【作者】 程坤；

【导师】 杨矫云； 束宇翔；

【作者基本信息】 合肥工业大学 ， 软件工程（专业学位）， 2021， 硕士

【摘要】 因果关系是一类具有明确导向的关系类型,尤其在健康领域具有更多的实际应用价值,比如分析疾病的致病因素等。目前,很多生物医学文献都包含因果关系知识,这部分因果关系值得被挖掘。为此,本文设计开发了一个面向健康领域的因果关系图构建系统,提取相关生物医学文献中的因果关系,并整合不同数据源,构建因果关系图,旨在帮助健康领域的从业人员决策判断,归因分析等。本文工作分为以下两部分:(1)提出了基于循环神经网络的因果关系提取方法,该方法分为四个步骤,分别是数据预处理、复杂句简化、因果实体确定以及因果关系提取。在数据预处理阶段,去除句子中注释,特殊符号等多余成分。在复杂句简化阶段,根据从句剥离以及并列句分解这两个方面进行简化。在因果实体确定阶段,确定简单句中的主语和宾语,对主宾进行语义扩充。在因果关系提取阶段,使用结合自注意力机制以及相似性特征的双向门控网络提取因果关系。实验结果表明,本文构建的因果关系提取模型提高了现有模型的提取精度,F值达到92.5%。(2)设计并实现了面向健康领域的因果关系图构建系统,该系统包含三个模块,分别是因果关系提取模块、因果关系图构建模块以及数据可视化模块。在因果关系提取模块中,自动化提取非结构化文本中包含的因果关系。在因果关系图构建模块中,手动或基于人工修正的自动化提取结果构建因果关系图。在数据可视化模块中,可视化展示已经构建完成的因果关系图。目前系统已经包含了护理,阿尔兹海默症,心脏病,心血管疾病,肝部疾病等健康领域的因果关系图数据。更多还原

【Abstract】 Causality is a type of relationship with a clear orientation,especially in the field of health,which has more practical application value,such as analyzing the pathogenic factors of diseases.At present,a lot of biomedical literature contains causal knowledge,this part of the causality is worth exploring.To this end,this dissertation designs and develops a causal graph construction system for the health field,which extracts the causality in relevant biomedical literature and integrates different data sources to construct a causal graph,aiming to help practitioners in the health field make decisions and analyze the reason,etc.The work of this dissertation is divided into the following two parts:(1)A recurrent-neural-network-based causality extraction method is proposed.The causality extraction method consists of four steps: data preprocessing,complex sentence simplification,causal entity determination,and causality extraction.In the data preprocessing phase,the extra components such as comments and special symbols in the sentence are removed.In the simplification phase of complex sentences,the simplification is carried out according to the stripping of clauses and the decomposition of compound sentences.In the causal entity determination phase,the subject and object in a simple sentence semantically are determined.In the causality extraction phase,the model of Bi-GRU combined with the self-attention mechanism and similarity features is used to extract the causality.The experimental results show that the proposed causality extraction model improves the extraction accuracy of the existing models,and the F value reaches 92.5%.(2)A causal graph construction system is designed and implemented for the health field.The system contains three modules,namely the causality extraction module,the causal graph construction module,and the data visualization module.In the causality extraction module,the causality contained in the unstructured text is automatically extracted.In the causal graph construction module,the causal graph is constructed manually or automatically based on the manual correction.In the data visualization module,the causal graph that has been constructed is displayed visually.At present,the system already contains causal graph data in the health fields such as nursing,Alzheimer’s disease,heart disease,cardiovascular disease,liver disease.更多还原