## Combining Multiple Clustering Systems (2004)

Venue: | In 8th European conference on Principles and Practice of Knowledge Discovery in Databases(PKDD), LNAI 3202 |

Citations: | 19 - 1 self |

### BibTeX

@INPROCEEDINGS{Boulis04combiningmultiple,

author = {Constantinos Boulis and Mari Ostendorf},

title = {Combining Multiple Clustering Systems},

booktitle = {In 8th European conference on Principles and Practice of Knowledge Discovery in Databases(PKDD), LNAI 3202},

year = {2004},

pages = {63--74}

}

### OpenURL

### Abstract

Three methods for combining multiple clustering systems are presented and evaluated, focusing on the problem of finding the correspondence between clusters of di#erent systems. In this work, the clusters of individual systems are represented in a common space and their correspondence estimated by either "clustering clusters" or with Singular Value Decomposition. The approaches are evaluated for the task of topic discovery on three major corpora and eight di#erent clustering algorithms and it is shown experimentally that combination schemes almost always o#er gains compared to single systems, but gains from using a combination scheme depend on the underlying clustering systems.

### Citations

8132 | Maximum likelihood from incomplete data via the em algorithm
- Dempster, Laird, et al.
- 1977
(Show Context)
Citation Context ... d) is the count of word w in document d. The cluster c that each document is generated from is assumed to be hidden. Training such a model is carried out using the Expectation-Maximization algorithm =-=[20]-=-. In practice, smoothing the multinomial distributions is necessary. The mixture of multinomials algorithm is the unsupervised analogue of the Naive Bayes algorithm and has been successfully used in t... |

1304 | Data clustering: a review
- Jain, Murty, et al.
- 1999
(Show Context)
Citation Context ...rse fields, such as genomics [1], lexical semantics [2], information retrieval [3] and automatic speech recognition [4], to name a few. A number of different clustering approaches have been suggested =-=[5]-=- such as agglomerative clustering, mixture densities and graph partitioning. Most clustering methods focus on individual criteria or models and do not address issues of combining multiple different sy... |

1227 | Molecular classification of cancer: class discovery and class prediction by gene expression monitoring
- Golub, Slonim, et al.
- 1999
(Show Context)
Citation Context ... single systems, but gains from using a combination scheme depend on the underlying clustering systems. 1 Introduction Clustering has an important role in a number of diverse fields, such as genomics =-=[1]-=-, lexical semantics [2], information retrieval [3] and automatic speech recognition [4], to name a few. A number of different clustering approaches have been suggested [5] such as agglomerative cluste... |

540 | An empirical comparison of voting classification algorithms: Bagging, boosting, and variants
- Bauer, Kohavi
- 1999
(Show Context)
Citation Context ...ombining multiple different systems. The problem of combining multiple clustering systems is analogous to the classifier combination problem, that has received increased attention over the last years =-=[6]-=-. Unlike the classifier combination problem, though, the correspondence between clusters of different systems is unknown. For example, consider two clustering systems applied to nine data points and c... |

527 |
Switchboard: Telephone speech corpus for research and development
- Godfrey, Holliman
- 1990
(Show Context)
Citation Context ...o one of 20 categories (newsgroups). The second corpus is a subset of Reuters-21578 3 , consisting of 1000 documents equally distributed among 20 topics. The third corpus is Switchboard-I release 2.0 =-=[18]-=-, a collection of 2263 5-minute telephone conversations on 67 possible topics. Switchboard-I and to a smaller extent 20Newsgroups, are characterized with a spontaneous, less structured style. On the o... |

479 |
Bayesian classification (AutoClass: theory and results
- Cheeseman, Stutz
- 1996
(Show Context)
Citation Context ...ogue of the Naive Bayes algorithm and has been successfully used in the past for document clustering [21]. Mixture models, in general, have been extensively used for data mining and pattern discovery =-=[22]-=-. The software package CLUTO4 was used for the optimization-based algorithms. Using CLUTO, a number of different clustering methods (hierarchical, partitional and graph-partitioning) and criteria can ... |

