Research On Personalized Recommendation Algorithms For Medical Health

With the development of medical information and Internet technology,a large number of medical health data have been generated.Relying on data analysis and mining useful information has become an important means to promote the development of medical health field.Recommendation system is an important application in this field.In this paper,we recommend drugs and nursing measures according to drug data and nursing measures data,so as to assist drug research and reduce the threshold of nursing measures.The specific work is as follows:1)Data acquisition and preprocessing of drugs and health care measures.This paper processed the open drug data set,obtained the required data,and labeled the disease characteristics;In addition,extracted the symptoms characteristic information and corresponding nursing measures from two health care guidelines,and analysed the characteristics of health nursing dataset through statistical analysis of data.2)A hybrid recommendation algorithm for drug repositioning based on the association of drug and disease characteristics is proposed.On the basis of collaborative filtering recommendation and demographic recommendation,the drug-disease characteristic matrix is constructed by introducing disease feature vectors.Finally,the similarity of drugs is calculated by using this matrix,so as to discover the curative effect of drugs and compare it with the actual curative effect of drugs.The comparison experiment shows that the algorithm reduced the impact of data sparsity and the recommendation effect is greatly improved,the overall recommendation accuracy rate is more than 95%.3)A recommendation algorithm of nursing measures was proposed.Using natural language toolkit and training word2 vec model to calculate text similarity to measure the similarity between symptoms,for a target symptom,two methods are used to find neighbor sets and mix them to produce final recommendation,and a comparative experiment is conducted on health care data sets.The experimental results showed that the hybrid method has obvious advantages,the overall recommendation accuracy is 18% higher than that of the natural language toolkit alone and 25% higher than that of the word 2vec model alone.