchathon
Find a question and/or topic that speaks to you, build a team and get to scheming!
Throughout the event, we’ll come back together to gain inspiration, build off of each other’s ideas, and align on our mission!
Shoutouts!
Huge shout out to the Fractal Tech Hub for providing the space for us to build together over the three days! We couldn't have come together without them :)
Check out what they're doing!
Resources
One way the discourse graph team has articulated their challenge to make discourse graphs more automated. (Plus a pre-labelled dataset of research papers!)
A project to Automate Metadata for research articles