摘要：在我国当前医疗服务网站快速发展的背景下，患者对医生的评价数据以及线上诊疗数据开始呈现体量大、增长快和多样化等大数据特征。一方面，这些信息为解决医患间信息不对称提供了帮助，另一方面，这些信息存在异质、缺失等不确定性，增加了患者选择医生的困难程度。如何选择合适的医生是患者在就诊过程中普遍面临的问题。因此，本文针对海量患者和医生，提出有效的医生推荐方法，从而为患者推荐合适医生。 论文的主要研究工作如下： (1)针对医疗服务信息中的文本信息，提出用图片犹豫模糊集表示的方法，对图片犹豫模糊集的概念、相关运算、比较规则、距离测度、相似度测度、熵测度以及集结算子等进行定义，并证明相关性质。 (2)针对现有医疗服务网站中存在的新患者(没有历史信息的新用户)没有历史评价信息的情况，考虑患者所患疾病，提出利用已有的文本评价信息，对相关医生进行排序，从而为新患者提供适当的医生推荐的方法，主要包括基于距离测度的图片犹豫TOPSIS医生推荐方法、基于距离测度的图片犹豫VIKOR医生推荐方法和基于距离测度的图片犹豫TODIM医生推荐方法。最后通过对实际数据的实例分析验证所提方法的合理性和有效性。 (3)针对现有医疗服务网站中给出过评价信息的老患者(拥有历史就诊记录且给出评价信息的历史用户)，根据其历史评价信息和其他患者对医生的评价信息，给出为老患者推荐老医生(获得评价数较多的医生)的方法，即基于图片犹豫模糊集和协同过滤的医生推荐方法。其中，通过考虑目标患者所患疾病为历史疾病和非历史疾病两种情况，分别提出基于患者的协同过滤医生推荐方法和基于医生的协同过滤医生推荐方法。最后，通过对真实抓取数据的实例分析验证所提方法在相应背景下的可行性和有效性。 (4)针对现有医疗服务网站中给出过评价信息的老患者，根据其历史评价信息和相关患者对医生的评价信息，给出为老患者推荐新医生(获得评价数较少或新注册医生)的方法，即基于图片犹豫模糊集和领域专家度的医生推荐方法。通过考虑新医生对患者提供线上咨询(医疗服务)及医生的个人学术成果(引文)两种信息来源，分别提出基于领域专家度和医疗服务信息的医生推荐方法和基于领域专家度和引文信息的医生推荐方法。最后，通过对真实抓取数据的实例分析验证所提方法在相应背景下的可行性和有效性。 (5)针对实际患者对医生评价数据的时间跨度较大，不同时段的评价信息对患者的影响程度不同的情况，提出考虑时间信息的医生推荐方法；考虑到实际医生推荐过程中往往需要短时间为新老患者同时提供包含新老医生在内的医生推荐列表的情况，提出基于图片犹豫模糊集的混合医生推荐方法。最后，通过对真实抓取数据的实例分析验证所提方法在相应背景下的可行性和有效性。 图51幅，表49个，参考文献112篇 关键词：图片犹豫模糊集；信息测度；集结算子；医生推荐方法；协同过滤；领域专家度 分类号：C934

 Abstract: With the rapid development of Chinese medical service websites, the data of diagnosis and doctor evaluation might show some features of big data that could be characterized by large quantity, high growth and diversity. On the one hand, the information on the Internet helps to solve the information asymmetry between doctors and patients. On the other hand, these data contain uncertain information, including heterogeneous information, missing information and so on, which make it a more difficult problem for patients to select appropriate doctors. The challenge of medical services is now to construct an effective doctor recommendation model with large numbers of patients and doctors. Therefore, in this paper, in order to recommend appropriate doctors to the target patient, some effective doctor recommendation methods have been proposed, considering numerous patients and doctors. Thesis work mainly includes the following aspects: (1)We propose the picture hesitant fuzzy set to deal with the textual information of medical service, then study on the definition, operation, distance measurements, similarity measurements, entropy measurements and aggregation operators for the picture hesitant fuzzy sets, with some proofs of corresponding properties. (2)We propose three doctor ranking approaches for the new patients (new users), namely, a distance-based picture hesitant TOPSIS doctor recommendation method, a distance-based picture hesitant VIKOR doctor recommendation method, and a distance-based picture hesitant TODIM doctor recommendation method. The new patients of the medical service websites are some users wanna to find doctors without personal historic information. Finally, some experiments are conducted to validate the reasonability and feasibility of proposed approaches. (3)We propose the doctor recommendation approaches based on picture hesitant fuzzy sets and the collaborative filtering, to recommend historic doctors for the historic patients. The historic doctors are doctors with enough evaluation information. The doctor recommendation approaches include patient-based collaborative filtering doctor recommendation approach and doctor-based collaborative filtering doctor recommendation approach, aiming at helping patients having historic disease and patients having new disease, respectively. Finally, some experiments are conducted to validate the reasonability and feasibility of proposed approaches. (4)We propose the doctor recommendation approaches based on picture hesitant fuzzy sets and domain expertise, to recommend new doctors for the historic patients. The new doctors are doctors with little online information. The doctor recommendation approaches include domain expertise-based doctor recommendation approach with medical service information and domain expertise-based doctor recommendation approach with paper information, aiming at helping patients having historic disease and patients having new disease, respectively. Finally, we conduct some experiments to validate the effectivity and feasibility of proposed approaches. (5)Considering the effect of time, we propose the doctor recommendation with time information. Because the realistic doctor recommendation requires to recommend historic and new doctors to historic and new patients simultaneously, we propose the hybrid doctor recommendation approach based on the picture hesitant fuzzy sets. Finally, we conduct some experiments to validate the effectivity and feasibility of proposed approaches. There are 51 figures, 49 tables and 112 references in this thesis. Key words: Picture hesitant fuzzy set; Information measures; Aggregation operators; Doctor recommendation method; Collaborative filtering ; Domain expertise Classification: C934