Dual Adversarial Co-Learning for Multi-Domain Text Classification
Authors: Yuan Wu, Yuhong Guo6438-6445
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct experiments on multi-domain sentiment classification datasets. The results show the proposed approach achieves the state-of-the-art MDTC performance. We conduct experiments on two multi-domain text classification datasets. |
| Researcher Affiliation | Academia | Yuan Wu, Yuhong Guo School of Computer Science, Carleton University, Ottawa, Canada Yuan Wu3@cmail.carleton.ca, yuhong.guo@carleton.ca |
| Pseudocode | Yes | Algorithm 1 Stochastic gradient descent training algorithm |
| Open Source Code | No | The paper does not contain any explicit statements about the release of open-source code or links to a code repository for the methodology described. |
| Open Datasets | Yes | We conducted experiments on two multi-domain text classification (MDTC) datasets: the multi-domain Amazon review dataset (Blitzer, Dredze, and Pereira 2007) and the FDU-MTL dataset (Liu, Qiu, and Huang 2017). |
| Dataset Splits | Yes | For the MDTC experiment on the Amazon review dataset, we followed the experiment setting in (Wu and Huang 2015) and conducted a 5-fold cross validation test. We randomly divide data in each domain into five partitions with equal size, where three partitions are used for training, one serves as the validation set, and the remaining one is used for testing. The data in each domain is randomly divided into three parts: training set (70%), validation set (10%) and testing set (20%). |
| Hardware Specification | No | The paper does not specify the hardware used for running the experiments (e.g., CPU, GPU models). |
| Software Dependencies | No | The paper mentions 'Adam optimizer (Kingma and Ba 2014)' and 'word2vec (Mikolov et al. 2013)' but does not provide specific version numbers for any software libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | The proposed model has two hyperparameters, α and γ. In the experiments, we set α = 0.1 and γ = 0.1. We set the extracted shared feature dimension as 128 and the extracted domain-specific feature dimension as 64. C1, C2 and D are MLPs with one hidden layer containing 128+ 64, 128+64, and 128 hidden units respectively. Re LU is used as the activation function. For the experiments on Amazon review dataset, we use MLPs with two hidden layers (with 1000 and 500 units respectively) as feature extractors. The input size of the MLPs is set to 5000. For the experiments on FDU-MTL, we use CNN with one convolutional layer as our feature extractor. It uses different kernel sizes (3, 4, 5), and the number of kernels are 200. The input of the convolutional layer is the 100-dim word embeddings, obtained using word2vec (Mikolov et al. 2013), of each word in the input sequence. We use Adam optimizer (Kingma and Ba 2014) to train our models, with the learning rate 0.0001. We use a batch size of 8, and 50 training iterations. |