Automated Rule Base Completion as Bayesian Concept Induction

Authors: Zied Bouraoui, Steven Schockaert6228-6235

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present experimental results that demonstrate the effectiveness of our method.
Researcher Affiliation Academia Zied Bouraoui CRIL CNRS & Univ Artois, France zied.bouraoui@cril.fr Steven Schockaert Cardiff University, UK Schockaert S1@Cardiff.ac.uk
Pseudocode No The paper does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any specific links or explicit statements about releasing source code for the described methodology.
Open Datasets Yes As knowledge bases we consider several well-known OWL ontologies, which we converted to existential rule bases. In particular, we consider two large-scale open-domain ontologies: SUMO and Open Cyc. We also test the performance on a number of smaller domain-specific ontologies: Wine ontology, Economy, Transport and Vicodi. As word embedding, We used a standard pre-trained 300-dimensional, which was learned using Skip-gram on the 100B words Google News corpus.
Dataset Splits Yes To split the rule bases into training and test rules, we use 10-fold cross validation. To set the parameters of our model (and the baselines), we select 10% of the training data as validation data, and only use the remaining 90% for training the model.
Hardware Specification No The paper does not specify any hardware details (e.g., CPU, GPU models, or memory) used for running its experiments.
Software Dependencies No The paper mentions 'pellet reasoner' and 'Skip-gram' but does not provide specific version numbers for these or any other software dependencies.
Experiment Setup Yes This validation data is used for selecting the parameter ยต and for choosing the number of dimensions in the vector space representations v R r (chosen from {10, 25, 50, 100}). For the classification experiments, we also tune a threshold on the probability for a rule to be predicted as valid.