在线诊断：针对临床征兆的诊断搜索模型研究

With the development of online medical services, people prefer to search the symptoms online from the Internet, before they ask the help from the professionals, with the aim to understand the condition they may have. This procedure is defined as online medical diagnosis (OMD). Different than the traditional face-to-face diagnosis, OMD makes use of the search model to compensate the missing of doctors in the process. Different than the general commercial search engine, OMD focuses on the medical domain and outputs the diagnosis other than the general information. Different than the professional medical search, the target users of OMD are those with no or scarce medical knowledge, the data of OMD are mainly from the Internet and social media platforms, and OMD is required to translate the symptoms into knowledge and make diagnosis decision....In this proposal, we focus on the online diagnosis search with medical symptoms, especially on symptom query representation, “symptom-knowledge” association rules and diagnosis decisions. We first introduce a labeled Latent Dirichlet Allocation (L-LDA) model for symptom query representation, where the analysis of user behaviors are embedded into the LDA model. Second, we propose a Bayesian learning based mining model for “symptom-knowledge” associations, as one of the functions of “doctors” in traditional clinic diagnosis. After that, we present a concept based learning to rank algorithm to extract the diagnosis from the general information, as the other function of “doctors”.

全民关注健康的时代，在线医疗的发展促进了互联网医疗数据的迅速增长，人们往往选择在线搜索的方式对其出现的临床征兆进行诊断／预诊断，称为在线诊断。和传统医疗诊断不同，在线诊断缺乏“医生”角色的智能支持，因此通过搜索模型来模仿“医生”职能。和一般搜索引擎不同，在线诊断强调医疗领域搜索，输出为诊断结果，而非一般信息普及。和专业数据库搜索不同，在线诊断主要针对无医疗知识用户，数据来源多样化以及要进行知识转换和诊断决策。..本申请课题关注临床征兆的在线诊断搜索模型，侧重征兆查询语句重写、“征兆－术语”知识关联和诊断结果输出。首先，针对无背景用户的无序征兆查询，课题将用户行为分析和主题模型相结合，设计一种基于用户行为的半监督主题模型。其次，对应“医生”角色的知识转换职能，提出一种基于贝叶斯学习的“知识－术语”关系挖掘算法。第三，对应“医生”角色的诊断决策职能，提出一种基于诊断概念特征学习的分类算法。

全民关注健康的时代，在线医疗的发展促进了互联网医疗数据的迅速增长，人们往往选择在线搜索的方式对其出现的临床征兆进行诊断／预诊断，称为在线诊断。和传统医疗诊断不同，在线诊断缺乏“医生”角色的智能支持，因此通过搜索模型来模仿“医生”职能。和一般搜索引擎不同，在线诊断强调医疗领域搜索，输出为诊断结果，而非一般信息普及。和专业数据库搜索不同，在线诊断主要针对无医疗知识用户，数据来源多样化以及要进行知识转换和诊断决策。..本申请课题关注临床征兆的在线诊断搜索模型，侧重征兆查询语句重写、“征兆－术语”知识关联和诊断结果输出。首先，针对无背景用户的无序征兆查询，课题将用户行为.分析和主题模型相结合，设计一种基于用户行为的半监督主题模型。其次，对应“医生”角色的知识转换职能，提出一种基于贝叶斯学习的“知识－术语”关系挖掘算法。第三，对应“医生”角色的诊断决策职能，提出一种基于诊断概念特征学习的分类算法。