越览（255）——精读期刊论文的5.案例研究：医生排名

分享兴趣，传播快乐，

增长见闻，留下美好。

亲爱的您，这里是LearingYard学苑！

今天小编为您带来“越览（255）——精读期刊论文

《D-Multi-granular Unbalanced Hesitant Fuzzy Linguistic

Term Sets and Their Application to

Multiple Attribute Decision Making》的

5 案例研究：医生排名”。

欢迎您的访问！

Share interest, spread happiness,

increase knowledge, and leave beautiful.

Dear, this is the LearingYard Academy!

Today, the editor brings the

＂Yue Lan (255)：Intensive reading of the journal article

＇D-Multi-granular Unbalanced Hesitant

Fuzzy Linguistic Term Sets and Their Application

to Multiple Attribute Decision Making’

5 Case studies: Doctor ranking.

Welcome to visit!

一、内容摘要（Summary of Content）

本期推文将从思维导图、精读内容、知识补充三个方面介绍精读期刊论文《D-Multi-granular Unbalanced Hesitant Fuzzy Linguistic Term Sets and Their Application to Multiple Attribute Decision Making》的5 案例研究：医生排名。

This issue of tweets will introduce 5 Case studies: Doctor ranking of "D-Multi-granular Unbalanced Hesitant Fuzzy Linguistic Term Sets and Their Application to Multiple Attribute Decision Making" from three aspects: mind map, intensive reading content, and knowledge supplement.

二、思维导图（Mind map）

三、精读内容（Intensive reading content）

本节主要通过在线医疗平台中的医生选择案例，验证前文提出的基于D-MGUHFLTS的多属性决策（MADM）方法的实用性与有效性。随着“互联网+医疗”的快速发展，春雨医生、丁香园、39健康网、好大夫在线等在线医疗平台逐渐普及，为患者提供了预约挂号、在线问诊、健康教育和疾病风险评估等服务，使患者能够更加便捷地获取医疗资源。然而，随着平台上医生数量不断增加，患者面临着如何从众多医生中选择最合适医生的问题，因此医生选择逐渐成为一种典型的多属性决策问题。

This section primarily uses a doctor selection case on an online medical platform to verify the practicality and effectiveness of the proposed D-MGUHFLTS-based multi-attribute decision-making (MADM) method. With the rapid development of "Internet + Healthcare," online medical platforms such as Chunyu Doctor, DXY.cn, 39health.com, and Haodf.com have become increasingly popular, providing patients with services such as appointment booking, online consultations, health education, and disease risk assessments, enabling them to access medical resources more conveniently. However, as the number of doctors on these platforms continues to increase, patients face the challenge of choosing the most suitable doctor from among many. Therefore, doctor selection has gradually become a typical multi-attribute decision-making problem.

与此同时，患者在在线医疗平台上留下的评论通常以自然语言形式表达，具有模糊性、犹豫性以及评价粒度不统一等特点。例如，不同患者可能分别使用“好”“较好”“非常好”等不同粒度的语言评价医生，传统的数值型决策方法难以准确表达这类复杂的语言信息。D-MGUHFLTS能够较好地描述不同粒度、不平衡以及犹豫性的语言评价，因此适合用于处理在线医疗评论信息。基于此，本文将患者的在线评论转换为D-MGUHFLTS形式，并进一步利用第4节提出的MADM方法对候选医生进行排序，以此展示该方法从实际语言评价信息处理到最终决策排序的完整过程，并验证D-MGUHFLTS及所提出MADM方法在实际决策问题中的应用价值。

Meanwhile, patient reviews on online medical platforms are typically expressed in natural language, characterized by ambiguity, hesitation, and inconsistent granularity of evaluation. For example, different patients may use different levels of granularity in their evaluations of doctors, such as "good," "good," and "very good." Traditional numerical decision-making methods struggle to accurately represent this complex linguistic information. D-MGUHFLTS can effectively describe linguistic evaluations of varying granularity, imbalance, and hesitation, making it suitable for processing online medical review information. Based on this, this paper converts patient online reviews into D-MGUHFLTS format and further utilizes the MADM method proposed in Section 4 to rank candidate doctors. This demonstrates the complete process of this method from processing actual linguistic evaluation information to final decision ranking, and verifies the application value of D-MGUHFLTS and the proposed MADM method in practical decision-making problems.

