Papers by Roman Kern
Causal Investigation of Public Opinion during the COVID-19 Pandemic via Social Media Text (2022.lrec-1)
Copied to clipboard
| Challenge: | Social distancing orders are the most effective strategy to reduce the spread of COVID-19 in 2020 . |
| Approach: | They propose to use NLP methods in a causal mediation scenario to emphasize the use of NLP and economics to decouple the effect of government restrictions on mobility from the effect that occurs due to public perception of the COVID-19 strategy. |
| Outcome: | The proposed model decouples the effect of government restrictions on mobility behavior from the effect that occurs due to public perception of the COVID-19 strategy in a country. |
Impact of Training Instance Selection on Domain-Specific Entity Extraction using BERT (2022.naacl-srw)
Copied to clipboard
| Challenge: | Named entity recognition (NER) tasks require a large number of training examples and handcrafted features. |
| Approach: | They propose to fine-tune pre-trained language models such as BERT to achieve up to 80% F1 when fine- tuned on only 70 training examples. |
| Outcome: | The proposed model achieves 80% F1 when fine-tuned on only 70 training examples, especially on biomedical domain. |
Ensemble Watermarks for Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Existing watermarks for large language models lack flexibility and struggle with attacks such as paraphrasing. |
| Approach: | They propose a multi-feature method for generating watermarks that combines multiple distinct watermark features into an ensemble watermark. |
| Outcome: | The proposed method achieves 98% detection rate after paraphrasing attack and 49% after paraphrase attack. |