Phd Activity of ANGELA LOCORO

PhD Courses and Credits

• Tecniche di Trasformazione di Spazi Vettoriali Multidimensionali per Applicazioni Statistiche
  Docente: Ing. Gabriele Moser, Dept. of Biophysical and Electronic Eng. (DIBE), Università di Genova, Italy
  esame: SOSTENUTO
• Machine Learning
  Docente: Ing. Marco Muselli, Istituto di Elettronica e di Ingegneria dell'Informazione e delle Telecomunicazioni (I.E.I.I.T.), Sezione di Genova, Italy.
  esame: SOSTENUTO
• Statistica e regressione non lineare
  Docente: Ing. Cristiano Cervellera (ISSIA-CNR, Genova)
  esame: SOSTENUTO

• Seminario di Linguistica Computazionale
  Docente: Prof. Alessandro Lenci

• Modularity in logical theories and ontologies, Prof. Carsten Lutz, University of Dresden
• Parsing beyond context-free grammars, Prof.ssa Laura Kallmeyer, University of Tubingen
• Lexical semantics: bridging the gap between semantic theory and computational simulations, Prof. Alessandro Lenci, Università di Pisa e Marco Baroni, Center for Mind/Brain Sciences, Università di Trento
• Logics for computation, Prof. Carlos Areces e Patrick Blackburn, INRIA Nancy Grand Est
• The syntax-semantics interface: theoretical insights and practical implementations, Prof. Markus Egg, University of Groningen e Prof. Valia Kordoni, University of Saarland
• Semantic relation extraction and its applications, Prof. Roxana Gjriu, University of Illinois
• Statistical language modeling for information access, Prof. Maarten De Rijke, University of Amsterdam
• Unification grammars, Prof. Shuly Wintner, University of Haifa

PhD Activity

The core part of my three years PhD programme was the design, implementation and experiments of ontology matching and alignment repair algorithms, as well as an analysis on ontology matching state-of-the-art tools, and the preliminary study, design and implementation of an approach using WordNet Domains for ontology and concepts classication.

All these different approaches, not necessarily separated, have been carried out for boosting the ontology matching process and for applying it to some application scenarios, in order to demonstrate the real benefits from using them. In concrete, my activity can be summarised according to the following aspects:

• Ontology Matching via Upper Ontologies. In many real Semantic Web applications, performing a direct alignment between two ontologies is not possible. In those cases, using a "reference ontology" as a "semantic bridge" may allow the application developer to align the two ontologies in an indirect way. Good candidates for acting as reference ontologies are Upper Ontologies, namely ontologies describing very general concepts that are the same across all domains. The study, design and implementation of ontology matching innovative methodologies using Upper Ontology and their systematic evaluation has been carried out, in order to state under which circumstances the exploitation of this background knowledge gives significant advantages with respect to other approaches that do not use it.
• Ontology Matching and Repair via Natural Language Processing (NLP). Achieving a better accuracy in creating correspondences between ontologies is a key factor justifying Semantic Web synergies with other research fields. A thorough analysis of methods using NLP in state-of-the-art ontology matching tools has been conducted, together with two new original contributions: the first is the design and implementation of an approach based on Word Sense Disambiguation (WSD) applied on concepts before the matching operation and on semantic inconsistencies detection used for refining alignments based on inheritance and disjointness reasoning; the second refers to the design and implementation of an approach based on both WSD and Upper Ontologies and exploited for repairing mappings.
• NLP and ontology matching: Integrated Approaches. The combination of both ontology-driven and unstructured semantic data is an emergent need in fields such as for example Social Netword Analysis. A concrete application aiming at concepts extraction from social network systems activities, such as those of tagging contents, through NLP and matching of such knowledge with domain ontologies through ontology matching, is one of the original work designed, implemented and tested during my research activity.
• WordNet Domains for ontology and concepts classification. The use of WordNet Domains is confined in the present days to text mining field. Exploiting WordNet Domains to automatically determine the domain of ontologies and using this information as a semantic layer to pre-process ontologies before matching them, is the original idea followed, designed and implemented to obtain even more accurate semantic reasoning methodologies for ontology classification and matching.