（二）数据处理（Data processing）

本部分主要介绍了如何从好大夫在线的真实患者评论中提取医生评价信息，并将非结构化的在线评论逐步转化为D-MGUHFLTS形式，为后续的医生多属性决策排序提供数据基础。整体上，这是整个案例研究中“数据采集—文本处理—主题提取—属性确定—意见提取—语言变量转换”的过程。

This section primarily introduces how to extract doctor evaluation information from real patient reviews on Haodf.com, and how to gradually transform unstructured online reviews into D-MGUHFLTS format, providing a data foundation for subsequent multi-attribute doctor decision ranking. Overall, this represents the process of "data collection—text processing—topic extraction—attribute determination—opinion extraction—linguistic variable transformation" in the entire case study.

首先，研究使用Python从好大夫在线平台收集了10名医生的6201条高血压患者在线评论。这些评论主要包括患者的文字评价和平台提供的印象标签。由于患者评论属于非结构化文本，研究首先利用LAC进行中文分词，并进一步删除标点符号和停用词，从评论中保留具有实际分析意义的词语。在完成文本预处理后，研究采用LDA主题模型对患者评论进行主题分析，并根据不同主题数量下的困惑度确定最终的主题数量为4个，从而识别患者评论中主要关注的内容。

First, the study used Python to collect 6201 online comments from 10 doctors on the Haodf.com platform, collected from patients with hypertension. These comments mainly included patients' written evaluations and impression tags provided by the platform. Since the patient comments were unstructured text, the study first used LAC (Laser-Based Character Analysis) for Chinese word segmentation, and then further removed punctuation and stop words, retaining words with practical analytical significance. After text preprocessing, the study employed LDA (Laser-Based Topic Analysis) to perform topic analysis on the patient comments, and determined the final number of topics to be four based on the perplexity under different topic counts, thereby identifying the main concerns in the patient comments.

随后，研究对LDA得到的主题词进行分析，并进一步确定医生评价所涉及的属性。例如，“建议、回复、咨询、诊断、指导”等词语主要反映医生与患者之间的沟通能力；“耐心、细致、认真”“热情、服务态度”“态度、和善”等词语主要反映医生的服务态度；而“专业、经验、医术、技能”等词语则体现医生的医疗技能。在结合平台本身提供的“态度满意度”“疗效满意度”等评价指标后，研究最终形成医生评价属性体系，为后续的多属性决策提供评价维度。

Subsequently, the study analyzed the keywords obtained from LDA and further identified the attributes involved in physician evaluations. For example, terms such as "suggestion, response, consultation, diagnosis, guidance" primarily reflect the communication skills between doctors and patients; terms such as "patience, meticulousness, conscientiousness," "enthusiasm, service attitude," and "attitude, kindness" primarily reflect the doctor's service attitude; while terms such as "professionalism, experience, medical skills, expertise" reflect the doctor's medical skills. By combining these with evaluation indicators such as "attitude satisfaction" and "treatment satisfaction" provided by the platform itself, the study ultimately formed a physician evaluation attribute system, providing evaluation dimensions for subsequent multi-attribute decision-making.

在确定评价属性之后，研究进一步利用百度AI平台的细粒度意见抽取功能，从患者评论中提取具体的评价关键词和评价值，并将这些信息映射到相应的医生评价属性上。例如，在“陈主任医术精湛，下次还会选择他”这一评论中，可以提取出“医疗技能”这一关键词以及“优秀”这一评价值。之后，通过百度AI平台中的语义相似度计算模型，将评论中的关键词与医生评价属性进行匹配，同时将患者使用的“优秀”等评价词转换为相应的语言术语，从而获得每名医生在不同属性下的语言评价值。

After determining the evaluation attributes, the study further utilized the fine-grained opinion extraction function of the Baidu AI platform to extract specific evaluation keywords and values from patient reviews and map this information to the corresponding doctor evaluation attributes. For example, in the review "Dr. Chen's medical skills are superb; I will choose him again next time," the keyword "medical skills" and the evaluation value "excellent" can be extracted. Then, using the semantic similarity calculation model in the Baidu AI platform, the keywords in the reviews are matched with the doctor evaluation attributes, and the evaluation words used by patients, such as "excellent," are converted into corresponding linguistic terms, thereby obtaining the linguistic evaluation values for each doctor under different attributes.

