الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 90 |  |  |
| | | from | 90 | | |
Abstract
Machine Reading Comprehension(MRC)is a task which evaluates the machine understanding of the natural languages.The MRC researching the medical eld has not received signi cant attention despiteits importance in practical applications.This work is concerned with the cloze- style problem which is one of the tasks of machine comprehension for the medical domain. A lot of research has been done to improve the performance of MRC starting from simple model to complex attention-based deep learning models.Recent deep learning model shav eshown promising results for the general domain.However these models were not evaluated on the medical do main andmost recent work didn’tconcentrate on the medical domain. In this work,anewdeep learning model
for MRC in the medical domain is proposed.It uses convoultionneural network(CNN),inaddition to self attention and gated attention mechanisms.It also incorporates character- level embedding andpre-trained medical word embedding.Theproposed mode lachievesstate-of-the art result son three datasets. The three dataset sunder consideration are BioMedical Knowledge Comprehension Title(BMKC T), BioMedical Knowledge Comprehension Last Sentence(BMKC LS) andClicrdatasets. The model achieves84.7%accuracy on the BMKC T dataset whichout- performs the previousstate-of-the-art result which was78.6%.Moreover the model achieves77.4%accuracy on the BMKC LS dataset which is better than the previous state-of-the-art result which was71.4%. The model also achieves41.8%F1scoreonClicr dataset which surpasses the previous state-of-the-art F1score which was35.3%.