考虑深度学习模型不确定性的在线医生推荐方法

doi:10.16381/j.cnki.issn1003-207x.2023.1494

摘要/Abstract

摘要：

基于深度学习模型的在线医生推荐已经成为提升互联网医疗平台患者服务水平的重要途径。由于关乎生命健康，与普通商品和服务推荐相比，医生推荐对结果的可靠性有着更高的要求。为此，本文考虑深度学习模型固有的不确定性，基于贝叶斯学习理论，提出了一种多模态贝叶斯神经协同过滤模型（multi-modal bayesian neural collaborative filtering， MM-BNCF）。首先，基于蒙特卡洛Dropout构建贝叶斯深度学习模型，将结构化患者反馈与非结构化医患文本数据作为输入，利用贝叶斯深度模型分别得到医生与患者表征；其次，基于贝叶斯深度模型分析医患双方匹配度，并评估医患匹配的不确定性；然后，基于匹配度均值生成医生推荐结果；最后，基于好大夫在线平台真实数据的实验结果验证了本文所提推荐方法能够更有效地适用于在线医生推荐任务。

Abstract:

The rapid development of Internet medical treatment has greatly improved the convenience of seeking medical advice， and at the same time， it has also caused patients to face the difficulty of finding a suitable doctor from the massive doctor database. Effective doctor recommendation has become an important way to improve the experience of patients on the Internet medical platform. As it is related to life and health， compared with ordinary product and service recommendation， doctor recommendation has higher requirements for the reliability of results. Deep learning models have been widely used in personalized recommendations in recent years. However， the inherent uncertainty of the deep learning models can easily lead to uncertain recommendation results， thereby affecting the reliability of the doctor recommendation. To address this， the inherent uncertainty of deep learning models is considered and a Multi-Modal Bayesian Neural Collaborative Filtering model （MM-BNCF） is proposed based on Bayesian learning theory. First， a Bayesian deep learning model is constructed based on Monte Carlo Dropout， with structured patient feedback and unstructured doctor and patient text data as input， to obtain representations of doctors and patients separately. Second， the Bayesian deep model is used to analyze the compatibility between doctors and patients， and to assess the uncertainty of doctor-patient matching. Then， doctor recommendation results are generated based on compatibility. Final， experimental results based on real data from the haodf.com online platform validate that the proposed recommendation method in this paper can be more effectively applied to online doctor recommendation tasks. There is theoretical and practical significance for improving the performance and patient satisfaction of Internet medical websites in this study.

Key words: online doctor recommendation, deep learning uncertainty, bayesian deep learning, multi-model data, internet medical care

中图分类号:

TP183

崔福来,柴一栋,姜元春, 等. 考虑深度学习模型不确定性的在线医生推荐方法[J]. 中国管理科学, 2025, 33(11): 151-161.

Fulai Cui,Yidong Chai,Yuanchun Jiang, et al. Online Doctor Recommendation Considering the Uncertainty of Deep Learning Models[J]. Chinese Journal of Management Science, 2025, 33(11): 151-161.

