step1: run preprocess.py step2: { run auxiliary_trigger.py; run dominant_trigger.py; run argument.py; } step3: run joint_predict.py
核心思路: 以RoBERTa_Large作为Context Encoder, 将问题拆解为触发词识别和触发词对应四论元检测 触发词识别和论元识别基于MRC(Span Extraction类机器阅读理解)思路实现
参考论文: ACL2020 《A Unified MRC Framework for Named Entity Recognition》
single model score: 0.78