Learning Knowledge Graph-based World Models of Textual Environments

Authors: Prithviraj Ammanabrolu, Mark Riedl

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental A zero-shot ablation study on neverbefore-seen textual worlds shows that our methodology significantly outperforms existing textual world modeling techniques as well as the importance of each of our contributions.
Researcher Affiliation Academia Prithviraj Ammanabrolu School of Interactive Computing Georgia Institute of Technology raj.ammanabrolu@gatech.edu Mark O. Riedl School of Interactive Computing Georgia Institute of Technology riedl@cc.gatech.edu
Pseudocode No The paper includes architectural diagrams (Figure 2, Figure 3) and mathematical formulations of loss functions, but no explicit pseudocode or algorithm blocks.
Open Source Code No The paper links to the Jericho World Dataset (https://github.com/Jericho World/Jericho World) but does not provide a concrete link or explicit statement about the open-sourcing of their Worldformer model's code.
Open Datasets Yes Dataset. We use the Jericho World Dataset [4].1 1https://github.com/Jericho World/Jericho World
Dataset Splits No The paper mentions 'training data' and a 'test set' with specific instance counts, but does not explicitly state the existence or size of a validation split or how it was created.
Hardware Specification No The paper does not specify any hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions various software components and models like BERT, GPT-2, ALBERT, Open IE, and Word Net, but does not provide specific version numbers for any of them.
Experiment Setup Yes All sequence models use a fixed graph vocabulary of size 7002... Additional details and hyperparameters for the models are found in Appendix A.2.