最后，为了进一步处理不同属性中存在的语言粒度不一致和语言术语数量不同的问题，研究分别为三个属性建立了不同的语言术语集。其中，第一个属性使用包含5个语言术语的语言集合，第二个属性使用包含6个语言术语的语言集合，第三个属性使用包含4个语言术语的语言集合。三个语言术语集的表达范围和粒度并不完全相同，例如有的评价尺度包含“差、一般、好、很好、完美”，有的则进一步加入“较好”等中间评价。这种不同属性采用不同粒度、不同结构的语言评价体系，正是D-MGUHFLTS所要处理的典型问题。研究最后采用已有研究中的语言术语隶属函数，并根据本案例进行调整，从而建立三个语言术语集的语义表示，为后续将医生评价构造为D-MGUHFLTS以及开展MADM排序奠定基础。

Finally, to further address the issues of inconsistent linguistic granularity and varying numbers of linguistic terms across different attributes, the study established different linguistic term sets for each of the three attributes. The first attribute used a set containing 5 terms, the second 6 terms, and the third 4 terms. The scope and granularity of these three term sets are not entirely identical; for example, some evaluation scales include "poor," "average," "good," "very good," and "perfect," while others further incorporate intermediate ratings such as "relatively good." This use of different granularities and structures for different attributes is a typical problem that D-MGUHFLTS aims to address. The study ultimately adopted linguistic term membership functions from existing research and adjusted them for this case study to establish semantic representations for the three term sets, laying the foundation for constructing D-MGUHFLTS for physician evaluations and conducting MADM ranking.

（三）信息转换（Information conversion）

本部分主要是在解决一个关键问题：患者的每一条在线评论对医生评价的重要程度并不相同，因此不能简单地把所有评论等权处理。 为了更加真实地反映不同评论对最终医生决策的影响，研究进一步引入“评论置信度（confidence level）”来衡量每条评论的重要程度，并最终将患者评论转化为带有置信度的D-MGUHFLTE，为后续构建混合决策矩阵做准备。

This section primarily addresses a key issue: the importance of each patient's online review to the physician's evaluation varies, therefore, all reviews cannot be simply treated with equal weight. To more accurately reflect the impact of different reviews on the final physician's decision, the study further introduces "review confidence level" to measure the importance of each review, and ultimately transforms patient reviews into D-MGUHFLTE with confidence levels, preparing for the subsequent construction of a hybrid decision matrix.

具体来说，研究认为，一条评论的时效性、信息丰富程度以及涉及的评价属性数量都会影响其可信程度。首先，评论发布得越晚，越能够反映医生当前的服务状态，因此评论时间越接近当前决策时间，其置信度越高；其次，一条评论包含的关键词越多，说明患者从更多方面对医生进行了评价，信息更加丰富，因此置信度越高；再次，一条评论涉及的医生评价属性越多，说明评论覆盖的信息更加全面，因此也具有更高的参考价值。此外，研究还引入了参考值（reference value）作为特征指标。因此，最终选择了评论时间、关键词数量、属性数量和参考值四个特征指标来衡量每条评论的重要程度。

Specifically, the study suggests that the timeliness, information richness, and number of evaluative attributes involved in a comment all affect its credibility. First, the later a comment is published, the more it reflects the doctor's current service status; therefore, the closer the comment's time is to the current decision-making time, the higher its confidence level. Second, the more keywords a comment contains, the more comprehensive the patient's evaluation of the doctor, resulting in richer information and thus higher confidence. Third, the more doctor evaluation attributes a comment involves, the more comprehensive the information covered, and therefore, the higher its reference value. Furthermore, the study introduced a reference value as a feature indicator. Therefore, the study ultimately selected four feature indicators—comment time, number of keywords, number of attributes, and reference value—to measure the importance of each comment.

