Semantic metadata enrichment : a semi-automatic approach for linking legacy metadata with knowledge organization systems / Bernd Moser
VerfasserMoser, Bernd
Begutachter / BegutachterinKlas, Wolfgang
Umfang2, IV, 73 Bl. : 1 CD-ROM ; Ill., graph. Darst.
HochschulschriftWien, Techn. Univ. u. Univ., Mag.-Arb., 2007
Bibl. ReferenzOeBB
Schlagwörter (DE)Ontologien / Thesauri / Klassifikation / maschinelles lernen
Schlagwörter (EN)ontologies / thesauri / metadata / machine learning / semantic enrichment / classification
URNurn:nbn:at:at-ubtuw:1-19198 Persistent Identifier (URN)
 Das Werk ist frei verfügbar
Semantic metadata enrichment [1.21 mb]
Zusammenfassung (Deutsch)

As with the amount of a rapidly growing digitised content, the need for advanced search mechanisms, that guide end users through the information flood, is also growing. Ontologies and semantic search mechanisms will play a key role in solving that issue, presumed that stored metadata records are ontology aware and machine processable also on a semantic level. Typically, this is not the case with existing legacy metadata records. A lot of enterprises are in possession of metadata records, which are useless to new discovery services, because of the missing ontology-awareness. An Ontology is a mechanism to formally represent knowledge by the use of classification. An ontology-aware metadata record contains, a relation which links the record to a concept of the ontology that represents the records content best. Such a record is also known as enriched.

To enrich such metadata records, a convenient way is needed. One can imagine that for the end-user this process would be too time consuming.

The idea therefore, is to develop a semi-automatic way, which makes a given metadata record ontology-aware. Semi-automatic enrichment can be done by using the power of machine-learning techniques. Machine-learning algorithms try to classify data on the basis of features. In the case of a metadata record, a feature would be its textual representation.

Different textual representations, lead to different classifications. A classifier can be trained by an user, so that the classifier knows which terms a data record must have, to relate it to a specific part of the ontology.