Papers by Peide Zhu
Unsupervised Domain Adaptation for Question Generation with DomainData Selection and Self-training (2022.findings-naacl)
Copied to clipboard
| Challenge: | Existing question generation models require large-scale and high-quality training data. |
| Approach: | They propose an unsupervised domain adaptation approach to combat the lack of training data and domain shift issue with domain data selection and self-training. |
| Outcome: | The proposed approach outperforms baselines on three large datasets with different domain similarities, using a transformer-based pre-trained QG model. |
Answer Quality Aware Aggregation for Extractive QA Crowdsourcing (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for creating extractive question answering datasets are crowdsourcing, but results are often inconsistent. |
| Approach: | They propose a method for aggregating answers from different crowd workers that takes into account the relations between the answer, question, and context passage. |
| Outcome: | The proposed method outperforms baselines by 16% on precision and effectively conduct answer aggregation for extractive question answering task. |
ControversialQA: Exploring Controversy in Question Answering (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing studies on controversy define it based on vague assumptions of its relation to sentiment . experimental results show controversy detection is essential and challenging . |
| Approach: | They propose a question-answering dataset that defines content controversy by user perception . they show controversy detection is essential and challenging . |
| Outcome: | The proposed dataset defines controversy by user perception, i.e., votes from plenty of users. |