## Multi-Label Collective Classification

Citations: | 4 - 1 self |

### BibTeX

@MISC{Kong_multi-labelcollective,

author = {Xiangnan Kong and Xiaoxiao Shi and Philip S. Yu},

title = {Multi-Label Collective Classification},

year = {}

}

### OpenURL

### Abstract

Collective classification in relational data has become an important and active research topic in the last decade, where class labels for a group of linked instances are correlated and need to be predicted simultaneously. Collective classification has a wide variety of real world applications, e.g. hyperlinked document classification, social networks analysis and collaboration networks analysis. Current research on collective classification focuses on single-label settings, which assumes each instance can only be assigned with exactly one label among a finite set of candidate classes. However, in many real-world relational data, each instance can be assigned with a set of multiple labels simultaneously. In this paper, we study the problem of multi-label collective classification and propose a novel solution, called Icml (Iterative Classification of Multiple Labels), to effectively assign a set of multiple labels to each instance in the relational dataset. The proposed Icml model is able to capture the dependencies among the label sets for a group of related instances and the dependencies among the multiple labels within each label set simultaneously. Empirical studies on real-world tasks demonstrate that the proposed multi-label collective classification approach can effectively boost classification performances in multilabel relational datasets. 1

### Citations

3800 | LIBSVM: a library for support vector machines. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm
- Chang, Lin
- 2001
(Show Context)
Citation Context ...oaches considers the high-order correlations among different labels. Such approaches includes random subset ensemble approaches [21, 22], Bayesian network based approach [28] and fullorder approaches =-=[4, 5, 2]-=-. Collective classification of single-label relational data has also been investigated by many researchers. The task is to predict the classes for a group of related instances simultaneously, rather t... |

529 | BoosTexter: A boosting-based system for text categorization
- Schapire, Singer
(Show Context)
Citation Context ... categories (As shown in Figure 2(b), xi denotes the i-th instance, {Y j i } is the set of labels assigned to xi). It has been shown useful in many real-world applications such as text categorization =-=[16, 23]-=- and bioinformatics [6]. 618 Copyright © SIAM. Unauthorized reproduction of this article is prohibited.• • • • • • • • • • • • • • • • Formally, a multi-label collective classification problem corres... |

360 | Discriminative probabilistic models of relational data
- Taskar, Abbeel, et al.
(Show Context)
Citation Context ...data mining approaches assume that instances are independent and identically distributed, and each testing instance is predicted with a class label independently. However, in many relational datasets =-=[25]-=- or information networks [8], the instances are implicitly or explicitly related, with complex dependencies. For example, in collaboration networks, the researchers who collaborate with each other are... |

146 | A kernel method for multi-labeled classification
- Elisseeff, Weston
- 2001
(Show Context)
Citation Context ...ure 2(b), xi denotes the i-th instance, {Y j i } is the set of labels assigned to xi). It has been shown useful in many real-world applications such as text categorization [16, 23] and bioinformatics =-=[6]-=-. 618 Copyright © SIAM. Unauthorized reproduction of this article is prohibited.• • • • • • • • • • • • • • • • Formally, a multi-label collective classification problem corresponds to predicting the... |

146 |
Link-based classification
- Lu, Getoor
- 2003
(Show Context)
Citation Context ...ntion in the last decade, where a group of related instances are classified simultaneously rather than independently. In the literature, collective classification problem has been extensively studied =-=[19, 25, 15, 27]-=-. Conventional approaches focus on single-label classification problems, which assume that each instance in the relational dataset has only one label among a finite set of candidate classes (As shown ... |

141 | Multi-label Text Classification with a Mixture Model Trained by EM - McCallum - 1999 |

131 | Iterative Classification in Relational Data
- Neville, Jensen
- 2000
(Show Context)
Citation Context ...ntion in the last decade, where a group of related instances are classified simultaneously rather than independently. In the literature, collective classification problem has been extensively studied =-=[19, 25, 15, 27]-=-. Conventional approaches focus on single-label classification problems, which assume that each instance in the relational dataset has only one label among a finite set of candidate classes (As shown ... |

112 | Learning multi-label scene classification
- Moutell, Luo, et al.
- 2004
(Show Context)
Citation Context ...oaches: The first type of approaches assumes all different labels are independent, which converts the multi-label problem into multiple independent binary classification problems (one for each label) =-=[1]-=-. Ml-knn[29] is one of the binary methods, which extends the kNN algorithm to a multilabel version using maximum a posteriori (MAP) principle to determine the label set predictions. (2) pairwise appro... |

