基于互联网与本地数据融合的“互联网+医疗健康”信息检索研究

The performance of current general-purpose search engines for the healthcare information is not high. With the emergence of “intelligent healthcare” and “health China”, the healthcare information retrieval is facing new opportunities and challenges. The healthcare data on Internet (called online data) has all typical characteristics of big data, but its quality varies greatly with the different sources. The healthcare data in local hospitals (called offline data) has high quality, but it cannot be accessed by the public. Both online and offline data are fragmented over time, class imbalance, and extremely sparse. Therefore, in order to satisfy the healthcare information needs of different user groups, this project will systematically study the basic theories and key technologies of the "Internet + big data + healthcare" information retrieval system based on the fusion of both online and offline healthcare data, including: research on deep learning based data fusion and knowledge representation of multiple heterogeneous data sources from both online and offline; research on the disease progression modeling, individual differences analysis, temporal and spatial differences analysis, as well as complex internal and external relationship mining; research on a new search result diversification algorithm to satisfy information needs of multiple user groups by integrating above studies. The final goal of this project is that through the fusion of online and offline healthcare data by adopting strong points while overcoming weak points, to mine the evolution process and implicit knowledge for healthcare problems, to make the information representation to be more suitable for the understanding of different user groups, to improve the quality and efficiency of medical information retrieval, as well as to provide new technical supports for human health and disease prevention and treatment.

当前通用搜索引擎在医疗健康信息检索方面的效果有限，随着智慧医疗、健康中国的提出，医疗健康信息的检索面临新的机遇与挑战。互联网上(线上)的医疗数据数量巨大，但质量良莠不齐；医院本地(线下)的数据质量较高，公众却无法访问；两者都在时间上碎片化、类不平衡、非常稀疏。因此，本项目从满足不同用户群体的医疗健康信息需求的角度出发，拟研究基于线上、线下数据融合的“互联网+大数据+医疗健康”信息检索中的基础理论及关键技术，包括：研究基于深度学习的线上、线下多源异构数据的融合及知识表示；研究医疗健康问题的演化过程建模、个体与时空差异分析、复杂关联关系挖掘；最后研究融合以上成果实现新型的面向多目标的多样化信息检索算法。本项目的目标是通过线上、线下数据的融合，优势互补，挖掘问题的演化规律及隐含的知识，使信息的表示适合不同群体的理解能力，提高人们获取信息的质量和效率，为人类保健及疾病的预防、治疗提供新的技术支撑。

当前人民生活水平迅速提高，医疗健康成为当前人们关注的一个极其重要的话题。而随着海量医疗健康信息的数字化，以及大数据、人工智能技术的快速发展，医疗健康信息的处理面临新的机遇。本课题围绕“互联网+大数据+医疗健康”的信息处理技术展开研究，四年来的主要研究内容归纳为如下四个方面：(1)研究了医疗信息检索，提出了一种基于医学知识图谱和深度学习算法的医疗信息检索方法。(2)研究了医疗问题/答案的匹配与医生推荐，提出了一个融合医疗知识和深度学习的医疗问题/答案智能匹配与医生推荐模型。(3)研究了电子病历的用药情况，提出了一种结合强化学习和图卷积网络的药物组合预测推荐算法。(4)实现了相应的原型系统，设计开发了一个医疗信息检索及辅助诊断系统，并在医院相关科室进行试用。. 课题基本按原定计划进行，达到了预期目标，取得了预期成果。相关研究成果在国际国内期刊和会议发表论文22篇，其中CCF A类论文11篇，CCF B类论文8篇。授权或受理发明专利9项，取得软件著作权2项。培养了博士生4名，硕士生9名，其中已毕业博士1名，硕士4名。