I started with evaluating the ontology matching algorithm last week. There are 830 source ontologies and 10 target ontologies in our dataset. I began by matching each source ontology with all of the target ontologies. However, this process was very time consuming and the algorithm took nearly 5 hours to make comparisons for only 60 source ontologies. Since we have 830 source ontologies, we decided that this approach was not feasible. My mentor suggested that we could combine the target ontologies into a single ontology and then proceed with matching. I began by using Protege tool to merge the ontologies. This process resulted in a single ontology but AgremmentMaker Lite (AML) was not able to load the classes or properties from this merged ontology. Another option to get around this issue was to make use of the ontologies that were already merged for the other part of this project that uses Linkipedia to identify entities. But since the merged ontologies were saved as N-Triples for their purpose while the ontology matching algorithm requires the input to be in OWL or RDF, I could not use them directly. Hence, I explored techniques that could convert N-Triples to OWL. Protege has this functionality, but for our case, it gave errors and showed problem in the target ontologies themselves. I found out that Jena API could also be used to read in N-Triples and write out OWL files. After configuring Jena in eclipse and reading in the N-Triples, I was able to get a OWL file but AML was still not able to identify classes and properties from this file. Hence, I reached to a conclusion that there might be some problem in the method for reading the OWL file in AML and not in the file itself.
Upon further debugging the code, I found out that AML incorrectly identifies the merged ontology as a SKOS formatted ontology and therefore uses a different method for extracting the classes and properties. After fixing this problem and handling few other exceptions, I was successfully able to load the merged ontology into AML. I plan to further proceed with evaluation of the entire dataset next week.