In plain words: A system reads text-game descriptions and outputs simple graph edits—adding or removing facts—so a knowledge map stays current as the story changes. Instead of rebuilding the whole graph each time, it was tested on over 300,000 game transitions collected for this purpose.

Abstract · Building Dynamic Knowledge Graphs from Text-based Games

We are interested in learning how to update Knowledge Graphs (KG) from text. In this preliminary work, we propose a novel Sequence-to-Sequence (Seq2Seq) architecture to generate elementary KG operations. Furthermore, we introduce a new dataset for KG extraction built upon text-based game transitions (over 300k data points). We conduct experiments and discuss the results.

Mikuláš Zelinka, Xingdi Yuan, Marc-Alexandre Côté, Romain Laroche, Adam Trischler
arXiv:1910.09532 · cs.CL, cs.LG · submitted Oct 21, 2019 · updated Jan 23, 2020
abstract · pdf · html · NeurIPS 2019, Graph Representation Learning (GRL) Workshop