图/表 10

图1

图2

表1

表2

表3

表4

表5

表6

图3

图4

参考文献 44

[1]	杨善林，范先群，丁帅，等. 医联网与智慧医疗健康管理［J］. 管理科学， 2021， 34（6）： 71-75.
	Yang S L， Fan X Q， Ding S， et al. Internet of healthcare systems and smart medical health management［J］. Journal of Management Science， 2021， 34（6）： 71-75.
[2]	中国互联网络信息中心. 第50次《中国互联网络发展状况统计报告》［R］. 研究报告，中国互联网络信息中心， 2022.
	China Internet Network Information Center. The 50th statistical report on China’s internet development［R］. Discussion Paper， China internet network information center， 2022.
[3]	杨善林，丁帅，顾东晓，等. 医疗健康大数据驱动的知识发现与知识服务方法［J］. 管理世界， 2022， 38（1）： 219-229.
	Yang S L， Ding S， Gu D X， et al. Healthcare big data driven knowledge discovery and knowledge service approach［J］. Journal of Management World， 2022， 38（1）： 219-229.
[4]	Chai Y， Bian Y， Liu H， et al. Glaucoma diagnosis in the Chinese context： An uncertainty information-centric Bayesian deep learning model［J］. Information Processing & Management， 2021， 58（2）： 102454.
[5]	Fan W， Zhou Q， Qiu L， et al. Should doctors open online consultation services？ An empirical investigation of their impact on offline appointments［J］. Information Systems Research， 2022， 34（2）： 629-651.
[6]	路薇，高盼，翟运开. 带有反馈调节的远程医疗专家自适化推荐［J］. 系统管理学报， 2023， 32（5）： 960-975.
	Lu W， Gao P， Zhai Y K. An adaptive recommendation method for telemedicine specialists with feedback adjustment［J］. Journal of Systems & Management， 2023， 32（5）： 960-975.
[7]	周鑫，熊回香，肖兵. 一种融合标签和患者咨询文本的医生推荐算法［J］. 情报科学， 2023， 41（3）： 145-154.
	Zhou X， Xiong H X， Xiao B. A physician recommendation algorithm based on the fusion of label and patient consultation text［J］. Information Science， 2023， 41（3）： 145-154.
[8]	Fang H， Zhang D， Shu Y， et al. Deep learning for sequential recommendation［J］. ACM Transactions on Information Systems， 2021， 39（1）： 1-42.
[9]	Lu J， Wu D， Mao M， et al. Recommender system application developments： A survey［J］. Decision Support Systems， 2015， 74： 12-32.
[10]	Yan Y， Yu G， Yan X. Online doctor recommendation with convolutional neural network and sparse inputs［J］. Computational Intelligence and Neuroscience， 2020， 2020（1）： 8826557.
[11]	Chiang J H， Ma C Y， Wang C S， et al. An adaptive， context-aware， and stacked attention network-based recommendation system to capture users’ temporal preference［J］. IEEE Transactions on Knowledge and Data Engineering， 2022， 35（4）： 3404-3418.
[12]	胡春华，邓奥，童小芹，等. 社交电商中融合信任和声誉的图神经网络推荐研究［J］. 中国管理科学， 2021， 29（10）： 202-212.
	Hu C H， Deng A， Tong X Q， et al. A graph neural network recommendation study combining trust and reputation in social E-commerce［J］. Chinese Journal of Management Science， 2021， 29（10）： 202-212.
[13]	刘世峰，康来松，宫大庆.面向群组的事件兴趣点推荐算法研究［J］.中国管理科学，2023，31（12）： 301-310.
	Liu S F， Kang L S， Gong D Q. An event POI recommendation system for groups in EBSN［J］. Chinese Journal of Management Science，2023，31（12）： 301-310.
[14]	Brownlee J. Probability for machine learning： Discover how to harness uncertainty with Python［M］. San Francisco：Machine Learning Mastery， 2019.
[15]	Kendall A， Gal Y. What uncertainties do we need in Bayesian deep learning for computer vision？［C］//Proceedings of the 31st International Conference on Neural Information Processing Systems， Long Beach， USA， December 4-9 ， Curran Associates Inc.， 2017： 5580-5590.
[16]	王若佳，王继民. 用户认知视角下在线问诊平台医生推荐研究［J］.图书情报工作，2023，67（10）： 128-138.
	Wang R J， Wang J M. Research on doctor recommendation of online “ask the doctor” platforms based on the perspective of users recognition［J］. Library and Information Service， 2023， 67（10）： 128-138.
[17]	Huang Y F， Liu P， Pan Q， et al. A doctor recommendation algorithm based on doctor performances and patient preferences［C］//Proceedings of the 2012 International Conference on Wavelet Active Media Technology and Information Processing （ICWAMTIP）， Chengdu December 17-19， IEEE， 2012： 92-95.
[18]	熊回香，李晓敏，李建玲. 基于医患交互数据的在线医生推荐研究［J］. 情报理论与实践， 2020， 43（8）： 159-166.
	Xiong H X， Li X M， Li J L. Research on online doctor recommendation based on doctor-patient interaction data［J］. Information Studies （Theory & Application）， 2020， 43（8）： 159-166.
[19]	Chen X， Liang Q， Chen Y， et al. Cognitive-based knowledge learning framework for recommendation［J］. Knowledge-Based Systems， 2024， 287： 111446.
[20]	Iftikhar H， Anwar S M， Majid M. A doctor recommendation system using patient’s satisfaction analysis［C］// Proceedings of Intelligent Technologies and Applications，Bahawalpur， Pakistan， November 6-8 ， Springer Singapore， 2020： 201-209.
[21]	Xue F， He X， Wang X， et al. Deep item-based collaborative filtering for top-N recommendation［J］. ACM Transactions on Information Systems， 2019， 37（3）： 1-25.
[22]	Liu K， Xue F， Guo D， et al. MEGCF： Multimodal entity graph collaborative filtering for personalized recommendation［J］. ACM Transactions on Information Systems， 2023， 41（2）： 1-27.
[23]	杨善林，李霄剑，张强，等. 人工智能与管理变革［J］. 中国管理科学， 2023， 31（6）： 1-11.
	Yang S L， Li X J， Zhang Q， et al. Artificial intelligence and management transformation［J］. Chinese Journal of Management Science， 2023， 31（6）： 1-11.
[24]	Reyzin L. Unprovability comes to machine learning［J］. Nature， 2019， 565（7738）： 166-167.
[25]	姚加权，冯绪，王赞钧，等. 语调、情绪及市场影响：基于金融情绪词典［J］. 管理科学学报， 2021， 24（5）： 26-46.
	Yao J Q， Feng X， Wang Z J， et al. Tone， sentiment and market impacts： The construction of Chinese sentiment dictionary in finance［J］. Journal of Management Sciences in China， 2021， 24（5）： 26-46.
[26]	余艳，张文，熊飞宇，等. 融合知识图谱与神经网络赋能数智化管理决策［J］. 管理科学学报， 2023， 26（5）： 231-247.
	Yu Y， Zhang W， Xiong F Y， et al. Fusion of knowledge graph and neural network to empower data-intelligence for management decisions［J］.Journal of Management Sciences in China，2023，26（5）： 231-247.
[27]	吴俊杰，刘冠男，王静远，等. 数据智能：趋势与挑战［J］.系统工程理论与实践，2020， 40（8）： 2116-2149.
	Wu J J， Liu G N， Wang J Y， et al. Data intelligence： Trends and challenges［J］. Systems Engineering-Theory & Practice， 2020， 40（8）： 2116-2149.
[28]	He X， Zemel R S， Carreira-Perpinan M A. Multiscale conditional random fields for image labeling［C］//Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition， 2004， Washington，DC，USA，June 27-July 2， IEEE， 2004： 1315232.
[29]	Gal Y， Ghahramani Z. Dropout as a Bayesian approximation： Representing model uncertainty in deep learning［C］//Proceedings of the 33rd International Conference on International Conference on Machine Learning， NY， USA， June 19-24， JMLR.org， 2016： 1050-1059.
[30]	Gal Y， Ghahramani Z. A theoretically grounded application of dropout in recurrent neural networks［C］//Proceedings of the 30th International Conference on Neural Information Processing Systems， Barcelona， Spain， December 5-10 ，Curran Associates Inc.，2016： 1027-1035.
[31]	Zhang X， Mahadevan S. Bayesian neural networks for flight trajectory prediction and safety assessment［J］. Decision Support Systems， 2020， 131： 113246.
[32]	He X， Liao L， Zhang H， et al. Neural collaborative filtering［C］//Proceedings of the 26th International Conference on World Wide Web， Perth， Australia， April 3-7， IW3C2， 2017： 173-182.
[33]	国家医疗保障局. 国家医疗保障局关于积极推进“互联网+”医疗服务医保支付工作的指导意见［EB/OL］.（2020-11-02）［2024-06-19］..
	National Healthcare Security Administration. Guiding opinions of the national healthcare security administration on actively promoting the medical insurance payment for “Internet+” healthcare services［EB/OL］. （2020-11-02）［2024-06-19］..
[34]	Guo J， Tang Y， Han K， et al. Hire-MLP： Vision MLP via hierarchical rearrangement［C］//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）， New Orleans， LA， USA June 18-24， IEEE， 2022： 816-826.
[35]	Long C， Yuan H， Fang J， et al. Learning global and multi-granularity local representation with MLP for sequential recommendation［J］. ACM Transactions on Knowledge Discovery from Data， 2024， 18（4）： 1-15.
[36]	Hochreiter S， Schmidhuber J. Long short-term memory［J］. Neural Computation， 1997， 9（8）： 1735-1780.
[37]	Song Y， Shi S， Li J， et al. Directional skip-gram： Explicitly distinguishing left and Right Context for word embeddings［C］//Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics： Human Language Technologies，New Orleans，Louisiana， June 1-6 ，ACL，2018：175-180.
[38]	Gatzioura A， Vinagre J， Jorge A M， et al. A hybrid recommender system for improving automatic playlist continuation［J］. IEEE Transactions on Knowledge and Data Engineering， 2021， 33（5）： 1819-1830.
[39]	Deshpande M， Karypis G. Item-based top-Nrecommendation algorithms［J］. ACM Transactions on Information Systems， 2004， 22（1）： 143-177.
[40]	Xue H J， Dai X Y， Zhang J B， et al. Deep matrix factorization models for recommender systems［C］//Proceedings of the 26th International Joint Conference on Artificial Intelligence， Melbourne， Australia， August， 19-25， AAAIPress， 2017： 3203-3209.
[41]	Robertson S E， Walker S. Some simple effective approximations to the 2-Poisson model for probabilistic weighted retrieval［C］//Proceedings of the 17th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval， Dublin Ireland，July 3-6，Springer，1994：232-141.
[42]	Huang P S， He X， Gao J， et al. Learning deep structured semantic models for web search using clickthrough data［C］//Proceedings of the 22nd ACM International Conference on Information & Knowledge Management， San Francisco California USA， October 27- November 1， ACM， 2013： 2333-2338.
[43]	He X， Chen T， Kan M Y， et al. TriRank： Review-aware explainable recommendation by modeling aspects［C］//Proceedings of the 24th ACM International on Conference on Information and Knowledge Management， Melbourne Australia， October 18-23 ACM， 2015： 1661-1670.
[44]	Liu L， Özsu M T. Encyclopedia of Database Systems［M］. New York： Springer New York， 2009.

