A Framework and Positive Results for IAR-answering
Authors: Despoina Trivela, Giorgos Stoilos, Vasilis Vassalos
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we have provided a prototype implementation and a preliminary evaluation obtaining encouraging results. and Evaluation Based on Algorithm 1 we created a prototype system... Our test ontologies consist of the seven ontologies used in (Hansen et al. 2015)... Our results for the rest of the ontologies are depicted in Tables 2 and 3. |
| Researcher Affiliation | Collaboration | Despoina Trivela Athens University of Economics and Business Athens, Greece despoina@aueb.gr Giorgos Stoilos Babylon Health London, SW3 3DD, UK Vasilis Vassalos Athens University of Economics and Business Athens, Greece vassalos@aueb.gr |
| Pseudocode | Yes | Algorithm 1 IAR-Rewriting Input: a CQ Q and an L-TBox T 1: Compute a negative closure Tcn of T 2: T cn := minimise(saturate(Tcn)) 3: Compute a rewriting R of Q w.r.t. T 4: R := saturate(R) 5: Rir := 6: for Q R do 7: Q := Q 8: for α Q where α is not an inequality atom do 9: for β1 . . . βm T cn do 10: if α = βkμ, μ a renaming, k in [1, m] then 11: Add (β1 . . . βm)μ to Q 12: end if 13: end for 14: end for 15: Rir := Rir {Q } 16: end for 17: return Rir |
| Open Source Code | No | The paper mentions creating a "prototype system" and that it uses external tools like Rapid and Grind, but it does not state that the code for their own system is open-source or publicly available, nor does it provide a link. |
| Open Datasets | Yes | Our test ontologies consist of the seven ontologies used in (Hansen et al. 2015)... Furthermore, we have used ELHdr fragments of the ontologies CARO,1 BFO2 and Dolce-Lite.3 [1:http://www.obofoundry.org/ontology/caro, 2:http://www.ifomis.org/bfo/1.1, 3:http://www.loa.istc.cnr.it/old/DOLCE] |
| Dataset Splits | No | The paper discusses testing on various ontologies and manually constructed test queries but does not specify train, validation, or test dataset splits in the context of model training/evaluation. |
| Hardware Specification | No | The paper reports computation times but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for running the experiments. |
| Software Dependencies | No | The system uses Rapid (Trivela et al. 2015) and Grind (Hansen et al. 2015), but no specific version numbers for these software dependencies are provided. |
| Experiment Setup | Yes | Based on Algorithm 1 we created a prototype system. It is using Rapid (Trivela et al. 2015) to compute a rewriting R for Q and T (line 3 of Algorithm 1) and Grind (Hansen et al. 2015) along with the approach described in Lemma 9 to decide whether it can compute a negative closure Tcn... The whole system currently supports ELHdr ontologies as this is the language supported by the current implementation of Grind... for each ontology we manually constructed five test queries. Each one of them contains at least one body atom that uses a predicate (concept or role) involved in a negative clause. More precisely, for axioms of the form B C we have constructed queries Q(x) A(x) and Q(x) D(x) such that T |= A B and T |= B D. We also tried to use concepts that appear low or high in the ontology hierarchy... In order to avoid it [increase in size of IAR-rewriting] we restricted its application to elements of R that contain roles that also appear in some negative clause. |