Design And Implementation Of Real-time Recommendation System For Medical Disputes Judgment Documents

The medical dispute judgment document(referred to as medical dispute case)is a legally effective document prepared by the Judicial Court in civil litigation,and it aims to resolve civil rights and obligations disputes between medical institutions and patients.Nowadays,the number of judgment documents for medical disputes is gradually increasing,and it is difficult for legal professionals to find suitable historical case references in a short period of time.At the same time,different users usually make different judgments on valid cases,and even users may have different judgments on valid cases at different stages.For these reasons,it is necessary to quickly and accurately analyze the existing medical dispute judgment documents,capture user preference changes in time,and make real-time recommendations of similar historical cases based on user preferences.This can provide scientific guidance for the judge’s trial,and can also help the parties understand the scope of their own responsibilities,and improve the efficiency of the mediation of medical disputes.However,the existing case recommendation system has the following problems:(1)Similarity calculation of medical dispute cases.The recommendation of similar cases requires the extraction of element information to calculate the case similarity.However,medical dispute cases are unstructured and long texts with scattered content and strong professionalism,it is difficult to accurately extract elements,which affects the accuracy of case similarity calculation.(2)Personalized recommendation for different legal users.The traditional user preference collection is to collect user feedback data through explicit feedback,but it is not guaranteed that users can feedback spontaneously in the actual scene.Therefore,it is difficult to collect enough feedback data,leading to inaccurate recommendation results.(3)Real-time recommendation of medical dispute cases.As the number of judgment documents and user feedback records continue to increase,the case similarity calculation and recommendation calculation are getting more and more time-consuming.The traditional centralized and offline batch computing system architecture has been unable to meet the needs of rapid calculation,and cannot perform real-time and accurate update recommendations.In order to solve the above problems,the specific research work of this thesis includes the following aspects:(1)Research on the calculation method of similarity of medical disputes cases.In order to solve the problem that the similarity calculation is not accurate due to the difficulty of case factor extraction,the case is firstly abstracted into a medical event that can be expressed structurally,the element information is extracted and the redundant information is filtered,a dynamic adjustment mechanism for event templates is provided to support the differential expression of different users’ attention to different elements;secondly,the BI-LSTM-CRF model was introduced to accurately extract case elements according to the medical event template;thirdly,the similarity calculation method of medical dispute cases based on medical events is proposed,and the candidate set of Top-K cases is returned according to the similarity ranking.(2)Research on personalized recommendation algorithm for medical dispute cases.To solve the problem of personalized recommendation for different legal users,firstly,the feedback behavior of users on the candidate set of similar cases is captured based on front-end buried point technology,and the user feedback behavior is quantified to generate the user’s rating on the case;Then,the recommendation model is used for iterative training and learning to obtain users’ attention to each element,and a personalized event template is formed for users.According to this template,users’ scores for cases are updated,and the Top-K case set is returned after sorting.(3)Realize the real-time recommendation system for medical dispute cases based on Spark.According to similar medical dispute case real-time recommendation problem,the real-time recommendation system of similar medical dispute case designed in this thesis is based on Spark distributed computing framework,and the system can process user feedback on the recommended list of history in time.The recommendation model is trained regularly to get the new weight of users’ attention to the elements,and the recommendation list is updated regularly accordingly to realize the dynamic update of recommendations and make the recommendation results close to users’ preferences continuously.Experimental results show that the calculation method of similarity of medical disputes cases based on medical events can improve the accuracy of the calculation results of case similarity to a certain extent.In the task of personalized recommendation of medical dispute cases,comparing the traditional matrix factorization recommendation algorithm with the element-based matrix factorization recommendation algorithm,the experiment shows that the element-based matrix factorization recommendation algorithm has higher accuracy.In the task of real-time recommendation of medical dispute cases,compared with the traditional centralized and offline batch calculation,the Spark based real-time recommendation system designed in this thesis has better timeliness than the system built on the traditional Hadoop distributed computing platform,which effectively improves the performance of the recommendation system.