Research And Implementation Of Doctor Recommendation Algorithm Combining User Explicit And Implicit Similarity

At this stage,online medical consultation services play an important role in pandemic prevention and control.The problem of "information overload" caused by a large number of medical datas reduces the efficiency of users in choosing doctors.The retrieval technology adopted by the existing medical platform can not understand the accurate personalized needs of users,which violates the demands of current users to quickly match doctors.Personalized recommendation of doctors has become the most effective solution.Collaborative filtering,as one of the mainstream recommendation algorithms,has been studied and improved a lot in the field of doctor recommendation,but there are still some problems in the application process,such as user cold start,score difference and data sparseness.Based on the above background and problems,the main research work of this thesis is as follows:1.In view of the problem of user cold start and rating difference,this thesis improves the similarity calculation method based on explicit information.Which includes two aspects,firstly,for the coarse granularity problem of user attribute similarity calculation,user attributes are divided into binary attributes,nominal attributes and numerical attributes according to different data categories,and their respective calculation methods are optimized.Secondly,aiming at the problem of rating difference caused by user rating criteria and project quality difference,the emotion analysis algorithm based on support vector machine is used to calculate the emotional tendency of user comments to correct user rating,and the user preference coefficient and common concern coefficient are introduced to alleviate the impact of item quality difference on similarity calculation results.2.To solve the problem of sparse rating data,existing algorithms generally use side information such as medical consultation text to expand the user’s neighborhood.The use of side information decreases the applicability of the algorithm and increases the risk of leaking private data.Based on above analysis,this thesis improved the similarity calculation method based on implicit information from three aspects,firstly,the implicit similarity between users is calculated through the implicit information of user rating matrix and doctor attribute matrix,this does not require external information sources,but also solves the problem of data sparsity.Secondly,aiming at the rating difference under implicit information,the common concern coefficient and rating preference factor are introduced to improve the accuracy of implicit similarity calculation,so as to obtain a more reasonable nearest neighbor set.Thirdly,taking into account the spread of similarity between users,it can avoid the situation that the implicit similarity between users is zero when the data is very sparse.3.After the dynamic fusion of explicit similarity and implicit similarity,an improved doctor recommendation algorithm is obtained,which is verified by experiments on the corresponding data sets.The results show that the improved doctor recommendation algorithm can effectively improve the accuracy of the recommendation results without obtaining additional side information.In addition,the improved algorithm is applied to the actual project,the doctor recommendation system is designed and developed,and the usability of the system is verified by testing tools.