LER

Dataset and source code for AAAI 2023 paper "Unsupervised Legal Evidence Retrieval via Contrastive Learning with Approximated Positive"

Overview

Dataset

The full dataset can be downloaded via Google Drive. You can also check the data_sample folder for a glimpse of our dataset.

Code

• Prepare the dataset
  • Download data.zip, unzip to get /data folder, and put it in the root path (LER)
  • Leave the file names unchanged (consistent with the config files in LER/config folder)
  • The directory should look like this:

    ├── config
    │   ├── ...
    │   └── ...
    ├── data
    │   ├── test
    │   │   ├── test_dev-set-200.json
    │   │   └── test_test-set-719.json
    │   └── train
    │       └── train_all_record-wo-test.jsonl
    ...
    

• Check the config folder for different experiment settings.
• Use train.sh and test.sh to train and evaluate the models.