4th International Semantic Web Conference (ISWC'2005)
Author: Raphael
CWI participants: Jacco, Lloyd, Raphael, Zhisheng
# participants: 500
The ISWC audience is still growing and getting bigger each year! This year, there were more people that in Hiroshima (which was already a record), that is about 500 people. The poster session was according to my point of view the best part of the conference: very long, high quality, each presenter had its own table and can make some demo. This should encourage us to submit posters for the next years.
For the group interest, unfortunately, like each year, they were very few (to not say 0) works on multimedia and semantic web! I detail below only the presentations on ontology alignment (2 full sessions this year, quite a hot topic) and a presentation of the goals of the next European FP7 research program. The proceedings of the conference is available in my office.
The two sessions dedicated to ontology alignment at ISWC were quite interesting.
Mikalai Yatskevich in "A Large Scale Taxonomy Mapping Evaluation", did the work for building the task 2 of the OAEI contest, that is align the web directories Google, Yahoo and Locksmart. I asked directly some of the data I would like to use: the whole hierarchies and the list of URLs belonging to each category. He told me that all these data are part of a project and that they cannot provide them before the end of the project, that is in a couple of years so I have to forget that ! I asked him also about the strange result he shows (very good in ISWC and bad in OAEI) with his system: and he answered me that he has lied in OAEI because in his lab, there is also a competting system (ctxMatch, I think) and his italian colleagues said it was not fair since he was the only one to have the data. So we should not consider the results he published in OAEI, thus I wonder why they publish this paper but it is another story ...
Marc Ehrig in "Bootstrapping Ontology Alignment Methods with APFEL" has presented the FOAM tool. FOAM has some similarities with oMAP in the sense that it uses also plenty of similarity measures computed according to different features of the ontologies. These features could be the labels (and then the similarity is based on String similarity), some OWL constructs (one classifier per OWL construct), the individuals, etc ... They do not use machine learning techniques which make oMAP different. The hard problem they encounter is that they finally obtain about 130 different similarity measures that should be aggregated and combined! They use for this purpose a learning algorithm, trained on various test sets (OAEI data + travel and tourism ontologies). They claim that the weights learned for each similarity measure are then valid for any ontology alignment problem. The tool is available on http://www.aifb.uni-karlsruhe.de/WBS/meh/foam/
Giorgos Stoilos in "A String Metric for Ontology Alignment" has presented a new string similarity which should perform better than the Edit distance, the Levenstein distance, or all the other string distances. What is interesting is that this string measure is compatible with the Jerôme API, and I took it (the greek guy gave me the source files). You can download it also on http://manolito.image.ece.ntua.gr/~gstoil/ I will include it in our framework ...
The second session was less interesting.
Finally, Willem Robert von Hage has manually aligned two very big food ontologies (expressed in OWL). He thought to bring that in the next OAEI data when the alignment will be confirmed by the expert of the domain.
Generally, EU would like to have more show case examples and very re-usable pieces of technological components. The projects accepted in the call 1 and 4 for the Knowledge part, dealt mainly with:
The new framework is expected to start at the end of 2006. The name of the program (IST) will change to ICT. All the documents concerning the FP7 are already on the web http://www.cordis.lu. Amongst the new instruments, EU would like to support individually VERY good (young) researchers and/or VERY good team, rather than only projects. The fundings will not be anymore split into disciplines.
The ICT in FP7 will be divided into: