De-biased Attention Supervision for Text Classification with Causality
Authors: Yiquan Wu, Yifei Liu, Ziyu Zhao, Weiming Lu, Yating Zhang, Changlong Sun, Fei Wu, Kun Kuang
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Through extensive experiments on two professional text classification datasets (e.g., medicine and law), we demonstrate that our method achieves improved classification accuracy along with more coherent attention distributions. |
| Researcher Affiliation | Collaboration | Yiquan Wu1*, Yifei Liu2*, Ziyu Zhao1, Weiming Lu1 , Yating Zhang3, Changlong Sun3, Fei Wu1, Kun Kuang1 1 College of Computer Science and Technology, Zhejiang University, China 2 College of Software Technology, Zhejiang University, China 3 Alibaba Group, China |
| Pseudocode | Yes | Algorithm 1: The pseudocode of DAS. |
| Open Source Code | Yes | To motivate other scholars to investigate this problem, we make the code and data publicly available 2. 2https://github.com/6666ev/DAS |
| Open Datasets | Yes | Legal Verdict4. This dataset is released by Chinese AI and Law Challenge (CAIL2018) (Zhong et al. 2018), and it has been widely used in Legal AI research. Medical Triage5. This dataset collects medical conversations. The input is patients questions and the output is the corresponding department. The statistics of the two datasets are presented in Tab. 2. 4https://github.com/thunlp/CAIL 5https://github.com/liangsbin/Chinese-medical-dialogue-data |
| Dataset Splits | Yes | To ensure fair evaluations, we partition each dataset randomly into training, validation, and test sets, maintaining an 80%:10%:10% ratio. |
| Hardware Specification | Yes | We conducted our experiments using two V100 GPUs. |
| Software Dependencies | No | The paper mentions using 'Gensim' but does not provide specific version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | The size of the keyword vocabulary is set to 1000. The setting for λ is 0.15. |