LogMap Parameters for Replicating OAEI2013

[Axel Magnuson]

Hello,

I am attempting to replicate the LogMap results from IMEI2013 listed here before I attempt to use it for my project.  However, I think I'm configuring it incorrectly.  So far, I have been unable to find any mappings on any datasets with logmap.  I will link one such attempt below:

runscript

parameters.txt

verbose output (stdout)

dataset source (unzipped into /data/test/in/rdft_datset)

The verbose output looks a little like it's using an incorrect input from http://logmap.cs.ox.ac.uk/ontology.owl for both files, rather than using the ones I specified, but it is hard to tell.  Any help would be greatly appreciated.  I really can't tell what I'm doing incorrectly, and unfortunately documentation is sparse.

Best Regards,

Axel Magnuson

[Ernesto Jimenez-Ruiz]

Hi Axel

I did it a bit of preprocessing to the datasets in order to be properly interpreted by the OWL API. This is why you did not get any results using the original dataset. See below the comments I sent to Alfio Ferrara (IMEI organiser).

<<<
To compute the instance mappings I needed to pre-process the input datasets since in the current RDF format the OWLAPI was interpreting the given instance entities as "Annotations" and not as "OWLNamedIndividual". Both LogMap and LogMapLite rely on the OWLAPI and only try to match named individuals.

If you try to load any of the given rdf ontologies with Protege, Protege only detects annotations and no property nor instance is listed.

Thus, I made the following preprocessing steps to create an OWL version of the given original.rdf, training.rdf and contest.rdf (see attached the extended datasets).

1. The ontologies import now dbpedia version 3.8 so that used data and object properties have a correspondence and individuals can be classified, using an OWL 2 reasoner, to the corresponding class type. LogMap closely relies on the associated type to each individual in order to validate candidate instance mappings.

2. I slightly modified dbpedia ontology (see
http://csu6325.cs.ox.ac.uk/ontologies/matching_25_08_2013/dbpedia_3.8_no_data_ranges.owl) to avoid inconsistencies when reasoning with the given datasets. I removed dataproperty ranges and functionality in the role "birthdate". In the dataset some individuals were using "birthdate" in two or more role assertion axioms while functionality states that at most can be used once in the same individual. Some datatype role assertion were also pointing to the wrong type (e.g. string) wtr the specified datarange in dbpedia (e.g. integer).

3. I explicitly stated that "curricuulum", "places" and "label" are OWLObjectProperties.

4. In testcase1 RDFT dataset, in both training.rdf and contest.rdf I found invalid  characters or scape sequences that could not be processed by OWL API and Protege (e.g. \u). For example

- In contest.rdf: line 1416 column 356, lines 16238 col 268

- In training.rdf: line 475 col 289, line 686 col 118, line1116 col 147

[Axel Magnuson]

Hi Ernesto,

Thanks for the prompt reply.  To generalize your process to other datasets, what necessary conditions should a dataset satisfy for LogMap to run on it well?  What I'm seeing is:

1. It must be correctly formed OWL to be interpreted by the OWLAPI. There isn't much room for data that doesn't conform to W3C, and invalid escape sequences must be cleaned.
2. It cannot have logical inconsistencies for the reasoning steps
3. Some manual configuration is necessary to specify OWLObjectProperties

How do you think a subset of Freebase would fare, since it seems to prefer its own ontology URI's over the W3C standard?

Best Regards,

Axel Magnuson