Based On Decision Relevance Multi-label Classification And Feature Selection Algorithm

In single-label learning task,each instance in the data set is associated with only one label.However,each instance is associated with a subset of labels simultaneously in multi-label learning task.The demand for multi-label classification methods continue to grow in many modern applications,such as document classification,the gene function categorization,semantic scene classification and so on.Classification and feature selection method are the two fundamental problems in multi-label learning region.In classification method,the paper proposes the method of multi-label fuzzy similarity-based nearest-neighbour classification using association rule(MLSNN and MLASNN).Firstly,in order to reduce the scale of label and avoid the label overlapping phenomenon,the association rule approach is employed to make the combination labels collapse to a set of sub-labels.Then by transforming the multi-label training data into the single-label representation data,the fuzzy similarity-based nearest-neighbour methods perform the classification label prediction.Finally,the resulting label set is the union of the predicted labels and their associated labels according to the extracted association rules.Empirical results suggest that the proposed approach can improve the performance compared with other multi-label classification algorithms.In feature selection method,the paper proposes the method of using association rules in the labels to implement a fuzzy-rough feature selection method for multi-label data set.Specifically,in order to reduce the scale of label and avoid the label overlapping phenomenon,the association rules between labels make the com'bination labels collapse to a set of sub-labels.Then each set of sub-labels is regarded as a unique class during the course of fuzzy-rough feature selection.Empirical results suggest that the proposed approach can improve the performance compared with other standard feature selection algorithms for multi-label problems.Conversion rate prediction model is an important multi-label learning task in the field of computational advertising.In order to improving the accuracy and comprehensiveness of the conversion rate predicted by model,the paper proposes a conversion prediction algorithm based on decision relevance.Firstly,for fully integrating the information of all conversion components,the weight of each conversion component is obtained by decision association rules.Then,conversion components are melt by label weight.Finally,the neural network model are trained by the label-fused instances to predict the conversion rate.The A/B test empirical results show that the proposed algorithm is superior to the traditional conversion prediction algorithm in terms of related metrics.