Research On Multi-view And Multi-label Learning Algorithm Based On Decision Fusion Strategy

In the era of the mobile Internet,not only has data information increased rapidly in scale,but its expression has also shown diversity and richness.People can often describe an object from multiple angles,that is,each object can not only be represented by the data features of multiple views,but also have multiple class labels.Multi-view multi-label learning is an important framework for studying such data problems.Because of its widespread existence in real life,it has become one of the focuses of machine learning research today.This paper focuses on the problem of multi-view multi-label learning,with decision fusion of multi-view multi-label learning algorithms as the core,aiming to improve the classification of the algorithm by effectively using the unique information of each view and the correlation between multiple views performance.The main research contents of this article are as follows:1.Aiming at the existing methods that tend to ignore the challenges caused by the complementarity and correlation of information between different data views,a multi-view multi-label learning algorithm based on decision fusion is proposed.This method comprehensively considers the effects of view-specific features,multi-view consensus,label correlation,and the contribution weight of each view on the model classification performance.First,learn the low-dimensional characteristic matrix of each data view.Then,use label correlation and multi-view consensus to build a multi-label classification model in each view,and learn the contribution weight of each view to the multi-label learning task.Finally,the multi-label prediction results in each view and the view contribution weights are fused to obtain the final label prediction result.Experiments are conducted on 5 public multi-view multi-label data sets,and the results show that the comprehensive performance of the algorithm is better.2.The existing methods lack communication between different views,so it is difficult to mine the real view sharing information,which is easy to cause poor model classification performance.To address this problem,a multi-view based on view sharing and unique features is proposed Multi-label learning algorithm.This method divides the sample features into two parts: shared features and specific features.The adversarial neural network is used to extract the shared features of the view,and the clustering integration method is used to extract the specific features of the view.This method not only enhances the communication between different views,but also maintains its own unique information.Experiments on a real multi-view multi-label data set verify the effectiveness of this method.