模型	参数	NDCG@5	NDCG@10	NDCG@15	MAP
ItemKNN	k=100	0.2640(36.55%)^***	0.2640(57.54%)^***	0.2643(76.20%)^***	0.3530(31.84%)^***
	k=50	0.2542(41.82%)^***	0.2546(63.35%)^***	0.2546(82.91%)^***	0.3478(33.81%)^***
	k=25	0.2466(46.19%)^***	0.2466(68.65%)^***	0.2466(88.85%)^***	0.3431(35.65%)^***
DMF	l=4	0.3206(12.45%)^***	0.3450(20.55%)^***	0.3879(20.06%)^***	0.4380(6.26%)^***
	l=3	0.3158(14.15%)^***	0.3376(23.19%)^***	0.3689(26.24%)^***	0.4311(7.96%)^***
	l=2	0.3052(18.12%)^***	0.3236(28.52%)^***	0.3743(24.42%)^***	0.4309(8.01%)^***
NeuMF	l=4	0.3525(2.27%)^**	0.4078(1.99%)^**	0.4527(2.87%)^***	0.4583(1.55%)^**
	l=3	0.3541(1.81%)^*	0.4059(2.46%)^**	0.4546(2.44%)^**	0.4576(1.70%)^**
	l=2	0.3421(5.38%)^***	0.3919(6.12%)^***	0.4386(6.18%)^***	0.4444(4.73%)^***
BM25	—	0.1248(188.86%)^***	0.1248(233.25%)^***	0.1248(273.16%)^***	0.2059(126.03%)^***
DSSM	l=4	0.3207(12.41%)^***	0.3993(4.16%)^**	0.4403(5.77%)^***	0.4029(15.51%)^***
	l=3	0.3434(4.98%)^**	0.4056(2.54%)^**	0.4538(2.62%)^**	0.4240(9.76%)^***
	l=2	0.3348(7.68%)^**	0.4088(1.74%)^**	0.4514(3.17%)^**	0.4216(10.39%)^***
MM-BNCF		0.3605	0.4159	0.4657	0.4654

