BROWSE
Detail
Journal Article
QLite: Lightweight Knowledge Graph Embedding Framework with Query Processing
- Authors
- Chun-Hee Lee, Dong-Oh Kang
- Issue Date
- 2025-12
- Citation
- IEEE Access, v.13, pp.217493-217503
- ISSN
- 2169-3536
- Publisher
- IEEE
- Language
- English
- Type
- Journal Article
- Abstract
- A vast number of studies on knowledge graph embedding have been conducted. However, most knowledge graph embedding models have high dimensional embedding vectors. To use the models in embedded systems such as mobile devices or robots, we need to reduce the size of the embedding models. Although there are several approaches to handle the lightweight knowledge graph embedding problem, the existing approaches have drawbacks while processing actual tasks (i.e., queries). For instance, to process queries in the embedded systems, they require a full scan of the entity embedding vectors. To overcome the drawbacks, we propose a lightweight knowledge graph embedding framework, called QLite, to simultaneously consider three factors: Model Accuracy, Model Space, and Query Processing Time. QLite provides simple methods to effectively reduce the models’ size without decoding. Moreover, QLite adopts a reordering module to avoid the full linear scan of the entities during query processing. We focus on TransE to show if the reordering module is effective. Finally, we experimentally show the efficiency and the effectiveness of QLite.
- Keyword
- Knowledge graph embedding, lightweight model, query processing, reordering
- KSP Keywords
- Embedding model, Embedding problem, Existing Approaches, High-dimensional, Knowledge graph embedding, Lightweight model, Linear scan, Mobile devices, Model accuracy, Model space, Query processing
This work is distributed under the term of Creative Commons License (CCL)
(CC BY)
(CC BY)
