Improving Neural Relation Extraction with Positive and Unlabeled Learning
Authors: Zhengqiu He, Wenliang Chen, Yuyi Wang, Wei Zhang, Guanchun Wang, Min Zhang7927-7934
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on a widely used real-world dataset demonstrate that this new approach indeed achieves significant and consistent improvements as compared to several competitive baselines. We design experiments to demonstrate the effectiveness of our system. The results on a widely used dataset indicate that our proposed model significantly outperforms the systems using RL-based instance selector in the previous studies. |
| Researcher Affiliation | Collaboration | Zhengqiu He, Wenliang Chen, Yuyi Wang, Wei Zhang, Guanchun Wang, Min Zhang Institute of Artificial Intelligence, School of Computer Science and Technology, Soochow University, China Alibaba Group, ETH Zurich, Laiye Network technology co.LTD zqhe@stu.suda.edu.cn, {wlchen, minzhang}@suda.edu.cn lantu.zw@alibaba-inc.com, yuwang@ethz.ch, arvid@laiye.com |
| Pseudocode | No | The paper describes the approach using text and mathematical equations, but it does not include any explicit pseudocode blocks, algorithm figures, or sections labeled 'Algorithm' or 'Pseudocode'. |
| Open Source Code | No | The paper does not include any statements or links indicating that the source code for the described methodology is publicly available, in supplementary materials, or provided via an anonymous link. |
| Open Datasets | Yes | We evaluate our model on a widely used benchmark dataset developed by Riedel, Yao, and Mc Callum (2010), which has also been used in many recent studies (Lin et al. 2016; Ji et al. 2017). The dataset contains 53 relation types, including a special relation NA standing for no relation between the entity pair. |
| Dataset Splits | Yes | The training data contains 522,611 sentences, 281,270 entity pairs and 18,252 relational facts. The testing set contains 172,448 sentences, 96,678 entity pairs and 1,950 relational facts. tune other hyperparameters via three-fold cross validation on training data. |
| Hardware Specification | No | The paper does not explicitly describe the specific hardware used for running the experiments, such as GPU models, CPU models, or details of cloud computing resources. |
| Software Dependencies | No | The paper mentions using 'Adam' for optimization and 'skip-gram method' for pretraining, but it does not provide specific version numbers for any software dependencies, libraries, or programming languages (e.g., Python 3.x, PyTorch 1.x). |
| Experiment Setup | Yes | After tuning, we choose the following settings in our experiments: the dimension of word embedding da is set to 50, the dimension of position embedding db is set to 5, the number of filters K is set to 230, and the window size l of filters is 3. The batch size is fixed to 50, the dropout probability is set to 0.5. And the combination weight α is set to 0.7, β is set to 0.1. When training, we apply Adam (Kingma and Ba 2014) to optimize parameters, and the learning rate is set to 0.001. |