Construction and Evaluation of a High-Quality Corpus for Legal Intelligence Using Semiautomated Approaches

This dataset is an extended version of from the aboved paper. The dataset contains more than 500,000 legal arguments which were automatically labeled by combining the following algorithms with majority vote:

1. Fine-Tuned Legal BERT on training data involving few Paraphrased test samples. Accuracy: 0.834933 
2. Fine-Tuned Legal BERT with custom algorithm involving a corruption of few train samples. Accuracy: 0.76775 
3. Fine-Tuned Legal BERT Accuracy: 0.76199 
4. Fine-Tuned BERT Accuracy: 0.76007 
5. GAN-BERT Accuracy: 0.7480
6. Co Training Accuracy: 0.66410 
7. EM-LightGBM Accuracy: 0.6564 
8. Pseudo Labeling Accuracy: 0.6339 

There are six labels in the dataset: 1) fact, 2) issue, 3)rule/law/holding, 4) analysis, 5) conclusion/opinion/answer, and 6) others. For more details about the labels and the algorithms, please read our paper on: IEEE Transactions on Reliability