Artificial Intelligence
Natural language processing
Semantic web

Improvement of the performance of semantic annotators

Student: Mohamed Chabchoub

Supervisor: Amal Zouaq

Co-supervisor(s): Michel Gagnon

Semantic annotators play an important role in the transition from the current Web to the Semantic Web. They take care of extracting structured information from raw texts and thus make it possible to point to knowledge bases such as DBpedia, YAGO or Babelnet. Many competitions are organized every year to promote research works in this field. We present in this thesis our system which was the winner of the Open Knowledge Extraction challenge at the European Semantic Web Conference 2016. For this competition, we implemented a generic approach tested with four semantic annotators. We particularly focus in this thesis on one semantic annotator, DBpedia Spotlight. We present its different limitations along with the approaches that we have developed to remedy them. We noted an improvement of an average of 20% of the current performance of DBpedia Spotlight on different corpora that mainly come from international newspapers, "Reuters News Stories", "MSNBC" and the "New York Times".

For more information, click here



Collective disambiguation and Semantic Annotation for Entity Linking and Typing

Mohamed Chabchoub, Michel Gagnon, Amal Zouaq

ESWC-16 Open Knowledge Extraction Challenge, Heraklion, Greece