Multi-label Learning Based On Neighborhood Models

Multi-label learning is a complex decision-making task,and the same object may belong to more than one category at the same time.Such tasks are widely used in text classification,image recognition,gene function analysis and so on,which is one of the hot issues in the field of international machine learning.The research of multi-label learning mainly focuses on reducing the complexity of feature space and tag space,and improving the accuracy of multi-label learning algorithm.The neighborhood rough set is an extension and extension of Pawlak classical rough set,and it can get rid of the constraint that rough set can only deal with nominal data.The neighborhood rough set model has a clear clas-sification boundary,which can better excavate the classification task structure.It has been widely applied in feature selection,rule learning and classifier designing and other fields in recent years.However,in the multi-label learning,neighborhood rough set model has not been extended.In this paper,we use the characteristics of neighborhood rough set model,and apply it to the feature selection and rule learning of multi-label learning problem.The main research results and innovations are as follows:Firstly,this paper proposes a multi-label learning feature selection method based on neighborhood rough set model.The neighborhood rough set model of single-label learning is extended to multi-label learning,and a neighborhood rough set model is constructed.The properties of the multi-label learning neighborhood rough set model are discussed.The monotonicity of the dependence and the lower approximation is proved.Then the forward greedy search strategy is designed.At the same time,two kinds of acceleration mechanism are introduced to improve the efficiency of the model.The experimental results show that the model has significant advantages in multi-label learning tasks such as image,text and audio.Secondly,this paper proposes a classification rule learning method for multi-label learning based on neighborhood coverage reduction.In general,the neighborhood of a sample is to use uniform parameters to control the neighborhood radius of the sample,while in the neighborhood coverage,different samples can use different neighborhood radii.In this paper,neighborhood coverage is defined in multi-label learning by using the property of neighborhood coverage.The final classification rules are obtained according to the coverage reduction obtained for each class.