Joint Classification With Heterogeneous Labels Using Random Walk With Dynamic Label Propagation

As one of the main research fields of data mining,classification problem is more and more applied in practical engineering.In order to meet the increasing demands of classification,this paper proposes an optimized joint classification algorithm based on multiple classification requirements of the same data set.The aim of this paper is to ensure the classification performance,that is,the convergence speed of the algorithm is fast enough,and at the same time to enhance the effect of joint classification to achieve the purpose of improving classification efficiency.This paper proposes a semi-supervised learning algorithm named RWDLP,which can solve the problem of multiple classification tasks in the same dataset.We review the research status of classification problems at home and abroad,introduce some common classification algorithms and applications,and then start from the following aspects:Firstly,we analyze and discuss Markov-chain-based random walk(RW)Process and application,and label propagation algorithm and its improved version of the dynamic label propagation algorithm(DLP)ideas and processes.By analyzing the common points of these two algorithms,this paper presents a RWDLP algorithm combining the characteristics of the two algorithms.The algorithm is based on Markov chain random walk,but in the process of walking,combined with the idea of dynamic label propagation algorithm,the dynamic updating step of transition probability matrix and the idea of near similarity are added to make the classification process more efficient.Which effectively utilizes the information of the label set,and also ensures the convergence speed of the iteration through the restart technique.Then,the RWDLP algorithm is combined with the joint classification of the scene together.Firstly,a mix-relevance graph is constructed,which contains instances and multiple label sets.Then,a random walk based on Markov chain is walking in the graph.The advantage of the algorithm proposed in this paper is to make full use of the inner information of label set and the relevance information of heterogeneous label sets,and to combine two or more classification tasks.The efficiency of classification is improved under the premise of ensuring iterative convergence speed.Finally,in the actual bioinformatics data set,the validity of the proposed algorithm is proved by comparison with other classical algorithms.The proposed joint classification algorithm is proved to be effective and practicable through experiments.With the increasing application of classification,more and more problems and demands are encountered,and the requirement of the algorithm is more and more common.The RWDLP algorithm proposed in this paper just meets these requirements.