Weak Supervision Enhanced Generative Network for Question Generation

Authors: Yutong Wang, Jiyuan Zheng, Qijiong Liu, Zhou Zhao, Jun Xiao, Yueting Zhuang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct detailed experiments to demonstrate the comparative performance of our approach. 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Baselines 4.4 Automatic Evaluation 4.5 Human Evaluation
Researcher Affiliation Academia Yutong Wang1 , Jiyuan Zheng1 , Qijiong Liu1 , Zhou Zhao1 , Jun Xiao1 and Yueting Zhuang1 1College of Computer Science and Technology, Zhejiang University, China {ytwang, jiyuanz, lqj, zhaozhou, yzhuang}@zju.edu.cn, junx@cs.zju.edu.cn
Pseudocode No The paper describes the model architecture and processes in text and with diagrams, but does not include any explicit pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about releasing source code or links to a code repository.
Open Datasets Yes The MS MARCO dataset is a large scale dataset collected from Bing. This dataset contains 1,010,916 questions and 8,841,823 related passages extracted from 3,563,535 web documents. ... The SQu AD dataset [Rajpurkar et al., 2016] is one of the most influential reading comprehension datasets which contains over 100k questions of 536 Wikipedia articles created by crowd-workers and the answers are continuous spans in the passages.
Dataset Splits Yes The whole dataset(train set and dev set) is randomly divided into a training set (80%), a development set (10%) and a test set (10%) at the article level.
Hardware Specification No The paper does not provide any specific hardware details (e.g., GPU/CPU models, memory specifications) used for running the experiments.
Software Dependencies No The paper mentions using GloVe pretrained embeddings and refers to standard components like Adam optimizer and Multi-head Attention, but does not provide specific version numbers for software dependencies such as deep learning frameworks or libraries.
Experiment Setup Yes For model hyperparameters, we have 4 convolutional filters with kernel size 7 for all convolutional blocks and all the self attention blocks are Multi-head Attention [Vaswani et al., 2017] with 8 attention heads. We adopt Adam optimizer with a learning rate of 0.001, β1 = 0.9 and β2 = 0.999 and batch size is set to 16. we train the model for a maximum of 20 epochs and use early stopping with the patience set to 5 epochs according to the BLEU-4 score on validation set. At the validation and test time, we use beam search with beam size set to 5.