dropout率	NDCG@5	NDCG@10	NDCG@15	MAP
0.9	0.1398	0.2054	0.2570	0.2110
0.8	0.1830	0.2529	0.3013	0.2600
0.7	0.2361	0.2949	0.3379	0.3111
0.6	0.2665	0.3114	0.3484	0.3415
0.5	0.3102	0.3414	0.3764	0.3939
0.4	0.3325	0.3645	0.4049	0.4281
0.3	0.3506	0.3938	0.4350	0.4489
0.2	0.3566	0.4055	0.4513	0.4582
0.1	0.3605	0.4159	0.4657	0.4654
0.0	0.3378	0.4043	0.4626	0.4564

感知器层数	NDCG@5	NDCG@10	NDCG@15	MAP
2	0.3549	0.4135	0.4649	0.4618
3	0.3554	0.4116	0.4607	0.4623
4	0.3605	0.4159	0.4657	0.4654
5	0.3586	0.4185	0.4677	0.4651
6	0.3581	0.4162	0.4644	0.4641

回看就诊数	NDCG@5	NDCG@10	NDCG@15	MAP
1	0.3553	0.4126	0.4604	0.4616
2	0.3588	0.4142	0.4643	0.4621
3	0.3533	0.4113	0.4615	0.4586
4	0.3605	0.4159	0.4657	0.4654
5	0.3585	0.4166	0.4670	0.4652
6	0.3524	0.4085	0.4607	0.4591
7	0.3532	0.4123	0.4631	0.4598
8	0.3588	0.4177	0.4684	0.4651
9	0.3602	0.4221	0.4728	0.4645

模型	NDCG@5	NDCG@10	NDCG@15	MAP
TextMatching	0.2982(20.89%)^***	0.3735(11.35%)^***	0.4167(11.76%)^***	0.3710(25.44%)^***
BNCF	0.3524(2.30%)^*	0.4085(1.81%)^*	0.4576(1.77%)^*	0.4590(1.39%)^*
MM-NCF	0.3378(6.72%)^***	0.4043(2.87%)^**	0.4626(0.67%)	0.4564(1.97%)^**
MM-BNCF	0.3605	0.4159	0.4657	0.4654