It is a key challenge to exploit the label coupling relationship in multi-label classification(MLC)problems.Most previous work focused on label pairwise relations,in which generally only global statistical information...It is a key challenge to exploit the label coupling relationship in multi-label classification(MLC)problems.Most previous work focused on label pairwise relations,in which generally only global statistical information is used to analyze the coupled label relationship.In this work,firstly Bayesian and hypothesis testing methods are applied to predict the label set size of testing samples within their k nearest neighbor samples,which combines global and local statistical information,and then apriori algorithm is used to mine the label coupling relationship among multiple labels rather than pairwise labels,which can exploit the label coupling relations more accurately and comprehensively.The experimental results on text,biology and audio datasets shown that,compared with the state-of-the-art algorithm,the proposed algorithm can obtain better performance on 5 common criteria.展开更多
In this paper a novel coupled attribute similarity learning method is proposed with the basis on the multi-label categorical data(CASonMLCD).The CASonMLCD method not only computes the correlations between different at...In this paper a novel coupled attribute similarity learning method is proposed with the basis on the multi-label categorical data(CASonMLCD).The CASonMLCD method not only computes the correlations between different attributes and multi-label sets using information gain,which can be regarded as the important degree of each attribute in the attribute learning method,but also further analyzes the intra-coupled and inter-coupled interactions between an attribute value pair for different attributes and multiple labels.The paper compared the CASonMLCD method with the OF distance and Jaccard similarity,which is based on the MLKNN algorithm according to 5common evaluation criteria.The experiment results demonstrated that the CASonMLCD method can mine the similarity relationship more accurately and comprehensively,it can obtain better performance than compared methods.展开更多
The adaptive local hyperplane (ALH) algorithm is a very recently proposed classifier, which has been shown to perform better than many other benchmarking classifiers including support vector machine (SVM), K-nearest n...The adaptive local hyperplane (ALH) algorithm is a very recently proposed classifier, which has been shown to perform better than many other benchmarking classifiers including support vector machine (SVM), K-nearest neighbor (KNN), linear discriminant analysis (LDA), and K-local hyperplane distance nearest neighbor (HKNN) algorithms. Although the ALH algorithm is well formulated and despite the fact that it performs well in practice, its scalability over a very large data set is limited due to the online distance computations associated with all training instances. In this paper, a novel algorithm, called ALH-Fast and obtained by combining the classification tree algorithm and the ALH, is proposed to reduce the computational load of the ALH algorithm. The experiment results on two large data sets show that the ALH-Fast algorithm is both much faster and more accurate than the ALH algorithm.展开更多
基金Supported by Australian Research Council Discovery(DP130102691)the National Science Foundation of China(61302157)+1 种基金China National 863 Project(2012AA12A308)China Pre-research Project of Nuclear Industry(FZ1402-08)
文摘It is a key challenge to exploit the label coupling relationship in multi-label classification(MLC)problems.Most previous work focused on label pairwise relations,in which generally only global statistical information is used to analyze the coupled label relationship.In this work,firstly Bayesian and hypothesis testing methods are applied to predict the label set size of testing samples within their k nearest neighbor samples,which combines global and local statistical information,and then apriori algorithm is used to mine the label coupling relationship among multiple labels rather than pairwise labels,which can exploit the label coupling relations more accurately and comprehensively.The experimental results on text,biology and audio datasets shown that,compared with the state-of-the-art algorithm,the proposed algorithm can obtain better performance on 5 common criteria.
基金Supported by Australian Research Council Discovery(DP130102691)the National Science Foundation of China(61302157)+1 种基金China National 863 Project(2012AA12A308)China Pre-research Project of Nuclear Industry(FZ1402-08)
文摘In this paper a novel coupled attribute similarity learning method is proposed with the basis on the multi-label categorical data(CASonMLCD).The CASonMLCD method not only computes the correlations between different attributes and multi-label sets using information gain,which can be regarded as the important degree of each attribute in the attribute learning method,but also further analyzes the intra-coupled and inter-coupled interactions between an attribute value pair for different attributes and multiple labels.The paper compared the CASonMLCD method with the OF distance and Jaccard similarity,which is based on the MLKNN algorithm according to 5common evaluation criteria.The experiment results demonstrated that the CASonMLCD method can mine the similarity relationship more accurately and comprehensively,it can obtain better performance than compared methods.
文摘The adaptive local hyperplane (ALH) algorithm is a very recently proposed classifier, which has been shown to perform better than many other benchmarking classifiers including support vector machine (SVM), K-nearest neighbor (KNN), linear discriminant analysis (LDA), and K-local hyperplane distance nearest neighbor (HKNN) algorithms. Although the ALH algorithm is well formulated and despite the fact that it performs well in practice, its scalability over a very large data set is limited due to the online distance computations associated with all training instances. In this paper, a novel algorithm, called ALH-Fast and obtained by combining the classification tree algorithm and the ALH, is proposed to reduce the computational load of the ALH algorithm. The experiment results on two large data sets show that the ALH-Fast algorithm is both much faster and more accurate than the ALH algorithm.
