大模型的药物不良反应实体识别能力评估OA

This study focused on the named entity recognition task in the field of adverse drug reactions and constructed a Chinese dataset named Chinese named entity recognition pharmacovigilance dataset(CNER-PH).Based on this dataset,mainstream named entity recognition(NER)models and large language models were systematically evaluated.The methods used included several pre-trained models such as bidirectional encoder representations from transformers(BERT)series and robustly optimized BERT pretrain-ing approach(RoBERTa),as well as large language models such as GPT-4o and DeepSeek.The analysis emphasized their recogni-tion ability under conditions of diverse entity types and unclear entity boundaries.The experimental results show that BERT series models are more stable in handling complex entity boundaries,and their overall F1 scores are higher than those of large language models.The findings reveal the strengths and weaknesses of different models in adverse drug reaction recognition and provide a ba-sis for future methodological improvements.