DARE: Disentanglement-Augmented Rationale Extraction
Authors: Linan Yue, Qi Liu, Yichao Du, Yanqing An, Li Wang, Enhong Chen
NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results on three real-world datasets and simulation studies clearly validate the effectiveness of our proposed method. Code is released at https://github.com/yuelinan/DARE. and In this section, we first compare DARE with some baselines on a beer reviews dataset, a movie reviews dataset, and a legal judgment prediction dataset. |
| Researcher Affiliation | Collaboration | Linan Yue1,2, Qi Liu1,2 , Yichao Du1,2, Yanqing An1,2, Li Wang1,2,3, Enhong Chen1,2 1: Anhui Province Key Laboratory of Big Data Analysis and Application, University of Science and Technology of China 2: State Key Laboratory of Cognitive Intelligence 3: Byte Dance |
| Pseudocode | Yes | Algorithm 1 MI Minimization with CLUB_NCE |
| Open Source Code | Yes | Code is released at https://github.com/yuelinan/DARE. |
| Open Datasets | Yes | We use the Beer Advocate [29] dataset containing more than 220,000 beer reviews as our dataset., We also make experiments on a movie review dataset [48] of the ERASER benchmark [15], We conduct our experiments on publicly available datasets of the Chinese AI and Law challenge CAIL2018 [41]. and Although the original Beer Advocate has been removed by the authors of this dataset, we can obtain Beer Advocate from http://people.csail.mit.edu/taolei/beer/, which is published by [24]. CAIL2018 has been published in https://github.com/ china-ai-law-challenge/CAIL2018. |
| Dataset Splits | No | The paper mentions a test set for the Beer Advocate dataset ('we take 994 reviews for three aspects as our test set') and discusses 'training' and 'testing', but it does not explicitly specify the proportions or counts for train/validation/test splits, nor does it describe a cross-validation setup or reference a predefined split with specific details for all three subsets. |
| Hardware Specification | Yes | We run all experiments on a single V100 GPU. |
| Software Dependencies | No | The paper mentions using RCNN as the architecture and refers to various methods, but it does not provide specific version numbers for any software dependencies like programming languages or libraries (e.g., 'Python 3.8', 'PyTorch 1.9'). |
| Experiment Setup | Yes | For a fair comparison, we set the lr as {0.13, 0.07, 0.07} in three aspect datasets, respectively, and initialize the epoch as 50, which are both consistent with the previous methods [4, 37]. |