fmLRE: A Low-Resource Relation Extraction Model Based on Feature Mapping Similarity Calculation
Authors: Peng Wang, Tong Shao, Ke Ji, Guozheng Li, Wenjun Ke
AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results demonstrate that fm LRE achieves the state-of-the-art performance compared with strong baselines on two public datasets. and Experiments Datasets and Settings To evaluate the proposed fm LRE model, we select two public datasets to conduct experiments: Sem Eval 2010 Task 8 dataset (Hendrickx et al. 2010) and the TACRED dataset (Zhang et al. 2017). |
| Researcher Affiliation | Academia | Peng Wang1* , Tong Shao1 , Ke Ji1, Guozheng Li1, Wenjun Ke1, 2 1Southeast University 2Beijing Institute of Computer Technology and Application pwang@seu.edu.cn, 22S151082@stu.hit.edu.cn, {keji, gzli}@seu.edu.cn, kewenjun2191@163.com |
| Pseudocode | Yes | Algorithm 1: Filtering mechanism |
| Open Source Code | Yes | The code of fm LRE and the datasets can be accessed via https://github.com/seukgcode/fm LRE. |
| Open Datasets | Yes | Considering the low-resource scenario, fm LRE only splits a small ratio of the data as labeled dataset, and the rest is used as the unlabeled dataset, which follows common experimental setups: 5%, 10%, and 30% of the data are split into labeled data on Sem Eval, and the rest is used as unlabeled data. On the TACRED dataset, 3%, 10%, and 15% of the data are split as labeled data. |
| Dataset Splits | No | The paper describes training and testing splits (labeled/unlabeled and Test sets) but does not explicitly mention a separate validation set or its split percentage. |
| Hardware Specification | No | The paper does not specify the hardware used for running the experiments, such as specific GPU or CPU models. |
| Software Dependencies | No | The paper mentions using the BERT model and MTB model, but does not specify software dependencies like programming language versions or specific library versions. |
| Experiment Setup | Yes | In the implementation of fm LRE, the max length of the sample sentences is 128, the batch size for training and division of unlabeled data is 32, and the initial learning rate of the model is 5e-5. In addition, there are some hyperparameters: the threshold θ of similarity is set to 0.7, the constraint range of reward is set to 0.2, and the number of filtering for noisy data n is set to 2. |