| In recent years,with the popularization of medical informatization and the promotion of medical and health big data research,data storage methods in the medical field have also turned to electronic,and the popularity of medical big data research has also risen accordingly.As one of the three major diseases that threaten human health,diabetes has caused a lot of research on diabetes data mining or data analysis.The types of data stored in the diabetes resource library are complex,including structured,semi-structured,and unstructured data,and there is a lack of sensitivity to distinguish diabetes data related to privacy issues.In addition,in many studies on the diagnosis and prediction of diabetes,structured diabetes research data used for predictive analysis is not anonymized or over-anonymized.On the one hand,the unanonymized data is directly used for research,which may cause the problem of leakage of sensitive information in the analysis and prediction of diabetes data or in the process of data release.On the other hand,excessive anonymization of the diabetes data table will affect the effectiveness of the diabetes analysis and research.Therefore,balancing diabetes data mining,high availability of data in predictive analysis,and privacy security are key issues in privacy protection research.For diabetes data of different structures,it is necessary to implement privacy protection technology research on diabetes big data from sensitivity classification,sensitivity rating,and further anonymization.For semi-structured and unstructured textual diabetes data,this paper proposes a text classification technology based on LSI-TF-IDF algorithm to achieve automatic classification of the sensitivity of diabetes text data.In the sensitivity classification of semi-structured,unstructured text-type diabetes data,the characteristics of diabetes data sensitive information are defined by the confidentiality,integrity and availability of diabetes-sensitive data.By improving the TF-IDF algorithm,the accuracy of the feature selection process is improved.And combined with the three classification methods of Naive Bayes,K nearest neighbor and support vector machine,the improved feature selection algorithm is compared with the traditional feature selection algorithm experimentally.Experiments show that the improved LSI-TF-IDF algorithm has better results for the sensitivity automatic classification of diabetes text data.For structured diabetes data,this paper proposes to quantify the sensitivity values of the attributes in the data table and formulate a sensitivity grading strategy.On this basis,(k,t)-closeness anonymous algorithm based on sensitivity hierarchical clustering is proposed.Based on the structured diabetes data table,the sensitivity grading rules of the diabetes data table are designed by calculating the sensitivity threshold of the record and judging the sensitivity of the identifier attribute.By grading the sensitivity of the diabetes data table,the t-closeness anonymous algorithm is improved to ensure that the degree of data loss in data mining,analysis and prediction research is as small as possible.The experimental results show that the(k,t)-closeness anonymous algorithm based on sensitivity hierarchical clustering has less information loss for the anonymity of diabetes data tables. |
