Sequence Generation with Label Augmentation for Relation Extraction

Authors: Bo Li, Dingyao Yu, Wei Ye, Jinglei Zhang, Shikun Zhang

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results show that RELA achieves competitive results compared with previous methods on four RE datasets.
Researcher Affiliation Academia 1National Engineering Research Center for Software Engineering, Peking University 2School of Software and Microelectronics, Peking University deepblue.lb@stu.pku.edu.cn, yudingyao@pku.edu.cn, wye@pku.edu.cn, jinglei.zhang@stu.pku.edu.cn, zhangsk@pku.edu.cn
Pseudocode No The paper describes its methods in prose and with supporting diagrams (e.g., Figure 2) but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes Code is available at https://github.com/pkuserc/RELA
Open Datasets Yes We evaluate our method on four RE datasets, i.e., TACRED (Zhang et al. 2017), Sem Eval2010 (Hendrickx et al. 2010), Google RE,7 and sci ERC (Luan et al. 2018).
Dataset Splits Yes Dataset #Train #Dev #Test #Class TACRED 68,124 22,613 15,509 42 Sem Eval 8,000 2,712 10 Google RE 38,112 9,648 9,616 5 sci ERC 3,219 455 974 7
Hardware Specification Yes We use Pytorch (Paszke et al. 2019) and Tesla T4 GPU with a batch size of 8.
Software Dependencies No The paper mentions software like 'Pytorch' and 'BART-large' and 'GPT-2' but does not provide specific version numbers for these software components. For example, it states 'We use Pytorch (Paszke et al. 2019)' but doesn't specify the PyTorch version number.
Experiment Setup Yes The maximum input/output sequence length is 256/32 tokens. As an optimiser, we used AdamW (Loshchilov and Hutter 2019) with a 1e-5 learning rate and a 0.2 warmup ratio. The training epoch is 20 for Semeval and 10 for other datasets.