Bert-like Models feasibility Exploration on Medical Short-Text Q&A Tasks

Explored the feasibility of Bert-like models for machine reading comprehension of small text in specialized areas; selected and blended different classes of Bert-like models to pursue better performance on small text data obtained from medical illustration videos; attempted to train field-specific MRC model for medical Q&A tasks, tuning parameters; tested and evaluated the performance; NLP skillsets increased by interacting with supervisor and Ph.D. students; attained a deeper understanding of DL & RL formulas; research contributed to Ningbo 2025 Sci & Tech Innovation Program.

预测步骤：

关黄母颗粒问答预测 sh test_bert_ghm.sh 丁苯酚问答预测 sh test_bert_dbf.sh 执行此脚本时，按照metrics.py 730行 提示 更改存放地址

参数：

--lm: 要加载的模型的文件夹名称 --do_train: 开启训练 --evaluate_during_training: 开启训练时的验证 --do_test: 开启预测 --version_2_with_negative: 开启适配于数据中有无答案数据 --threads: 数据处理所使用的线程数

Prediction steps.

关黄母 granules quiz prediction sh test_bert_ghm.sh butanol quiz prediction sh test_bert_dbf.sh When executing this script, follow metrics.py line 730 Prompt Change storage address

Parameters.

--lm: the name of the folder where the model is to be loaded --do_train: turn on training --evaluate_during_training: turn on validation during training --do_test: turn on prediction --version_2_with_negative: turn on adaptation to data with or without answer data --threads : the number of threads used for data processing