A master thesis from Aalborg University
ORB-E: Ensemble Method with Query-Specific Weight Assignment Depending on Query Characteristics
Author(s)
Term
4. term
Education
Publication year
2023
Submitted on
2023-06-15
Pages
13 pages
Abstract
The performance of link prediction on Temporal Knowledge Graphs (TKGs) has improved in the last decade via development of several new and diverse Knowledge Graph Embedding (KGE) methods. In this paper, the strengths and weaknesses of several temporal KGE methods are examined, with focus on the temporal data density of the overall Knowledge Graph (KG) and temporal properties of relations determined by the structure of the surrounding KG. The results of this analysis are used to create a new ensemble voting method ORB-E with query-specific model weights based on the characteristics of the query and model performance relative to that characteristic. The characteristics are target of prediction, temporal data density, properties of relations and overall scores of models. The rules that determine the weights of each model reflect which query characteristics have the most importance for different methods, datasets, and overall. We find that prediction target is the most influential characteristic, and that query-specific weight assignment has better performance than a static weight assignment. Link prediction preformance is increased for time prediction on temporally dense data but not other prediction targets and the theoretical expressivity of methods does not always reflect the actual performance of the models. Additionally, the domain of time predictions is analyzed to determine their accuracy under different circumstances and we find that most methods are not capable of predicting time within an acceptable error margin. A strategy that utilizes the continuous nature of time information and combines predictions of multiple models is presented, and can be used to improve time predictions when models have equal precision.
Keywords
Documents
Colophon: This page is part of the AAU Student Projects portal, which is run by Aalborg University. Here, you can find and download publicly available bachelor's theses and master's projects from across the university dating from 2008 onwards. Student projects from before 2008 are available in printed form at Aalborg University Library.
If you have any questions about AAU Student Projects or the research registration, dissemination and analysis at Aalborg University, please feel free to contact the VBN team. You can also find more information in the AAU Student Projects FAQs.