A Feedback Clustering Method Based on Semantic Feature Analysis of Clusters

Abstract

The invention belongs to the technical field of data analysis and mining, and particularly relates to a feedback clustering method on the basis of cluster semantic feature analysis. The feedback clustering method includes steps of 1), acquiring the optimal attribute weights according to feedback attribute weighted K-means clustering; 2), acquiring the optimal clustering results according to the cluster semantic feature analysis. The feedback clustering method has the advantages that the attribute attention of users can be reflected by attribute weights, representative, distinguishable and intelligible attribute items can be selected from the clustering results by the aid of cluster semantic feature analysis processes and can be demonstrated, and accordingly the problem of unintelligible clustering results can be solved for the users; domain knowledge, experience and business analysis targets of the users are blended on the basis of the clustering results, weights are adjusted on the basis of improved particle swarm algorithms, and accordingly the attribute weights can be optimized; clustering numbers can be optimized, the optimized weights and the optimized clustering numbers are clustered again, and accordingly clustering results which meet the requirements of the analysis targets of the users can be obtained.

Description

[0001] Technical field [0002] The invention belongs to the technical field of data analysis and mining, and in particular relates to a feedback clustering method based on cluster semantic feature analysis. [0003] technical background [0004] With the advent of the era of big data, using data analysis technology and data mining technology to analyze and predict a certain situation, and provide various decision-making and service support has become one of the research focuses of big data applications. However, different users may often have different analytical business goals. For example, in the field of medical and health care, pharmaceutical personnel are more concerned about the relationship between diabetes and drug resistance, while clinicians are more concerned about how long it takes for diabetic patients to be accompanied by other drug reactions. In this way, for different analysis objectives, the evaluation of clustering results is different. This requires evalua...

AI Technical Summary

Problems solved by technology

Although there are many existing clustering methods, there is still a lack of targeted research and comprehensive consideration on the interpretation of clustering results and clustering optimization problems based on user feedback, and the use effect is not good.

Images