Classifier chains for multi-label classification

By Jesse Read, Bernhard Pfahringer, G. Holmes, Eibe Frank. Machine-mediated learning, 2009

2,723 citations recorded by the source corpus.

The widely known binary relevance method for multi-label classification, which considers each label as an independent binary problem, has often been overlooked in the literature due to the perceived inadequacy of not directly modelling label correlations. Most current methods invest considerable complexity to model interdependencies between labels. This paper shows that binary relevance-based methods have much to offer, and that high predictive performance can be obtained without impeding scalability to large datasets. We exemplify this with a novel classifier chains method that can model label correlations while maintaining acceptable computational complexity. We extend this approach further in an ensemble framework. An extensive empirical evaluation covers a broad range of multi-label datasets with a variety of evaluation metrics. The results illustrate the competitiveness of the chaining method against related and state-of-the-art methods, both in terms of predictive performance and time complexity.

Subfields: Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/e75a71274522980dce85b8ad00369d0a33667ea4

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 Information Systems

Citations in the corpus

LEAP: Learning to Prescribe Effective and Safe Treatment Combinations for Multimorbidity (2017, methodology): “• Classifier Chains: Classifier Chains [28] is a popular multi-label learning method that models the correlation between labels by a feeding both input and previous classification results into the latter classifiers.”
Towards a taxonomic coherence between average nucleotide identity and 16S rRNA gene sequence similarity for species demarcation of prokaryotes. (2014, methodology): “The F measure was originally introduced for measuring classification performance in information retrieval processes (van Rijsbergen, 1979) and has been used frequently in assessing the performance of binary or multilabel classifiers (Lan et al., 2012; Read et al., 2011).”
Multi-label Zero-Shot Learning with Structured Knowledge Graphs (2017, methodology): “Approaches such as [38, 7] take cross-label correlation by as-”
Large-Scale Multi-label Text Classification - Revisiting Neural Networks (2013, background): “However, this so-called binary relevance approach ignores dependencies between the labels, so that current research in multi-label classification concentrates on the question of how such dependencies can be exploited [23, 3].”
Statistical topic models for multi-label document classification (2011, methodology): “…accurate classification in multi-label problems, yet has been problematic in the past for datasets with large numbers of labels, as summarized in Read et al. (2009): The consensus view in the literature is that it is crucial to take into account label correlations during the classification…”
Multi-label learning by exploiting label dependency (2010, methodology): “As shown in Table 7, for data sets with regular number of examples (|S| < 5000), Lead is significantly superior to the compared algorithms in 31.4% (Bsvm), 31.4% (Ml-knn), 68.6% (Bp-mll) and 54.3% (Ecc) cases, and is inferior to them in much less 0.0% (Bsvm), 17.1% (Mlknn), 8.6% (Bp-mll) and 17.1% (”
Inverse random under sampling for class imbalance problem and its application to multi-label classification (2012, background): “not be relevant to any known subset of labels or where label relationships may change over the test data [45].”
TGIF: A New Dataset and Benchmark on Animated GIF Description (2016, positive): “[29] as our visual classifier.”
On the Stratification of Multi-label Data (2011, methodology): “This was observed when the binary relevance approach was used, but could be generalizable to other algorithms, especially those learning a binary model for each label in one of their steps [18,4].”
Lift: Multi-Label Learning with Label-Specific Features (2011, background): “Few notable exceptions are [30] and [38] where a total of 11 and 15 data sets have been considered respectively.”
Automated Selection and Configuration of Multi-Label Classification Algorithms with Grammar-Based Genetic Programming (2018, methodology): “In order to evaluate its effectiveness, Auto-MEKAGGP was tested in 10 datasets and compared to two well-known MLC algorithms: Binary Relevance (BR) [16] and Classifier Chain (CC) [12].”
On label dependence and loss minimization in multi-label classification (2012, methodology): “To reduce the influence of the label order, Read et al. (2009) propose to average the multi-label predictions of CC over a (randomly chosen) set of permutations.”
SGM: Sequence Generation Model for Multi-label Classification (2018, methodology): “Classiﬁer chains (CC) proposed by Read et al. (2011) converts the MLC task into a chain of binary classiﬁcation problems to model the correlations between labels.”
Learning shared subspace for multi-label dimensionality reduction via dependence maximization (2015, methodology): “Algorithms belonging to this group include RAkEL [22] and ensembles of classifier chains [23].”
Online Metric Learning for Multi-Label Classification (2020, background): “Many studies (Dembczynski, Cheng, and Hüllermeier 2010; Read et al. 2011; Bhatia et al. 2015; Yen et al. 2016; Liu, Tsang, and Müller 2017) have shown that multi-label learning methods that do not capture label dependency usually achieve degraded prediction performance.”