Semantic Model And Extraction Of Clinical Risk Statement In Medical Literature

With the rapid development of biomedicine,the medical literature shows an exponential and explosive growth trend.Medical literature is a huge treasure trove of medical knowledge.It is more and more important to explore the risk knowledge of medical literature,because risk expression is closely linked to the characteristics of evidence-based medicine.Currently,the research of semantic model of medical literature risk statement is not particularly much.At present,the extraction of medical literature risk knowledge mainly lies in two aspects.On the one hand,medical personnel analyze the risk knowledge in a large number of medical texts,then integrate risk knowledge and build a literature review.On the other,the researchers of the natural language processing deal with the abstracts of the medical literature,and extract the influencing elements and influenced elements about the percentages of the risk,but the accuracy and recall rate are relatively low,the accuracy rate is between 50 and 70%,the recall rate between 30 and 50%,accuracy rate and recall rate have room for improvement.The article focuses on the risk knowledge in the medical literature of breast cancer,and extracts the risk events from the medical literature.The risk events include the influencing elements,influenced elements,keyword description elements,source description elements,degree description elements,and time description elements.First of all,we construct the risk sentence patterns to extract the risk sentences from medical texts.Then,risk events are extracted by dictionary-hidden Markov model,dictionary-conditional random field,role labeling BIO-hidden Markov model and role labeling BIO-conditional random field,respectively.The results of dictionary-sequence labeling algorithm are used for baseline.By comparing the baseline,we can recognize that the method of role labeling BIO-conditional random field is the best,the accuracy of various events is higher than 70%,and the average F1 value reaches 70.6%.