Research And Implementation Of Question Answering System For Medical Field

We conducted a series of researches on medical field question answering algorithms in this subject and proposed a BAN-BiGRU-CNN question answering model based on Bridge Architecture Network(BAN)and Negative Sampling in Batch(NSB)model.At the same time,we proposed a Dynamic Multi Answer Extraction(DMAE)model to further mine the performance of the model.The main work of this paper is divided into the following four points:(1)Aiming at the certain gap between the pre-trained model and the question-answering task,we proposed a Bridge Architecture Network(BAN),and using BAN,we designed a question answering model based on BAN-BiGRU-CNN which models the sequence through BiGRU and extracts fractional feature using CNN.Experiments show that on the cMedQA validation set based on the Hit@1 indicator,using different pretrained models with BAN-BiGRU-CNN improves the respective baseline indicators by 0.84%-3.63%,and the training time is reduced by an average of 46.29%.(2)Aiming at the problem that the question answering corpus training takes a long time and leads to a long iteration cycle,we proposed an NSB negative sampling in the batch question-answering model.It makes full use of the advantages of reconstructed training data and batches negative sampling to improve the accuracy and greatly reduce the training time.Experiments show that the training time is reduced by 93.90%on average after optimization of different models.At the same time,aiming at the problem that the calculation time of the loss function is too high,we proposed a loss calculation parallelization optimization scheme to further optimize the model training time at the loss calculation level.(3)Aiming at the insufficient performance of the traditional question answering model,the DMAE model is proposed and designed.DMAE abandons the traditional model evaluation method of selecting the same number of answers for each question,and can dynamically determine the answer quantity according to the difference in the question-answering evaluation score.Experiments show that DMAE can recall 42.22%of the wrong answers to the questions judged by the model.(4)Design and implement a prototype medical field questionanswering system,which is based on the BAN-NSB question answering model and the DMAE model for question answering modeling.Using the clustering algorithm and the method of recall+sorting retrieval strategy to realize automatic intelligent medical question answering.