This is the official implementation of the paper "LR2: a Model-Agnostic Logical Reasoning Framework for Legal Case Retrieval" based on PyTorch.
Here we mainly provide the implementation of LR2-G, for LR2-D and the neural module, you can refer to "https://github.com/ke-01/NS-LCR".
-
Get the results of the neural model through training.
-
Get the results of the law-level module and case-level module.
-
Use the fusion module to get the final results.
The Dataset details is shown in dataset file
You can get the law-level and case-level results with the following instructions.
python Law-level/law_main.py
python Law-level/law_main.py --data_type elam
python Case-level/case_main.py
python Case-level/case_main.py --data_type elamRefer to Fusion folder.
You can get the LR2-G results with the following instructions.
python Fuison/fusion_main.py --data_type Lecard --base_model_path ./base_model_res/Lecard_bert_res.json
python Fuison/fusion_main.py --data_type Lecard --base_model_path ./base_model_res/Lecard_bertpli_res.json
python Fuison/fusion_main.py --data_type Lecard --base_model_path ./base_model_res/Lecard_lawformer_res.json
python Fuison/fusion_main.py --data_type Lecard --base_model_path ./base_model_res/Lecard_shaobert_res.json
python Fuison/fusion_main.py --data_type ELAM --base_model_path ./base_model_res/elam_bert_res.json
python Fuison/fusion_main.py --data_type ELAM --base_model_path ./base_model_res/elam_bertpli_res.json
python Fuison/fusion_main.py --data_type ELAM --base_model_path ./base_model_res/elam_lawformer_res.json
python Fuison/fusion_main.py --data_type ELAM --base_model_path ./base_model_res/elam_shaobert_res.jsonYou can evaluate explanations with the following instructions.
python eva_exps/eva_exps.py --data_type Lecard --eva_type logic
python eva_exps/eva_exps.py --data_type Lecard --eva_type rel
python eva_exps/eva_exps.py --data_type Lecard --eva_type flu
python eva_exps/eva_exps.py --data_type Lecard --eva_type complete
python eva_exps/eva_exps.py --data_type ELAM --eva_type logic
python eva_exps/eva_exps.py --data_type ELAM --eva_type rel
python eva_exps/eva_exps.py --data_type ELAM --eva_type flu
python eva_exps/eva_exps.py --data_type ELAM --eva_type completeWe conducted the experiments based on the following environments:
- CUDA Version: 11.4
- torch version: 2.2.0
- OS: Ubuntu 18.04.5 LTS
- GPU: NVIDIA Geforce RTX A6000
- CPU: Intel(R) Xeon(R) Silver 4214 CPU @ 2.20GHz