1 GAT_GCN_code.ipynb —— 基于GAT和GCN模型的司法判决预测算法
model文件保存GAT和GCN的best_model_path
aux_files文件定义停用词文件的路径STOPWORDS_PATH
1.1 基于GAT模型的司法判决预测算法
F1 in test: 0.828065190694157
precision in test: 0.8405425486020889
recall in test: 0.834963701242771
accuracy in test: 0.834963701242771
1.2 基于GCN模型的司法判决预测算法
F1 in test: 0.7988395395765523
precision in test: 0.8130721153032072
recall in test: 0.8079242032730405
accuracy in test: 0.8079242032730405
2 RoBERTa+DPR_code.ipynb —— 基于RoBERTa和DPR模型的司法判决预测算法
验证集评估结果: {'eval_loss': 12.467425346374512, 'eval_runtime': 200.9035, 'eval_samples_per_second': 85.27, 'eval_steps_per_second': 5.331, 'epoch': 10.0}
预测准确率：0.8871662360034454
Macro F1: 0.8271501624283938
Micro F1: 0.8871662360034454
3 数据集data
详见 https://github.com/china-ai-law-challenge/CAIL2018