Multi-label Learning Based On Kernel Extreme Learning Machine Autoencoder

Initially,multi-label learning was to solve the problem of semantic divergence encountered in the process of document classification.Since its introduction,it has gradually become an important topic in data mining and information retrieval.Many classification problems encountered in real life are related to multi-label.The main feature of multi-label learning is to increase the type of sample classification,from the previous single label to multiple labels,making the description of the sample more accurate.In multi-label learning,the effective use of limited sample space information has been the research direction of many scholars.In order to improve the robustness of the algorithm and the efficiency of extracting data features,many scholars have made new optimization suggestions in different aspects.Research shows that feature reconstruction can improve the performance of the algorithm to a certain extent,so the information of feature space and label space is unified.Aiming at the complex calculation process of the traditional self-encoding neural layer,which makes the algorithm time complexity relatively large,a nuclear limit self-encoder is introduced.In the real world,there is often a certain implicit relationship between tags,and analyzing the relationship between tags is also a hotspot of multi-label learning research.Based on this,this article starts research,the main work is as follows:(1)The difficulty of the multi-label learning algorithm is how to accurately obtain the relationship between the label information of each group of samples to be trained.On this basis,to obtain the prediction results of the unknown data set,in simple terms,it is to extract the data association with the existing data set Relationship,and make predictions to get prediction results.Considering that the features reconstructed by features and labels can improve the classification performance of the algorithm,a kernel extreme learning self-encoding algorithm(ML-KELMAE)is proposed.In this,a kernel extreme learning machine self-coding neural network is used.First,label information is added to the input nodes of the neural network,then the nuclear extreme learning machine self-coding neural network is used to take the input features as the target output,and finally the singular value decomposition is used to solve the classification problem.The results on multiple multi-label benchmark data sets show that this method has certain advantages.(2)In the real world,there is often a certain hidden relationship between the individuals in the tag set and the individuals.This relationship between them has a more or less certain influence on the classification results,etc.It is very necessary to add relationship factors to the consideration.Therefore,we consider adding label correlation on the basis of the ML-KELMAE algorithm,adding label correlation information to the original label matrix,and the obtained matrix replaces the original matrix.On the basis of the optimization of the correlation characteristics of the label by the algorithm,the consideration of the correlation factors between the label and the features is added.