Multi-Label Classification: An Overview

By Grigorios Tsoumakas, Ioannis Manousos Katakis. 2007

2,985 citations recorded by the source corpus.

Multi-label classification methods are increasingly required by modern applications, such as protein function classification, music categorization, and semantic scene classification. This article introduces the task of multi-label classification, organizes the sparse related literature into a structured presentation and performs comparative experimental results of certain multilabel classification methods. It also contributes the definition of concepts for the quantification of the multi-label nature of a data set.

Subfields: Artificial Intelligence, Molecular Biology, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Molecular Biology

In Evaluation & Benchmarks

Citations in the corpus

node2vec: Scalable Feature Learning for Networks (2016, background): “In a typical node classification task, we are interested in predicting the most probable labels of nodes in a network [33].”
Anchors: High-Precision Model-Agnostic Explanations (2018, background): “Even for a “simpler” output space such as the one present in the multi-label classification setting (Tsoumakas and Katakis 2006), it is not clear if the best option would be to explain each label individually or the set of predicted labels as a single label.”
Optimal Decision Trees for the Algorithm Selection Problem: Integer Programming Based Approaches (2019, background): “…for each instance, this modeling does not capture some important problem aspects: firstly, it is often the case that many algorithms may produce equivalent results for a given instance, this aspect can be fixed by working with multi-label classification algorithms (Tsoumakas and Katakis, 2007).”
DOC: Deep Open Classification of Text Documents (2017, background): “Note that although multi-label classification (Huang et al., 2013; Zhang and Zhou, 2006; Tsoumakas and Katakis, 2006) may also leverage multiple sigmoid functions, Eq.”
Structured Prediction Energy Networks (2015, background): “Section A.4 describes details for improving the accuracy and efficiency of SPENs in practice.”
Multi-Task CNN Model for Attribute Prediction (2015, background): “If the problem is adapted as is, a challenge arises when the number of labels increases and the potential output label combinations become intractable [54].”
ETHOS: a multi-label hate speech detection dataset (2020, methodology): “or generalised, as well as the HS category to which it is oriented. 4.4.1 Algorithms. The algorithms handling multi-label data can be either problem transformation techniques or adaptation approaches [40]. MlkNN [50] and MLARAM [5] adaptation methods, as well as Binary Relevance (BR) and Classifier ”
Reinforcement Learning Via Practice and Critique Advice (2010, positive): “The multi-label learning problem (Tsoumakas and Katakis 2007) differs in that the goal is to learn a classifier that outputs all of the labels in sets and no others.”
Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning (2017, methodology): “Because these attributes are not mutually exclusive (especially for images containing multiple individuals), this task is a multilabel classification (55, 56) problem.”
Multilabel Classification using Bayesian Compressed Sensing (2012, background): “Most of the existing multilabel methods learn a decision function or weight vector per label and then combine the decision functions in a certain manner to predict labels for an unseen point [3, 4, 2, 5, 6].”
Ensemble deep learning: A review (2021, background): “This kind of classification problem is named multi-label classification [188].”
Learning shared subspace for multi-label dimensionality reduction via dependence maximization (2015, methodology): “Some prominent examples include binary relevance method [7], pair-wise method [16,17] and label embedding method [18–20].”
Statistical topic models for multi-label document classification (2011, methodology): “Most discriminative approaches to multi-label classification have employed some variant of the “binary problemtransformation” technique, in which the multi-label classification problem is transformed into a set of binary-classification problems, each of which can then be solved using a suitable bina”
iAMP-2L: a two-level multi-label classifier for identifying antimicrobial peptides and their functional types. (2013, background): “It is instructive to point out that, for a multi-label system like the current one, the absolute-true success rate for each of the individual AMP functional types is meaningless and misleading [62,70].”
Kernel Methods and Machine Learning (2014, methodology): “Typical methods include label powerset (LP) [276], binary relevance (BR) [275], ensembles of classifier chains (ECC) [216], and compressive sensing [102].”
Block-Row Sparse Multiview Multilabel Learning for Image Classification (2016, background): “Here, we follow [39], [40] to categorize the previous multilabel learning”
Bayes Optimal Multilabel Classification via Probabilistic Classifier Chains (2010, methodology): “…apart from the fact that complexity is not our main concern here, one may argue that several other methods suffer from the same problem, including the label power-set approach (Tsoumakas & Katakis, 2007), as also graphical models estimating the joint distribution (Ghamrawi & McCallum, 2005).”