Repository with implementation and evaluation data of LEAPME, by Daniel Ayala, Inma Hernández David Ruiz, and Erhard Rahm.
Contains the following files/folders:
- datasets: Contains the 4 datasets used for validation. Includes the original files, a TAPON-compatible version, OWL versions, and a json file version. Also includes the mappings to reference ontologies from which positive pairs were generated.
- features: Contains the features computed for the application of LEAPME. Contains both the instance features computed by TAPON (including the added embeddings ones), and the final property-pair features.
- implementations: Contains:
- The python implementation of LEAPME for both use cases, which corresponds to the training and application of a neural network that takes the already computed features.
- The python implementation of the technique by Nezhadi et al. corresponding to both use cases. The first file (use-case-1) computes the features associated to property pairs. The second file (use-case-2) uses this file.
- The python implementation (as a jupyter notebook) of SemProp, including auxiliary files used for computing LSH (which is used by a matcher).
- The python implementation (as a jupyter notebook) of the proposal by Duan et al., including an auxiliary file for computing LSH.
- The java script that was used to add embedding features to instances. Note that this script uses the method "getEmbeddings(word)" which can be found in the java file "GloveEmbeddingsCalculator.java".
- The R script that computes property features and property pair features from the instance ones.
- UC1/2-results: The detailed evaluation results for all setups.