Multi-view Subspace Clustering Algorithm Based On Bi-level Optimization And Its Application

The fact that multi-view data can provide richer information to represent their underlying structure for the same dataset as opposed to single-perspective data has received considerable attention in the clustering problem.Traditional approaches to multi-view data clustering typically assume a complete feature set for all viewpoint samples.However,in actual applications,it often occurs that the sample loses the feature set in some perspectives,resulting in a significant degradation of the clustering performance,which leads to the problem of partial multi view clustering.The subspace approach can map data of high-dimensional features to a low-dimensional subspace,which can better handle multi-view data and has a widespread application in multiview learning.However,most such work combines various regularization terms from multiple perspectives into a single-layer objective,resulting in extremely complex optimization models that do not guarantee convergence,especially in partial data cases.In response to the above questions,this paper proposes a Bi-level Collaborative Factorization(BCF)framework to improve the limitations of existing partial multi-view clustering approaches based on factor decomposition.The BCF model divides the data into shared and partial samples and models matrix decomposition across multiple views and under a single view,respectively,in the upper and lower layers.An average-type iterative scheme was designed to derive the BCF algorithm,and its convergence was theoretically demonstrated.Numerous experimental results on the benchmark dataset show that the BCF algorithm proposed in this paper is superior to other algorithms.Public sentiment analysis is one of the important application areas of clustering problem.For the study of public sentiment of news,it is necessary to further analyze the news from the angle of headline,content,topic,etc.Therefore,the partial multi-view clustering algorithm has important significance for public sentiment analysis.In this paper,we apply a partial multiview subspace clustering algorithm based on double-layer optimization to public opinion analysis,and the results show that the algorithm in this paper can process multi-view news data and play an important role in public sentiment analysis.