SCU-MESCLab at ROCLING 2023 ''MultiNER-Health'' Task : Named Entity Recognition Using Multiple Classifier Model

SCU-MESCLab at ROCLING 2023 ''MultiNER-Health'' Task : Named Entity Recognition Using Multiple Classifier Model

本研究旨在命名實體識別模型的任務上，設計出多重分類模型，並運用在醫學領域。其中訓練資料以BIO格式進行標記，我們採用不同組合的模型進行選擇，以六種模型進行評估及批選，最後篩選出最佳三種模型分別為BERT-based NER Model、RoBERTa (base)-based NER Model，以及RoBERTa (large) + BiLSTM + CRF Model，並應用於多重分類模型，使RUN1獲得最佳預測，其平均精確度為68.69%、平均召回率為67.64%，平均F1-score為68.13。

This study aims to design a multi-class classification Model for the task of named entity recognition and apply it in the medical field. The training data is labeled in the BIO format. We employed various combinations of Models for selection, evaluating, and choosing among six Models. Finally, we identified the top three Models: a BERT-based NER Model, a RoBERTa (base)-based NER Model, and a RoBERTa (large) + BiLSTM + CRF Model. These Models were applied to a multi-class classification setup, with RUNl achieving the best predictions. The average precision for RUNl is 68.69%, the average recall is 67.64%, and the average Fl score is 68.13%.