Distribution Matching for Rationalization
Authors: Yongfeng Huang, Yujun Chen, Yulun Du, Zhilin Yang13090-13097
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirically, the proposed distribution matching approach consistently outperforms previous methods by a large margin. Our data and code are available1. Experiments Datasets To evaluate the performance of our DMR framework, we use the multi-aspect beer and hotel datasets, which are commonly used in the field of rationalization. |
| Researcher Affiliation | Collaboration | Yongfeng Huang2*, Yujun Chen1, Yulun Du1, Zhilin Yang1 1Recurrent AI, Beijing 2Tsinghua University, Beijing huangyf17@tsinghua.org.cn, chenyujun@rcrai.com, duyulun@rcrai.com, kimi yang@rcrai.com |
| Pseudocode | No | The paper describes the proposed method and framework using narrative text and a figure, but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our data and code are available1. 1https://github.com/kochsnow/distribution-matching-rationality |
| Open Datasets | Yes | We evaluate the proposed distribution matching approach on widely-used rationalization benchmarks the beer review dataset (Mc Auley, Leskovec, and Jurafsky 2012) and hotel review dataset (Bao et al. 2018). |
| Dataset Splits | No | The paper mentions 'train and validate the models using the same beer review dataset but with different data split and processing' and 'We train our models using a balanced training dataset as in (Chang et al. 2019)'. However, it does not explicitly provide the specific percentages, sample counts, or detailed methodology for these splits within the paper's text. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, memory, or cloud computing resources used for the experiments. |
| Software Dependencies | No | The paper mentions general model architectures and techniques (e.g., 'recurrent neural networks', 'Transformers', 'CMD', 'teacher-student distillation') but does not specify any software dependencies with version numbers (e.g., Python, TensorFlow, PyTorch versions). |
| Experiment Setup | Yes | Following previous work (Chang et al. 2019), the hidden unit size and the embedding dimension of the teacher discriminator are set as 100, while those of the generators and the student discriminator are set as 102 for two extra class label dimensions. |