This repo [UNDER CONSTRUCTION] contains computational analyses of and analytical-creative approaches to the

MARGENTO community-based interplatform events #GraphPoem

held at the Digital Humanities Summer Institute (DHSI) since 2017 (as well as other venues since 2001).

The events involve data-sharing, interactive coding, and livestreaming simultaneously across platforms such as JupyterHub (for coding), Twitter (botting, replaced with Facebook since 2025), and Facebook (posting/SM-interaction & livestreaming).

The conglomerate of interplatform SM users/coders, AI (and other ML agents and algorithms), and (poetry) data, is seen as one hybrid-network variably evolving community--the #GraphPoem community.

Studying this community as a dynamical system occasioned making an argument in favor of their radical potential by means of interplatform complexity and creativity. The "complexity of resistance" and "complexity of assertion" in such communities escape hegemonic control and extractivism since they are impossible to learn computationally (Tanasescu 2024 https://www.taylorfrancis.com/chapters/edit/10.4324/9781003320838-3/dynamical-systems-interplatform-intermediality-chris-tanasescu?context=ubx&refId=a352b9ec-5706-43ce-84c4-827290868465).

This repo illustrates that with a case-in-point, #GraphPoem@DHSI23, by developing a mathematical learner--an infinite/continuum-depth graph neural ordinary differential equation (GNODE or GDE) network (see https://arxiv.org/abs/1911.07532)--which tries to simulate the evolution of the community during such a specific event.

The learner is fed the (multiple) platform logs and the development of the poetry corpus (as particpants co-contribute to it and the code analyzes and then passes outputs on to the bots) and predicts the community across windows of continuous time. Although it performs adequately when evolution remains near-linear, the error margin surges when a rupture occurs: an evental spike (Badiou) or singularity (Agamben).