Stage-oe-small.jpg

Thema4979: Unterschied zwischen den Versionen

Aus Aifbportal
Wechseln zu:Navigation, Suche
Zeile 11: Zeile 11:
 
|Ausschreibung=Master_Thesis_EA_DDB.pdf
 
|Ausschreibung=Master_Thesis_EA_DDB.pdf
 
|Beschreibung DE=A Knowledge Graph (KG) contains real-world information organized to enable data sharing, understanding and reasoning. In order to make a KG consistent, integration with existing KGs is required. One such task, Entity Alignment1 (EA), attempts to map equivalent entities from one KG (source) to another KG (target).
 
|Beschreibung DE=A Knowledge Graph (KG) contains real-world information organized to enable data sharing, understanding and reasoning. In order to make a KG consistent, integration with existing KGs is required. One such task, Entity Alignment1 (EA), attempts to map equivalent entities from one KG (source) to another KG (target).
 +
</br>
 
</br>
 
</br>
 
EA techniques are often hampered by entity name variations, multilingualism, and the heterogeneity of source-target KGs. Recent developments in Machine Learning and Natural Language Processing have been shown to mitigate these challenges.
 
EA techniques are often hampered by entity name variations, multilingualism, and the heterogeneity of source-target KGs. Recent developments in Machine Learning and Natural Language Processing have been shown to mitigate these challenges.
 +
</br>
 
</br>
 
</br>
 
In particular, KG Embeddings3 (KGEs) have proven invaluable through their ability to encode an entity’s graph substructure, describing its neighborhood and the relations attached to it. These dense representations enable similarity-based scoring functions to compute equivalence between entities.  
 
In particular, KG Embeddings3 (KGEs) have proven invaluable through their ability to encode an entity’s graph substructure, describing its neighborhood and the relations attached to it. These dense representations enable similarity-based scoring functions to compute equivalence between entities.  
 +
</br>
 
</br>
 
</br>
 
In addition, pre-trained Language Models4 (PLMs) encode the semantics found in the literals attached to entity names and attributes. They have been shown to address name variations and multilingualism.
 
In addition, pre-trained Language Models4 (PLMs) encode the semantics found in the literals attached to entity names and attributes. They have been shown to address name variations and multilingualism.
 +
</br>
 
</br>
 
</br>
 
The goal of this thesis is to explore and combine the aforementioned techniques in order to integrate the German Digital Library4 Knowledge Graph (DDB-KG) to existing and well-established KGs, such as, the Integrated Authority File (GND) and Wikidata.  
 
The goal of this thesis is to explore and combine the aforementioned techniques in order to integrate the German Digital Library4 Knowledge Graph (DDB-KG) to existing and well-established KGs, such as, the Integrated Authority File (GND) and Wikidata.  
 +
</br>
 
</br>
 
</br>
 
Prerequisites:
 
Prerequisites:
Zeile 25: Zeile 30:
 
</br>- Interest in Natural Language Processing
 
</br>- Interest in Natural Language Processing
 
</br>- Interest in Machine Learning approaches
 
</br>- Interest in Machine Learning approaches
</br>Interest in Semantic Web technologies
+
</br>-Interest in Semantic Web technologies
 
}}
 
}}

Version vom 1. Dezember 2022, 09:57 Uhr



Potato, Potahto, Iliad, Ilias: Entity Alignment with Knowledge Graph Embeddings and Language Models




Informationen zur Arbeit

Abschlussarbeitstyp: Master
Betreuer: Mary Ann TanRussa BiswasHarald Sack
Forschungsgruppe: Information Service Engineering
Partner: FIZ
Archivierungsnummer: 4979
Abschlussarbeitsstatus: Offen
Beginn: 02. Januar 2022
Abgabe: unbekannt

Weitere Informationen

A Knowledge Graph (KG) contains real-world information organized to enable data sharing, understanding and reasoning. In order to make a KG consistent, integration with existing KGs is required. One such task, Entity Alignment1 (EA), attempts to map equivalent entities from one KG (source) to another KG (target).

EA techniques are often hampered by entity name variations, multilingualism, and the heterogeneity of source-target KGs. Recent developments in Machine Learning and Natural Language Processing have been shown to mitigate these challenges.

In particular, KG Embeddings3 (KGEs) have proven invaluable through their ability to encode an entity’s graph substructure, describing its neighborhood and the relations attached to it. These dense representations enable similarity-based scoring functions to compute equivalence between entities.

In addition, pre-trained Language Models4 (PLMs) encode the semantics found in the literals attached to entity names and attributes. They have been shown to address name variations and multilingualism.

The goal of this thesis is to explore and combine the aforementioned techniques in order to integrate the German Digital Library4 Knowledge Graph (DDB-KG) to existing and well-established KGs, such as, the Integrated Authority File (GND) and Wikidata.

Prerequisites:
- Good programming skills in Python
- Interest in Cultural Heritage
- Interest in Natural Language Processing
- Interest in Machine Learning approaches
-Interest in Semantic Web technologies


Ausschreibung: Download (pdf)