医学文献阅读增强深度学习方法

Deep learning method for reading enhancement of medical documents

针对专业医学术语的理解问题,提出一种新的方法。结合关键词层级划分树(HST)和基于注意力机制的卷积神经网络(lwCNN),分别用于抽取医学文献的关键内容和检索与医学文献相关的社区问答。首先利用关键词层级划分树对医学文献中的内容划分得到不同的片段,并对各个片段中的内容抽取代表性关键词,进而使用lwCNN对社区问答数据进行检索并做相关性排序,最后生成注解并链接到对应的文本片段中,藉此帮助读者理解医学文献。实验结果表明:本文方法相比经典的信息检索算法和优秀的深度学习算法,有更好的检索效果,它的P@5、MAP和MRR评价指标值均优于对比算法。

For the understanding of professional medical terminology,a new method combining a keyword hierarchical segmentation tree(HST)and an attentive convolutional neural network LWCNN is proposed.Firstly,HST is used to divide the content of medical document into different segments,and representative keywords are extracted from each segment.Then,LWCNN is used to retrieve and relevance rank the community Q&A data.Finally,the annotations are generated and linked to the corresponding segment,which can help readers to understand the medical document.The experimental results show that the method has better retrieval effect than classical information retrieval algorithms and excellent deep learning algorithms,and its P@5,MAP and MRR evaluation index values are better than the comparison algorithms.