Graph-Evolving Meta-Learning for Low-Resource Medical Dialogue Generation
Authors: Shuai Lin, Pan Zhou, Xiaodan Liang, Jianheng Tang, Ruihui Zhao, Ziliang Chen, Liang Lin13362-13370
AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiment results on the CMDD dataset and our newly-collected Chunyu dataset testify the superiority of our approach over state-of-the-art approaches. |
| Researcher Affiliation | Collaboration | Shuai Lin1, Pan Zhou2, Xiaodan Liang1,3*, Jianheng Tang1, Ruihui Zhao4, Ziliang Chen1, and Liang Lin1,3 1Shenzhen Campus of Sun Yat-sen University, 2Salesforce Research, 3Dark Matter AI Inc., 4Tencent Jarvis Lab |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described, nor does it explicitly state that the code is released. |
| Open Datasets | Yes | Here we conduct extensive experiments on the CMDD dataset (Lin et al. 2019) and the newly-collected Chunyu dataset to demonstrate the benefits of GEML. ... The raw data of Chunyu is obtained from the Gastroenterology department of the Chinese online health community Chunyu3. It contains 15 gastrointestinal diseases and 62 symptoms in total. ... 3https://www.chunyuyisheng.com/ |
| Dataset Splits | Yes | For Chunyu, as shown in Fig.1, high-resource diseases with more than 500 training instances are treated as source diseases and the remaining four low-resource ones as target diseases, whose size of adaptation data is ranging from 80 200. For CMDD, we adopt the standard leave-one-out setup, i.e., using four diseases for meta-training and the one target disease left for adaptation (with the data size of 150 dialogues). ... Each task T k i T has only a few dialogue samples, which can be further split into the training (support) set DTi tr and the validation (query) set DTi va. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments. |
| Software Dependencies | No | The paper mentions using the Allen NLP toolkit and pkuseg toolkit, but it does not specify version numbers for these or any other software dependencies needed to replicate the experiment. |
| Experiment Setup | Yes | We set both dimensions of the hidden state and word embedding to 300 for LSTM. Adam optimization is adopted with the initial learning rate of 0.005 and the mini-batch size of 16. The maximum training epochs are set to 100 with the patience epoch of 10 for early-stopping. The best hyperparameter λ to balance the generation loss and the entity loss is 8. |