在获得这四个指标的数据后，研究首先利用Max–Min归一化方法对不同量纲的数据进行标准化处理，使不同指标能够进行统一比较。随后采用熵权法确定四个特征指标的客观权重，得到权重向量为(0.1,0.4,0.2,0.3)。其中，关键词数量的权重最高，为0.4，说明在该案例中评论所包含的信息丰富程度对评论置信度的影响最大；参考值的权重为0.3，属性数量为0.2，评论时间为0.1。最后，将每条评论在四个指标上的标准化结果与对应权重进行加权求和，得到每条患者评论的置信度 vi。

After obtaining the data for these four indicators, the study first standardized the data of different dimensions using the Max-Min normalization method, enabling unified comparison of different indicators. Then, the entropy weight method was used to determine the objective weights of the four feature indicators, resulting in a weight vector of (0.1, 0.4, 0.2, 0.3). The number of keywords had the highest weight of 0.4, indicating that the richness of information contained in the comment had the greatest impact on the comment confidence in this case. The reference value had a weight of 0.3, the number of attributes had a weight of 0.2, and the comment time had a weight of 0.1. Finally, the standardized results of each comment on the four indicators were weighted and summed with their corresponding weights to obtain the confidence level vi for each patient's comment.

在此基础上，研究进一步将前面通过百度AI意见抽取得到的评论评价值转换为语言术语值，并将每条评论的置信度融入语言评价中，从而将普通的语言评价转化为D-MGUHFLTE。例如，某医生在“医疗技能”这一属性下获得了一条“好”的患者评价，并且该评论的置信度为0.2，那么这条评论就可以表示为(s2,0.2)。其中，s2表示“好”这一语言术语，0.2则表示该评论在决策中的置信度。对所有患者评论进行类似处理之后，就可以得到包含不同医生、不同属性以及不同语言评价和置信度信息的混合决策矩阵，即表8，为后续利用D-MGUHFLTS进行医生排序奠定基础。

Building upon this foundation, the study further transforms the comment evaluation values obtained through Baidu AI opinion extraction into linguistic term values, and incorporates the confidence level of each comment into the linguistic evaluation, thus converting ordinary linguistic evaluation into D-MGUHFLTS. For example, if a doctor receives a "good" patient evaluation under the attribute "medical skills" with a confidence level of 0.2, this evaluation can be represented as (s2, 0.2). Here, s2 represents the linguistic term "good," and 0.2 represents the confidence level of the evaluation in the decision-making process. After similar processing of all patient comments, a hybrid decision matrix containing information on different doctors, attributes, linguistic evaluations, and confidence levels is obtained, as shown in Table 8. This lays the foundation for subsequent doctor ranking using D-MGUHFLTS.

（四）医生程序排名（Doctor ranking program）

本段主要介绍了利用前面获得的D-MGUHFLTS评价信息和决策矩阵，对10名医生进行综合评价与排序的过程。首先将评价信息转换为比例2元组集合，并根据决策者对属性重要性的判断，利用AHP确定三个属性的权重为(0.26,0.63,0.11)。随后考虑属性之间的关联关系，采用D-MGUHFWABM算子对各医生不同属性下的评价信息进行聚合，得到每名医生的综合评价值。最后通过得分函数计算各医生的综合得分，并按照得分进行排序，最终得到D1 > D8 > D3 > D5 > D7 > D6 > D4 > D10 > D2 > D9，因此D1为最适合患者选择的医生。

This section mainly introduces the process of comprehensively evaluating and ranking 10 doctors using the previously obtained D-MGUHFLTS evaluation information and decision matrix. First, the evaluation information is converted into a set of proportional 2-tuples. Based on the decision-maker's judgment of attribute importance, the weights of the three attributes are determined using AHP (Adaptive Hierarchy Process) as (0.26, 0.63, 0.11). Then, considering the correlation between attributes, the D-MGUHFLTS WABM operator is used to aggregate the evaluation information of each doctor under different attributes, obtaining a comprehensive evaluation value for each doctor. Finally, the comprehensive score of each doctor is calculated using a scoring function, and the doctors are ranked according to their scores, resulting in D1 > D8 > D3 > D5 > D7 > D6 > D4 > D10 > D2 > D9. Therefore, D1 is the doctor most suitable for the patient's selection.

