Re17: Read the paper Challenges for Information Extraction from Dialogue in Criminal Law

Gods are silent-personal CSDN blog directory

Paper name: Challenges for Information Extraction from Dialogue in Criminal Law
Paper ACL official download address: https://aclanthology.org/2021.nlp4posimpact-1.8/

This article is a 2021 ACL paper, tasked with extracting factual information (11 hand-picked features) from the hearing transcript text, testing unsupervised methods, weakly supervised methods, and methods using pre-trained modelseffect on this task.
The data set is self-made data, it is a free-form dialogue of California parole hearings, part of the data is annotated with 11 features.

The algorithm used in this article is:

1. an unsupervised data programming paradigm extended to weak supervision: Unsupervised Snorkel, Supervised WSLF (Logistic Regression)
2. pretrained question answering models based on DistilBERT and Longformer: QA1-2
3. classification models based on BERT each fine-tuned to predict a single task: Task-FT

Date and numeric are binned when the F1 value is calculated.