Multi-Labeled Relation Extraction with Attentive Capsule Network

Authors: Xinsong Zhang, Pengshuai Li, Weijia Jia, Hai Zhao7484-7491

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results show that the proposed approach can indeed extract the highly overlapped features and achieve significant performance improvement for relation extraction comparing to the state-of-the-art works.
Researcher Affiliation Academia Xinsong Zhang,1 Pengshuai Li,1 Weijia Jia,2,1 Hai Zhao1 1Department of Computer Science and Engineering, Shanghai Jiao Tong University, Shanghai 200240, China 2State Key Lab of Io T for Smart City, University of Macau, Macau 999078, China {xszhang0320, pengshuai.li}@sjtu.edu.cn and {jia-wj, zhaohai}@cs.sjtu.edu.cn
Pseudocode Yes Algorithm 1 Attention-based Routing Algorithm
Open Source Code No No explicit statement or link providing access to the source code for the methodology described in this paper.
Open Datasets Yes We conduct experiments on two widely used benchmarks for relation extraction, NYT-10 (Riedel, Yao, and Mc Callum 2010) and Sem Eval-2010 Task 8 dataset (Hendrickx et al. 2009). The NYT-10 dataset is generated by aligning Freebase relations with the New York Times (NYT) corpus, in which sentences from the years 2005-2006 are for training while those from 2017 for testing. The dataset consists of amounts of multi-labeled sentences. The Sem Eval-2010 Task 8 dataset is a small dataset which has been well-labeled for relation extraction. The details of both datasets are shown in Table 1.
Dataset Splits No The paper does not explicitly provide validation dataset splits. While it mentions train/test splits and 10-fold cross-validation in evaluation, specific validation set details are missing.
Hardware Specification No No specific hardware details (like GPU/CPU models or memory) are provided for running the experiments.
Software Dependencies No Software components like word2vec and Adam optimizer are mentioned, but specific version numbers for any libraries or frameworks are not provided.
Experiment Setup Yes Table 2 lists our hyper-parameter setting. Parameters NYT-10 Sem. batch size 50 50 word dimension p 50 50 position dimension q 5 5 hidden state dimension sh 256 256 capsule dimensions [du, dr] [16,16] [16,16] iterations z 3 3 sliding-margin γ 0.4 0.4 down-weighting λ 1.0 0.5 learning rate 0.001 0.001 dropout probability 0.0 0.7 L2 regularization strength 0.0001 0.0