107 | Collective classification in network data
- Sen, Namata, et al.
- 2008
(Show Context)
Citation Context ...ttributes and relational features for inference [25]. For a detailed 620 Copyright © SIAM. Unauthorized reproduction of this article is prohibited.review of collective classification please refer to =-=[24]-=-. 3 Problem Definition Before presenting the collective classification model for multi-label relational data, we first introduce the notations that will be used throughout this paper. Suppose we are g... |

78 | ML-KNN: A lazy learning approach to multi-label learning
- Zhang, Zhou
(Show Context)
Citation Context ... first type of approaches assumes all different labels are independent, which converts the multi-label problem into multiple independent binary classification problems (one for each label) [1]. Ml-knn=-=[29]-=- is one of the binary methods, which extends the kNN algorithm to a multilabel version using maximum a posteriori (MAP) principle to determine the label set predictions. (2) pairwise approaches: The s... |

73 | Collective multi-label classification
- Ghamrawi, McCallum
- 2005
(Show Context)
Citation Context ...ersion using maximum a posteriori (MAP) principle to determine the label set predictions. (2) pairwise approaches: The second type of approaches exploit the pairwise relation between different labels =-=[9]-=-. For example, Elisseeff and Weston [6] presented a kernel method Rank-svm by minimizing a loss function named ranking loss to properly rank label pairs. (3) High-order approaches: The third type of a... |

71 | Parametric Mixture Models for Multi-Labeled Text
- Ueda, Saito
- 2002
(Show Context)
Citation Context ...classification on relational data. We briefly discuss both of them. Multi-label learning deals with the classification problem where each instance can belong to multiple differentclassessimultaneously=-=[26,10,12,7,3,13]-=-. The goal of multi-label classification is to predict each instance with a set of multiple labels in the space of all label sets, i.e. the power set of all labels, which is exponential to the number ... |

61 | Discriminative methods for multi-labeled classification
- Godbole, Sarawagi
- 2004
(Show Context)
Citation Context ...classification on relational data. We briefly discuss both of them. Multi-label learning deals with the classification problem where each instance can belong to multiple differentclassessimultaneously=-=[26,10,12,7,3,13]-=-. The goal of multi-label classification is to predict each instance with a set of multiple labels in the space of all label sets, i.e. the power set of all labels, which is exponential to the number ... |

60 | Collective classification with relational dependency networks
- Neville, Jensen
- 2003
(Show Context)
Citation Context ...e based approaches [19, 15] and Gibbs sampling approaches [18]. Many local classifiers have been used for local methods, e.g. logistic regression [15], Naive Bayes [19], relational dependency network =-=[20]-=-, etc. (2) Global methods: The second type of approaches optimizes global objective functions on the entire relational dataset, which also uses both attributes and relational features for inference [2... |

55 | Random k-labelsets: An ensemble method for multilabel classification
- Tsoumakas, Vlahavas
(Show Context)
Citation Context ...classification on relational data. We briefly discuss both of them. Multi-label learning deals with the classification problem where each instance can belong to multiple differentclassessimultaneously=-=[26,10,12,7,3,13]-=-. The goal of multi-label classification is to predict each instance with a set of multiple labels in the space of all label sets, i.e. the power set of all labels, which is exponential to the number ... |

47 | Classifier chains for multi-label classification
- Read, Pfahringer, et al.
- 2009
(Show Context)
Citation Context ...erly rank label pairs. (3) High-order approaches: The third type of approaches considers the high-order correlations among different labels. Such approaches includes random subset ensemble approaches =-=[21, 22]-=-, Bayesian network based approach [28] and fullorder approaches [4, 5, 2]. Collective classification of single-label relational data has also been investigated by many researchers. The task is to pred... |

35 | Maximal margin labeling for multi-topic text categorization
- Kazawa, Izumitani, et al.
- 2005
(Show Context)
Citation Context |

34 | Semi-supervised multi-label learning by constrained non-negative matrix factorization
- Liu, Jin, et al.
- 2006
(Show Context)
Citation Context ...luation Metrics The performance evaluation for multi-label classification requires more complicated criteria than conventional single-label classification problems. Here we adopt some metrics used in =-=[9, 11, 14, 30, 5]-=- to evaluate the classification performance in a multi-label relational data. Assume we have a multi-label relational dataset DU containing n multi-label instances (xi, Yi), where Yi ∈{0, 1} q (i = 1,... |

29 | Correlated label propagation with application to multi-label learning
- Kang, Jin, et al.
- 2006
(Show Context)
Citation Context ...luation Metrics The performance evaluation for multi-label classification requires more complicated criteria than conventional single-label classification problems. Here we adopt some metrics used in =-=[9, 11, 14, 30, 5]-=- to evaluate the classification performance in a multi-label relational data. Assume we have a multi-label relational dataset DU containing n multi-label instances (xi, Yi), where Yi ∈{0, 1} q (i = 1,... |

