TUTORIAL DAY - December 18, 2007

Tutorial 1: 9.30-12.50 - Room "Aula II" - Palazzo delle Aule

Chris Welty
IBM Watson Research Center

Title: Ontology Engineering with OntoClean

Designing and specifying ontologies is an engineering task that, like any other, benefits from the experiences of learning from failure. OntoClean is a formally specified methodology for ontological analysis, whose primary motivation was to explain why the errors that many experienced ontology designers recognize are, in fact, errors. In this tutorial I will cover some of the basics of OntoClean, and move through examples of principled ontology design.

Tutorial 2: 14.30-16.30 - Room "Aula II" - Palazzo delle Aule

Marko Grobelnik, Blaz Fortuna, Dunja Mladenic
Jozef Stefan Institute, Ljubljana, Slovenia

Title: What Semantic Web researchers need to know about Machine Learning?

The tutorial will cover basic topics from the field of Machine Learning explained in an intuitive way relevant for Semantic Web researchers and practitioners. In the first part the topics will cover brief top level overview of the Machine Learning field, its algorithms, and data types being analyzed. In the second part we will cover relation to Semantic Web and Web2.0. In the last part we will perform hands-on exercise with some of the tools for modeling text semantics and social networks in analytical way.

Tutorial 3: 14.30-16.30 - Room "Hume" - DIB, 2nd floor

Claudia d'Amato, Nicola Fanizzi, Francesca A. Lisi
Dipartimento di Informatica, Universita' degli Studi di Bari, Italy

Title: Inductive Reasoning on Ontologies and Rules for the Semantic Web

Building ontologies as well as rules on top of ontologies for the Semantic Web is a very demanding task. When performing this task, Semantic Web practitioners could take benefit from the application of Machine Learning (ML) methods and techniques, e.g. in order to support Ontology Construction and Refinement. Ontology and rule languages for the Semantic Web are logically founded on fragments of first-order logic, namely Description Logics (DLs) and hybrid languages combining DLs and Horn Clausal Logic (HCL). The tutorial will provide a survey of ML proposals for inductive reasoning with DLs and hybrid DL-HCL languages. Attendees are expected to be knowledgeable in standard languages for Semantic Web ontologies and rules, more precisely in their logical foundations (DLs and HCL). The tutorial will focus mainly on theoretical aspects but the presentation will be rich in illustrative examples, animations, etc. in order to assure that the attendees fully comprehend the potential of ML as an inductive reasoning means for the Semantic Web.

Tutorial 4: 16.45-18.45 - Room "Aula II" - Palazzo delle Aule

Roberto Basili
Department of Computer Science, Systems and Production
AI Research Group (ART)
University of Roma, Tor Vergata
ISTC-CNR, Roma, Italy


Alessandro Moschitti
Department of Information and Communication Technology
Universita' degli studi di Trento, Italy

Title: Support Vector Machines and Kernel Methods for Robust Extraction of Textual Knowledge

The future development of the Semantic Web will depend on the availability of automatic tools able to extract semantic information from distributed resources. As most of the Web content is still textually expressed, the ability of automatically extracting semantic information from linguistic data is critical to the Semantic Web success. Recognition of relational (e.g. predicate argument) structures or named entities and their mapping into descriptive metadata are just some examples of these processes. Unfortunately, accurate extraction from unstructured text is still a challenging task as the resulting semantic data is often incomplete and/or noisy.
Traditional solutions for such problems are based on uncertainty models and probabilistic reasoning. However, these require the definition of complex models, especially when dealing with structured data. Moreover, the applied smoothing techniques are often not enough robust to provide reliable information to the target reasoning system.
Discriminative approaches such as maximum margin algorithms, e.g. Support Vector Machines, and kernel methods represent a very important alternative. The former, provide us with a well defined theory and learning algorithms robust to noise and to irrelevant features. The latter offer a powerful approach to the data representation problem, as kernel functions can be used to define similarities between complex objects at more abstract levels.
In this talk, we will first outline the traditional ideas characterizing the inductive approaches to natural language processing. Then, we will present the theory of Support Vector Machines and Kernel Methods, emphasizing the practical and experimental aspects. In particular, we will show (a) the theory behind several effective kernels for Natural Language Applications, e.g. Tree Kernels, String Kernels and Lexical Semantic Kernels, and (b) their practical implementation in the SVM-Light-TK (tree kernels) toolkit. The experiments on two different language tasks such as Semantic Role Labeling and Question Answering on standard benchmarks will illustrate the most important properties of the proposed techniques in the light of applications fundamental for the Semantic Web.

Tutorial 5: 16.45-18.45 - Room "Hume" - DIB, 2nd floor

Umberto Straccia
ISTI-CNR, Italy

Title: Managing Uncertainty and Vagueness in Semantic Web Languages

Managing uncertainty and/or vagueness is starting to play an important role in Semantic Web research. The tutorial presents the state of the art in representing and reasoning with uncertain and/or vague knowledge in the semantic web. The goal of the tutorial is to make attendees familiar with the main concepts and techniques for representing and reasoning with uncertain and vague knowledge in semantic web ontology and rule languages, which will help the attendees to gain insights on the main features of the formalisms and tools proposed so far.