Entity-Agnostic Representation Learning for Parameter-Efficient Knowledge Graph Embedding
Authors: Mingyang Chen, Wen Zhang, Zhen Yao, Yushan Zhu, Yang Gao, Jeff Z. Pan, Huajun Chen
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results show that EARL uses fewer parameters and performs better on link prediction tasks than baselines, reflecting its parameter efficiency. |
| Researcher Affiliation | Collaboration | 1College of Computer Science and Technology, Zhejiang University 2School of Software Technology, Zhejiang University 3Donghai Laboratory 4Huawei Technologies Co., Ltd. 5School of Informatics, The University of Edinburgh 6Alibaba-Zhejiang University Joint Institute of Frontier Technologies |
| Pseudocode | No | The paper describes methods verbally and through equations but does not include any explicitly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code is available at https://github.com/zjukg/EARL. |
| Open Datasets | Yes | Our model is evaluated on several KG benchmarks with various sizes and characteristics, and the dataset statistics are shown in Table 1. Specifically, FB15k-237 (Toutanova et al. 2015) is derived from Freebase (Bollacker et al. 2008)... WN18RR (Dettmers et al. 2018) is a subset of Word Net (Miller 1995)... Co DEx (Safavi and Koutra 2020)... YAGO3-10 (Mahdisoltani, Biega, and Suchanek 2015)... |
| Dataset Splits | Yes | Table 1: Dataset statistics. The number of entities, relations, training triples, validation triples, and test triples. |
| Hardware Specification | Yes | We conduct our experiments on NVIDIA RTX 3090 GPUs with 24GB RAM |
| Software Dependencies | No | The paper mentions using PyTorch and DGL, but does not provide specific version numbers for these software dependencies. It only cites their original papers. |
| Experiment Setup | Yes | For entity-agnostic encoding, we use 2-layer GNNs, and the default number of k for k NRes Ent encoding is 10. We set the number of reserved entities as 10% of the number of all entities for each dataset... For model training, the learning rate is set to 0.001; the batch size is set to 1024; the number of negative samples (i.e., n) is set to 256; the margin is set to 15 for YAGO3-10 and 10 for other datasets. |