摘要
[目的/意义]针对在线问诊平台中医生推荐满意度较低的问题,探究如何将信息技术与用户认知相结合以提升医生推荐系统的效果,有助于优化在线问诊平台的用户体验。[方法/过程]首先,基于1500名医生的基本信息和78万余条用户提问,对比TF-IDF、Doc2Vec和Word2Vec三种词向量模型的医生推荐效果,以最优模型构建医生推荐系统原型:然后,通过用户实验和访谈获取用户使用该系统的行为数据,深入挖掘在线问诊平台医生推荐情境中的用户认知与意义构建过程;最后,从用户角度提出模型优化思路,实现原型系统的改进。[结果/结论]基于Word2Vec词向量模型的医生推荐效果最优,前10位医生候选集中88%的医生有能力回答用户问题:用户实验结果显示,科室信息与医生专业极大影响用户选择,医生曾回答过的相似问题是用户的重要参考信息:基于以上结果,提出并实现建立科室预测分类器以及为健康医学关键词赋予较高权重的两种模型优化思路,并通过匹配度指数对医生推荐结果进行优化排序。结果表明,两种方法均可提高医生推荐系统的准确度,证明用户认知与人工智能算法结合具有可行性。
[Purpose/Significance]In view of the low satisfaction of online doctor recommendation,this paper explores how to combine information technology and user cognition to improve the effect of the doctor recommendation system,which helps to optimize the user experience of online“Ask the Doctor”platform.[Method/Process]First,we established a doctor recommender prototype system based on the relevance theory and the NLP method based on the information of 1500 doctors and more than 780 thousands user questions;Then,did a qualitative study to analyze user's thoughts in the process of using the recommender based on the sense-making;Finally,we optimized the recommender though considering the users'perspectives.[Result/Conclusion]Word2Vec model has the best effect in the doctor recommendation task,which was up to 88%doctors in TOP10 doctor candidates are able to answer user questions.The user experiment results show that most users attach great importance to the doctor's department and areas of expertise while similar questions answered by doctors.When judging the similarity of questions,users mainly pay attention to the medical terms,and avoid the rrelevant medical keywords.Based on these,two model optimizations were identified,including(1)a function of predicting departments was incorporated into the system,and doctors belonging to these departments were ranked forward,(2)a healthcare wordlist was built and higher weights were given to these words when calculating text similarity.Results show that these two methods improved the accuracy of the doctor recommender system,which indicates that the integration of the Al-related algorithms and the user's thoughts can be well implemented.
作者
王若佳
王继民
Wang Ruojia;Wang Jimin(School of Management,Beijing University of Traditional Chinese Medicine,Beijing 100105;Department of Information Management,Peking University,Beijing 100871)
出处
《图书情报工作》
北大核心
2023年第10期128-138,共11页
Library and Information Service
