On Training Knowledge Graph Embedding Models

J 2021

On Training Knowledge Graph Embedding Models

MOHAMED, Sameh K, Vít NOVACEK a Emir MUNOZ

Základní údaje

Originální název

On Training Knowledge Graph Embedding Models

Autoři

MOHAMED, Sameh K, Vít NOVACEK (203 Česká republika, garant, domácí) a Emir MUNOZ

Vydání

Information, Switzerland, MDPI, 2021, 2078-2489

Další údaje

Jazyk

angličtina

Typ výsledku

Článek v odborném periodiku

Obor

10201 Computer sciences, information science, bioinformatics

Stát vydavatele

Švýcarsko

Utajení

není předmětem státního či obchodního tajemství

Odkazy

URL

Kód RIV

RIV/00216224:14330/21:00121336

Organizační jednotka

Fakulta informatiky

DOI

http://dx.doi.org/10.3390/info12040147

UT WoS

000643062300001

Klíčová slova anglicky

loss functions; knowledge graph embeddings; link prediction

Štítky

Artificial Intelligence, knowledge graphs

Příznaky

Recenzováno

Změněno: 23. 5. 2022 14:41, RNDr. Pavel Šmerk, Ph.D.

Anotace

V originále

Knowledge graph embedding (KGE) models have become popular means for making discoveries in knowledge graphs (e.g., RDF graphs) in an efficient and scalable manner. The key to success of these models is their ability to learn low-rank vector representations for knowledge graph entities and relations. Despite the rapid development of KGE models, state-of-the-art approaches have mostly focused on new ways to represent embeddings interaction functions (i.e., scoring functions). In this paper, we argue that the choice of other training components such as the loss function, hyperparameters and negative sampling strategies can also have substantial impact on the model efficiency. This area has been rather neglected by previous works so far and our contribution is towards closing this gap by a thorough analysis of possible choices of training loss functions, hyperparameters and negative sampling techniques. We finally investigate the effects of specific choices on the scalability and accuracy of knowledge graph embedding models.

Přiložené soubory

information-12-00147.pdf

Vyhledat podobné dokumenty

Citovat

MOHAMED, Sameh K, Vít NOVACEK a Emir MUNOZ. On Training Knowledge Graph Embedding Models. Information. Switzerland: MDPI, 2021, roč. 12, č. 4, s. 147-165. ISSN 2078-2489. Dostupné z: https://dx.doi.org/10.3390/info12040147.

@article{1757476,
   author = {Mohamed, Sameh K and Novacek, Vít and Munoz, Emir and Muñoz, Emir},
   article_location = {Switzerland},
   article_number = {4},
   doi = {http://dx.doi.org/10.3390/info12040147},
   keywords = {loss functions; knowledge graph embeddings; link prediction},
   language = {eng},
   issn = {2078-2489},
   journal = {Information},
   title = {On Training Knowledge Graph Embedding Models},
   url = {https://www.mdpi.com/2078-2489/12/4/147},
   volume = {12},
   year = {2021}
}

TY  - JOUR
ID  - 1757476
AU  - Mohamed, Sameh K - Novacek, Vít - Munoz, Emir - Muñoz, Emir
PY  - 2021
TI  - On Training Knowledge Graph Embedding Models
JF  - Information
VL  - 12
IS  - 4
SP  - 147
EP  - 147
PB  - MDPI
SN  - 20782489
KW  - loss functions
KW  - knowledge graph embeddings
KW  - link prediction
UR  - https://www.mdpi.com/2078-2489/12/4/147
N2  - Knowledge graph embedding (KGE) models have become popular means for making discoveries in knowledge graphs (e.g., RDF graphs) in an efficient and scalable manner. The key to success of these models is their ability to learn low-rank vector representations for knowledge graph entities and relations. Despite the rapid development of KGE models, state-of-the-art approaches have mostly focused on new ways to represent embeddings interaction functions (i.e., scoring functions). In this paper, we argue that the choice of other training components such as the loss function, hyperparameters and negative sampling strategies can also have substantial impact on the model efficiency. This area has been rather neglected by previous works so far and our contribution is towards closing this gap by a thorough analysis of possible choices of training loss functions, hyperparameters and negative sampling techniques. We finally investigate the effects of specific choices on the scalability and accuracy of knowledge graph embedding models.
ER  -

MOHAMED, Sameh K, Vít NOVACEK a Emir MUNOZ. On Training Knowledge Graph Embedding Models. \textit{Information}. Switzerland: MDPI, 2021, roč.~12, č.~4, s.~147-165. ISSN~2078-2489. Dostupné z: https://dx.doi.org/10.3390/info12040147.

Podrobný výpis o publikaci