Semi-Automated Construction of Linked Data Repository Using Unstructured Web Portals' Text

U radu je najprije sagledano trenutačno stanje u semantičkom webu. Opisani su osnovni koncepti, tehnologije i poznatiji repozitoriji. Nastavlja se s pregledom koncepata strojnog učenja te načina na koji on može pomoći semantičkom webu. Pregledava se postojeće modele za prepoznavanje relacija te se izgrađuje i evaluira vlastiti. Zatim se opisuje servis koji koristi model za obradu tekstova s pet odabranih web portala s vijestima. Nabrajaju se komponente servisa te se pojedinačno opisuju. Nakon toga se opisuje izrađena ekstenzija za preglednik Chrome koja služi kao korisničko sučelje te stranica za vizualizaciju relacija. Na kraju se razmatra mogućnost kontinuiranog poboljšavanja modela.The thesis first makes assessment of the current state in the field of semantic web. Basic concepts, technologies and repositories that are wider known are described. It is continued with overview of machine learning concepts and its potential benefits for the semantic web. Existing models for relation extraction are reviewed and own model is built and evaluated. Then the service which uses the model to process texts from five chosen news portals is described. Service components are listed and individually described. After that Chrome extension, which is used as a user interface, and page for relations visualization are described. At the end, possibility of continuous model improvement is considered