Open Rule Induction

Authors: Wanyun Cui, Xingran Chen

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conducted extensive experiments to verify the quality and quantity of the inducted open rules.
Researcher Affiliation Academia Wanyun Cui , Xingran Chen Shanghai University of Finance and Economics cui.wanyun@sufe.edu.cn, xingran.chen.sufe@gmail.com
Pseudocode Yes Algorithm 1: Supported beam search
Open Source Code Yes Code and datasets are available at https://github.com/chenxran/Orion
Open Datasets Yes Code and datasets are available at https://github.com/chenxran/Orion
Dataset Splits No The paper mentions using several relation extraction datasets (Google-RE, TREx, NYT10, WIKI80, Few Rel, Sem Eval) and their own 'Open Rule155' dataset. However, it does not explicitly provide specific training/validation/test split percentages or sample counts for any of these datasets in the main text that would allow reproduction of data partitioning.
Hardware Specification Yes All the experiments run over a cloud of servers. Each server has 4 Nvdia Tesla V100 GPUs.
Software Dependencies No The paper mentions using specific software components like 'Bart', 'Spacy NER library', and 'Exp BERT', but it does not provide specific version numbers for any of these software dependencies.
Experiment Setup Yes For each converted premise atom, we use Orion to induct k = 5, 10, 20 corresponding open rules. We following the settings of Exp BERT and use k = 29, 41 hypothesis atoms inducted by Orion as for Disease and Spouse, respectively. In addition, we modified Exp BERT to allow the training process to fine-tune the parameters that were frozen in the original Exp BERT, as we found that this will improve the model s effectiveness.