四、知识补充（Knowledge supplement）

在实际的网络评价决策中，如何判断一条评论是否具有较高的参考价值是一个重要问题。传统方法往往默认所有评论具有相同的重要性，但实际上不同评论的信息质量存在明显差异。例如，评论发布时间越接近决策时点，越能够反映对象当前的真实状态；评论中包含的关键词越多，通常意味着提供的信息更加丰富；同时，一条评论涉及的评价属性越多，也说明其能够从更多维度反映被评价对象。因此，可以通过构建多个评论特征指标，对评论的重要程度进行量化。

In practical online evaluation decisions, determining the reference value of a comment is crucial. Traditional methods often assume all comments have equal importance, but in reality, the information quality of different comments varies significantly. For example, the closer a comment's posting time is to the decision-making point, the more accurately it reflects the current state of the object; the more keywords a comment contains, the richer the information it provides; and the more evaluation attributes a comment involves, the more dimensions it reflects the evaluated object. Therefore, the importance of comments can be quantified by constructing multiple comment feature indicators.

本案例采用评论时间、关键词数量、属性数量和参考值四个指标衡量评论的有效性，并通过Max–Min方法消除不同指标之间的量纲差异，再利用熵权法确定各指标的客观权重。熵权法的基本思想是根据指标数据的离散程度确定权重：指标差异越明显，能够提供的有效信息越多，其权重通常越高。经过加权计算后，可以得到每条评论对应的置信度，从而避免简单地将所有患者评论视为同等重要。

This case study uses four indicators to measure the effectiveness of reviews: review time, number of keywords, number of attributes, and reference values. The Max-Min method is used to eliminate dimensional differences between the indicators, and then the entropy weight method is employed to determine the objective weight of each indicator. The basic idea of the entropy weight method is to determine the weight based on the dispersion of the indicator data: the more significant the difference between indicators, the more effective information they provide, and their weight is usually higher. After weighted calculation, the confidence level corresponding to each review can be obtained, thus avoiding simply treating all patient reviews as equally important.

进一步地，评论置信度可以与语言评价信息结合，形成“语言术语+置信度”的评价形式。例如，患者认为某医生“医疗技能很好”，同时该评论具有较高的可信程度，那么最终的评价不仅体现“很好”这一语言判断，也能够反映该评论本身的重要性。这种方法能够更充分地保留网络评论中的模糊性、不确定性和信息质量差异，因此特别适用于在线医疗评价、商品评论、服务评价等非结构化信息较多的决策场景。

Furthermore, review confidence can be combined with linguistic evaluation information to form an evaluation format of "linguistic terminology + confidence level." For example, if a patient believes a doctor has "excellent medical skills," and the review has a high degree of credibility, then the final evaluation not only reflects the linguistic judgment of "excellent" but also the importance of the review itself. This method can more fully preserve the ambiguity, uncertainty, and information quality differences in online reviews, and is therefore particularly suitable for decision-making scenarios with a large amount of unstructured information, such as online medical reviews, product reviews, and service reviews.

从更广泛的角度来看，评论置信度实际上连接了文本信息处理与多属性决策两个环节。前一阶段通过文本挖掘、主题模型和意见抽取从大量评论中识别“评价了什么”，后一阶段则通过特征指标和权重计算进一步判断“这条评价有多大参考价值”，最终再将二者融合到D-MGUHFLTS决策框架中。这样既避免了直接丢失原始语言信息，也提高了网络评价数据在后续决策分析中的有效性。

From a broader perspective, comment confidence actually connects two stages: text information processing and multi-attribute decision-making. The first stage identifies "what is being evaluated" from a large number of comments through text mining, topic modeling, and opinion extraction. The second stage further determines "how valuable this evaluation is" through feature indicators and weight calculations. Finally, the two are integrated into the D-MGUHFLTS decision framework. This avoids directly losing original linguistic information and improves the effectiveness of online evaluation data in subsequent decision analysis.

今天的分享就到这里了。

如果您对文章有独特的想法，

欢迎给我们留言，让我们相约明天。

祝您今天过得开心快乐！

That's all for today's sharing.

If you have a unique idea about the article,

please leave us a message,

and let us meet tomorrow.

I wish you a nice day!

文案|yyz

排版|yyz

审核|hzy

翻译：谷歌翻译

参考资料：百度百科、Chat GPT

参考文献：Yongzhu Lu, Xihua Li. D-Multi-granular Unbalanced Hesitant Fuzzy Linguistic Term Sets and Their Application to Multiple Attribute Decision Making [J]. International Journal of Fuzzy Systems, 2025, 27(5): 1357-1372.

本文由LearningYard学苑整理发出，如有侵权请在后台留言！