Inductive Reasoning about Ontologies Using Conceptual Spaces
Authors: Zied Bouraoui, Shoaib Jameel, Steven Schockaert
AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Overall we can see that our method outperforms similarity based reasoning, especially in terms of precision for concepts with many instances. In particular, if the majority of instances of A are not instances of B, there are very limited guarantees that the most similar instances to the entity being categorized will be instances of B, which is a result of the context-dependent nature of similarity. Note that, for our method, by setting the threshold for the confidence degree higher than 0, we can increase precision although at the cost of lower recall. The results for both variants of TBox reasoning are shown in Table 3. |
| Researcher Affiliation | Academia | Zied Bouraoui Cardiff University, UK Bouraoui Z@Cardiff.ac.uk Shoaib Jameel Cardiff University, UK Jameel S1@Cardiff.ac.uk Steven Schockaert Cardiff University, UK Schockaert S1@Cardiff.ac.uk |
| Pseudocode | No | The paper provides mathematical formulations and descriptions of the proposed method, but it does not include any explicitly labeled |
| Open Source Code | Yes | We have implemented2 our method, using Java including OWLAPI3 and Pellet reasoner4 for deductive reasoning tasks (e.g. determining the instances or subconcepts of a given concept, checking consistency), and using a standard approach to repairing inconsistencies which uses the confidence scores as a penalty. Footnote 2 links to http://www.cs.cf.ac.uk/semanticspaces/ which hosts a |
| Open Datasets | Yes | To evaluate the effectiveness of the proposed approach, we have used the OWL version of the SUMO ontology5. This is a relatively large open-domain ontology, covering a total of 4558 concepts, 86457 individuals, 5330 inclusion axioms and 167381 ABox assertions. An advantage of SUMO is that for several concepts, an explicit mapping to Word Net is provided (which has, in turn, been linked to Wikidata). |
| Dataset Splits | Yes | For the evaluation of ABox reasoning, we use three-fold cross validation7 as follows. Let Y be the entities that are asserted to be instances of B in SUMO, and let X be the entities that are asserted to be instances of A (where Y X). We split Y = Y1 Y2 Y3 in three sets of (approximately) the same size. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used to run the experiments, such as GPU/CPU models or memory specifications. |
| Software Dependencies | No | We have implemented2 our method, using Java including OWLAPI3 and Pellet reasoner4 for deductive reasoning tasks... Footnote 3 links to http://owlapi.sourceforge.net and Footnote 4 links to https://github.com/stardog-union/pellet. While Java, OWLAPI, and Pellet are mentioned as software components, no specific version numbers are provided for any of them. |
| Experiment Setup | No | The paper describes the mathematical model and inference rules, such as using a normal-inverse-χ2 prior and making inductive inference when the confidence score is strictly positive. However, it does not specify concrete experimental setup details like hyperparameters (e.g., learning rates, batch sizes, epochs) for training or running the model. |