Efficient Concept Induction for Description Logics
Authors: Md Kamruzzaman Sarker, Pascal Hitzler3036-3043
AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The goal of our experimental evaluation was to test the hypothesis that the ECII algorithm leads to a favorable tradeoff between runtime improvements and loss in accuracy, compared to DL-Learner. |
| Researcher Affiliation | Academia | Md Kamruzzaman Sarker, Pascal Hitzler Data Semantics (Da Se) Laboratory, Dept. of Computer Science and Engineering, Wright State University {sarker.3, pascal.hitzler}@wright.edu |
| Pseudocode | No | No structured pseudocode or algorithm blocks were found. |
| Open Source Code | Yes | The ECII system and all experimental data and results, including ontologies and configuration files are available online.3 https://github.com/md-k-sarker/ecii-expr. |
| Open Datasets | Yes | The evaluation scenarios from (Lehmann and Hitzler 2010) were Michalski s trains (Michalski 1980), Forte Family (Richards and Mooney 1995), Poker (http://www.ics.uci. edu/ mlearn/MLRepository.html), Moral Reasoner (http:// mlearn.ics.uci.edu/databases/moral-reasoner/), and Yinyang family relationship (Iannone, Palmisano, and Fanizzi 2007)... as well as the scenario from (Sarker et al. 2017) which makes use of the ADE20k (Zhou et al. 2017) dataset. |
| Dataset Splits | No | The paper mentions 'validation data ontology' and 'training data ontology' for the ADE20k dataset but does not provide specific details on how these splits were created (e.g., percentages, sample counts, or methodology for partitioning). |
| Hardware Specification | Yes | All experiments were conducted on a 2.2. GHz core I7 machine with 16GB RAM. |
| Software Dependencies | Yes | To evaluate our approach, we implemented the ECII system in Java (version 1.8) which makes it platform independent. We made use of the OWL API (Horridge and Bechhofer 2011) (version 4.5) |
| Experiment Setup | Yes | For DL-Learner we used the CELOE (Lehmann et al. 2011) algorithm and the fast instance check (DL FIC) variation of it, which is another approximation approach which trades time for correctness. We terminated DL-Learner at the first occurance of a solution with accuracy 1.0, making use of the stop On First Definition parameter. For some large ontologies DL-Learner could not produce a solution with accuracy 1.0 within 4,500 seconds (i.e., 75 minutes); in these cases we terminated the algorithm after 4,500 seconds, using the max Execution Time In Seconds parameter. [...] For ECII we used the default settings, for Ks i.e., k1 = k2 = k3 = 3 and k4 = k5 = 50 and varied the keep Common Types as true and false, as mentioned earlier. |