Knowledge Graph Completion Model Based on Con-distillation

DOI

10.2991/978-94-6463-490-7_61How to use a DOI?

Keywords

Transformer; Knowledge Graph; Bidirectional Encoder Representation from Transformers

Abstract

To improve the ability of the knowledge graph completion model to predict long-tail entities and keep its prediction indices for popular entities, this paper proposes a knowledge graph completion model based on co-knowledge distillation. Overall, the model consists of three parts as shown below. In detail, the first part is the neighborhood-information-based transformer model (NT), which aims to learn the general representation of the current knowledge graph. In this regard, NT takes the first-order subgraph of the knowledge graph as input to learn the representation of entities from neighboring nodes, subsequently learning the task target of link prediction through masking training. The second part is the relational-path-based bidirectional encoder representations from transformers (BERT) model, hereinafter the PB model, which expands the retrieval range for long-tail entity information by learning multiple relational paths of triples, thereby improving the characterization ability of long-tail entities. Lastly, the proposed model enables the two foregoing models to learn from each other in a lightweight co-knowledge distillation way, so that their prediction ability for popular entities and long-tail entities can be improved simultaneously.

Copyright

© 2024 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Min Xia
AU  - Tao Zhang
AU  - Zhonghai Wu
PY  - 2024
DA  - 2024/08/31
TI  - Knowledge Graph Completion Model Based on Con-distillation
BT  - Proceedings of the 2024 3rd International Conference on Artificial Intelligence, Internet and Digital Economy (ICAID 2024)
PB  - Atlantis Press
SP  - 570
EP  - 582
SN  - 2589-4919
UR  - https://doi.org/10.2991/978-94-6463-490-7_61
DO  - 10.2991/978-94-6463-490-7_61
ID  - Xia2024
ER  -