Research On Micro-video Multi-label Classification Based On Deep Matrix Factorization

With the coming of the big data era,various new multimedia information products emerge in endlessly.As the most typical new media product,micro-video is more and more popular because it adapts to the needs of mass fragmented consumption.Microvideo data often contains very rich information,how to use intelligent technology for in-depth data mining,will be an important factor to promote the rapid development of micro-video industry.Multi-label classification is one of the basic direction of research on micro-video content understanding,and based on the deep matrix factorization framework,we introduce the idea of the codec used to further excavate the potential information in the micro-video.We hope to obtain high-level semantic features hidden in micro-video to improve the accuracy of micro-video multi-label classification tasks.The main work of this paper includes:In order to solve the problem that the complex information contained in microvideo data is difficult to accurately describe,this paper proposes a Deep Matrix Factorization with Auto-Encoder Decoder Framework.On the basis of deep matrix factorization,the codec module is introduced to mine potential semantic features and enhance the ability of classification algorithm to learn potential features.The low-rank learning strategy is introduced to eliminate redundant information in potential features and improve the generalization ability of the whole model.In order to improve the performance of the algorithm in micro-video multi-label classification,inverse matrix is constructed to learn the correlation between labels and the direction of learning potential features guided by triples with supervised discriminant constraints.Finally,experimental tests on MLSV2018 micro-video multi-label dataset confirm the effectiveness of the proposed algorithm.In order to solve the problem of shallow mining of the association between multilabel in micro-video,this paper proposes a Deep Matrix Factorization with Adaptive Graph Learning.On the basis of the deep matrix factorization codec module and lowrank learning module,the algorithm uses the feature of adaptive graph learning to mine the correlation between multi-label of micro-video with the help of visual features,and further improves the generalization ability of reconstruction matrix obtained by multilayer factorization of visual features.Finally,experiments on MLSV2018 micro-video multi-label dataset confirm the excellent performance of this algorithm.