GDPNet: Refining Latent Multi-View Graph for Relation Extraction

Authors: Fuzhao Xue, Aixin Sun, Hao Zhang, Eng Siong Chng14194-14202

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental On Dialog RE and TACRED, we show that GDPNet achieves the best performance on dialogue-level RE, and comparable performance with the state-of-the-arts on sentence-level RE.
Researcher Affiliation Academia 1 School of Computer Science and Engineering, Nanyang Technological University, Singapore 2 Institute of High Performance Computing, Agency for Science, Technology and Research, Singapore
Pseudocode No No pseudocode or algorithm blocks were found in the paper.
Open Source Code Yes Our code is available at https://github.com/Xue Fuzhao/GDPNet.
Open Datasets Yes On Dialog RE and TACRED, we show that GDPNet achieves the best performance on dialogue-level RE, and comparable performance with the state-of-the-arts on sentence-level RE.
Dataset Splits Yes Model Dev set Test set F1 (σ) F1c (σ) F1 (σ) F1c (σ)
Hardware Specification Yes All experiments are conducted on a desktop with Intel i7-8750H CPU, DDR4 16GB memory, and a single NVIDIA Ge Force RTX 1070 GPU. We also reproduced our results on Quadro RTX 8000 GPU.
Software Dependencies Yes The GDPNet is implemented by using Py Torch 1.4 with CUDA 10.1. Our implementation also uses the Soft DTW2 toolkit.
Experiment Setup Yes The hyper-parameter settings on the two datasets, Dialog RE and TACRED, are listed as follows. Parameter Dialog RE TACRED Epoch 20 10 Batch Size 24 32 Learning rate 3e-5 2e-5 Dropout 0.5 0.5 Hidden units of graph 300 300 Number of views 3 3 Number of DTWPool layers 3 3 Pooling ratio lower bound 0.7 0.8 Weight of Soft DTW loss 1e-6 2e-4