This repository contains the dialogue corpus, source code, and evaluation output for the study "Bridging Information Gaps in Dialogues With Grounded Exchanges Using Knowledge Graphs" accepted at SIGDIAL 2024.
Notebook: The Jupyter notebook (cg-kg-llm-prediction.ipynb) contains the experimental procedures including model configuration, prompts, as well as the analysis procedures.
/dialogue_corpus - contains the annotated dialoque corpus (annotated_dialogue_corpus.xlsx) including separate JSON files for each conversation
/domain_datasets - contains the tabular datasets serving as knowledge base
/clsruns - contains raw generated output of grounding act classification experiments
/gkruns - contains raw generated output of grounded knowledge identification experiments
/evaluation - contains the evaulation output of the experiment analysis \
- Open the Jupyter notebook using a compatible environment or platform.
- Follow the step-by-step instructions to reproduce the experiments.