29 | Cautious inference in collective classification
- McDowell, Gupta, et al.
- 2007
(Show Context)
Citation Context ...related instances as shown in Figure 5, i.e. “SLRelFeature” for inter-instance single-label dependencies and “CLRelFeature” for inter-instance cross-label dependencies. In the spirit of ICA framework =-=[15, 17, 18]-=-, the inference procedure of our Icml method has two parts: bootstrap and iterative inference asshowninFigure4. (1) At the beginning of the inference procedure, the label sets of all the unlabeled ins... |

27 | Bayes optimal multilabel classification via probability classifier chains
- Dembczynski, Cheng, et al.
- 2010
(Show Context)
Citation Context ...ed instances. Conventional multi-label classification approaches assume, explicitly or implicitly, that instances are unrelated and the label sets of the testing instances are predicted independently =-=[6, 28, 5]-=-. However in the context of relational data, the label sets of related instances are not independent, which should be predicted simultaneously. Multiple Labels: Another fundamental problem in multi-la... |

24 | Semi-supervised multi-label learning by solving a sylvester equation
- Chen, Song, et al.
(Show Context)
Citation Context |

23 | Combining instance-based learning and logistic regression for multilabel classification
- Cheng, Hüllermeier
- 2009
(Show Context)
Citation Context ...oaches considers the high-order correlations among different labels. Such approaches includes random subset ensemble approaches [21, 22], Bayesian network based approach [28] and fullorder approaches =-=[4, 5, 2]-=-. Collective classification of single-label relational data has also been investigated by many researchers. The task is to predict the classes for a group of related instances simultaneously, rather t... |

21 |
Multi-label classification using ensembles of pruned sets
- Read, Pfahringer, et al.
- 2008
(Show Context)
Citation Context ...erly rank label pairs. (3) High-order approaches: The third type of approaches considers the high-order correlations among different labels. Such approaches includes random subset ensemble approaches =-=[21, 22]-=-, Bayesian network based approach [28] and fullorder approaches [4, 5, 2]. Collective classification of single-label relational data has also been investigated by many researchers. The task is to pred... |

18 |
Multi-label learning by exploiting label dependency
- Zhang, Zhang
(Show Context)
Citation Context ...ed instances. Conventional multi-label classification approaches assume, explicitly or implicitly, that instances are unrelated and the label sets of the testing instances are predicted independently =-=[6, 28, 5]-=-. However in the context of relational data, the label sets of related instances are not independent, which should be predicted simultaneously. Multiple Labels: Another fundamental problem in multi-la... |

16 | On community outliers and their efficient detection in information networks
- Gao, Liang, et al.
(Show Context)
Citation Context ... that instances are independent and identically distributed, and each testing instance is predicted with a class label independently. However, in many relational datasets [25] or information networks =-=[8]-=-, the instances are implicitly or explicitly related, with complex dependencies. For example, in collaboration networks, the researchers who collaborate with each other are more likely to share simila... |

13 | Cautious collective classification
- McDowell, Gupta, et al.
(Show Context)
Citation Context ... approaches involves an iterative process to update the labels and the relational features of the related instances, e.g. iterative convergence based approaches [19, 15] and Gibbs sampling approaches =-=[18]-=-. Many local classifiers have been used for local methods, e.g. logistic regression [15], Naive Bayes [19], relational dependency network [20], etc. (2) Global methods: The second type of approaches o... |

10 | Multi-label feature selection for graph classification
- Kong, Yu
- 2010
(Show Context)
Citation Context |

7 | Multi-label classification without the multi-label cost
- Zhang, Yuan, et al.
- 2010
(Show Context)
Citation Context ...luation Metrics The performance evaluation for multi-label classification requires more complicated criteria than conventional single-label classification problems. Here we adopt some metrics used in =-=[9, 11, 14, 30, 5]-=- to evaluate the classification performance in a multi-label relational data. Assume we have a multi-label relational dataset DU containing n multi-label instances (xi, Yi), where Yi ∈{0, 1} q (i = 1,... |

3 |
Link Mining: Models, Algorithms, and Applications
- Yu, Han, et al.
- 2010
(Show Context)
Citation Context ...ntion in the last decade, where a group of related instances are classified simultaneously rather than independently. In the literature, collective classification problem has been extensively studied =-=[19, 25, 15, 27]-=-. Conventional approaches focus on single-label classification problems, which assume that each instance in the relational dataset has only one label among a finite set of candidate classes (As shown ... |