Publication: Enhancing CNN Based Knowledge Graph Embedding Algorithms Using Auxiliary Vectors: A Case Study of Wordnet Knowledge Graph
Issued Date
2020-06-01
Resource Type
Other identifier(s)
2-s2.0-85091888072
Rights
Mahidol University
Rights Holder(s)
SCOPUS
Bibliographic Citation
17th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology, ECTI-CON 2020. (2020), 763-766
Suggested Citation
Chanathip Pornprasit, Pattararat Kiattipadungkul, Peeranut Duangkaew, Suppawong Tuarob, Thanapon Noraset Enhancing CNN Based Knowledge Graph Embedding Algorithms Using Auxiliary Vectors: A Case Study of Wordnet Knowledge Graph. 17th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology, ECTI-CON 2020. (2020), 763-766. doi:10.1109/ECTI-CON49241.2020.9158288 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/59943
Research Projects
Organizational Units
Authors
Journal Issue
Thesis
Title
Enhancing CNN Based Knowledge Graph Embedding Algorithms Using Auxiliary Vectors: A Case Study of Wordnet Knowledge Graph
Other Contributor(s)
Abstract
© 2020 IEEE. Knowledge graphs (KGs) have been utilized by various business fields. One example is Google that stores data in knowledge graphs for searching and retrieval tasks. Even though these graphs have reached an impressive size, they are far from completeness. Missing relations in knowledge graphs is a severe problem for algorithms that operate over knowledge graphs. There are many researchers trying to develop knowledge graph embedding methods so that they can handle different types of relations. ConvKB is one of the knowledge graph embedding methods that utilize convolution neural networks (CNN). However, this method lacks the ability to handle symmetric relations. Being inspired by this limitation, we would like to enhance this method by proposing ConvKB+, which is obtained by modifying ConvKB's CNN structure and introducing an additional relation vector. Our experiment results show that our method outperforms ConvKB by achieving higher MRR on some symmetric relations of the WN18RR dataset.