Low-Resource NER by Data Augmentation With Prompting

Authors: Jian Liu, Yufeng Chen, Jinan Xu

IJCAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results have widely confirmed the effectiveness of our approach.
Researcher Affiliation Academia Jian Liu , Yufeng Chen and Jinan Xu Beijing Key Lab of Traffic Data Analysis and Mining Beijing Jiaotong University, School of Computer and Information Technology, China jianliu@bjtu.edu.cn, chenyf@bjtu.edu.cn, jaxu@bjtu.edu.cn
Pseudocode No The paper describes its approach using descriptive text and mathematical equations (e.g., equations 1-7) but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes We have made our code available at https: //github.com/jianliu-ml/few NER for further investigation.
Open Datasets Yes We use three NER datasets for evaluation: Co NLL 2003 [Tjong Kim Sang and De Meulder, 2003], Onto Notes 5.0 [Hovy et al., 2006], and a real-world low-resource dataset, Ma Scip [Mysore et al., 2019].
Dataset Splits Yes for each dataset we sample out 50, 150, and 500 sentences (at least one mention of each entity type is included) to create the small (S), medium (M), and large (L) training sets (we use F to indicate the full training set), and we use precision (P), recall (R), and F1 as evaluation metrics. Table 2 gives statistics of the three datasets. ... We use the development set to tune the best iteration step.
Hardware Specification No The paper mentions using 'BERT-base cased version' as the backbone and discusses model architectures, but does not provide specific hardware details such as GPU or CPU models used for the experiments.
Software Dependencies No The paper mentions using 'BERT-base cased version', 'Bi LSTM-CRF', 'Glo Ve embeddings', and 'Adam' but does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes In label-conditioned word replacement, we empirically set T = 10... In the uncertainty-guided self-training method, we set the number of forward passes K to 10... and select N = 200... We set the batch size to 50... and the learning rate to 1e-2... The batch size is set to 10... and the learning rate is set to 1e-5... We apply Adam [Kingma and Ba, 2015] for model optimization.