398 | ensembles – a knowledge reuse framework for combining multiple partitions
- Strehl, Ghosh
- 2002
(Show Context)
Citation Context ...ce, but also on the combination of multiple runs. 2 Related Work Combining multiple clustering systems has recently attracted the interest of several researchers in the machine learning community. In =-=[7]-=-, three different approaches for combining clusters based on graph-partitioning are proposed and evaluated. The first approach avoids the correspondence problem by defining a pairwise similarity matri... |

378 | Automatic word sense discrimination
- Schütze
- 1998
(Show Context)
Citation Context ...ins from using a combination scheme depend on the underlying clustering systems. 1 Introduction Clustering has an important role in a number of diverse fields, such as genomics [1], lexical semantics =-=[2]-=-, information retrieval [3] and automatic speech recognition [4], to name a few. A number of different clustering approaches have been suggested [5] such as agglomerative clustering, mixture densities... |

238 | Grouper: A Dynamic Clustering Interface to Web Search Results
- Zamir, Etzioni
(Show Context)
Citation Context ...n scheme depend on the underlying clustering systems. 1 Introduction Clustering has an important role in a number of diverse fields, such as genomics [1], lexical semantics [2], information retrieval =-=[3]-=- and automatic speech recognition [4], to name a few. A number of different clustering approaches have been suggested [5] such as agglomerative clustering, mixture densities and graph partitioning. Mo... |

235 | Refining initial points for K-Means clustering
- Bradley, Fayyad
- 1998
(Show Context)
Citation Context ...ned by multiple runs of the k-means algorithm with different initial conditions. An agglomerative pairwise cluster merging scheme is used, with a heuristic to determine the corresponding clusters. In =-=[16]-=-, a twostage clustering procedure is proposed. Resampling is used to obtain multiple solutions of k-means. The output centroids from multiple runs are clustered with a new k-means run. A disadvantage ... |

162 | Learning to classify text from labeled and unlabeled documents
- Nigam, McCallum, et al.
- 2000
(Show Context)
Citation Context ...tinomial distributions is necessary. The mixture of multinomials algorithm is the unsupervised analogue of the Naive Bayes algorithm and has been successfully used in the past for document clustering =-=[21]-=-. Mixture models, in general, have been extensively used for data mining and pattern discovery [22]. The software package CLUTO4 was used for the optimization-based algorithms. Using CLUTO, a number o... |

158 | Consensus clustering: A resampling-based method for class discovery and visualization of gene expression microarray data
- Monti, Tamayo, et al.
- 2003
(Show Context)
Citation Context ...ns assigned to cluster i, mj the number of observation of class j and D the total number of observations. It can be shown that 0 < NMI ≤ 1 with NMI=1 corresponding to perfect classification accuracy=-=. (8)-=-s5 Experiments The multiple clustering system combination schemes that are introduced in this paper are general and can, in principle, be applied to any clustering problem. The task we have chosen to ... |

101 | Bagging to improve the accuracy of a clustering procedure
- Dudoit, Fridlyand
(Show Context)
Citation Context ...be assigned to the combined hyperedge they most strongly belong to. Clustering hyperedges is performed by using graph-partitioning algorithms. The same core idea can also be found in [10, 14– 16]. I=-=n [10]-=-, different clustering solutions are obtained by resampling and are aligned with the clusters estimated on all the data. In both [14, 15], the different clustering solutions are obtained by multiple r... |

97 | Random projection for high dimensional data clustering: A cluster ensemble approach - Fern, Brodley - 2003 |

96 | Data clustering using evidence accumulation - Fred, Jain - 2002 |

82 | Empirical and theoretical comparisons of selected criterion functions for document clustering
- Zhao, Karypis
(Show Context)
Citation Context ...2 Clustering Algorithms A number of different clustering systems were used, including the mixture of multinomials (MixMulti) and the optimization-based clustering algorithms and criteria described in =-=[19]-=-. The MixMulti algorithm clusters documents by estimating a mixture of multinomial distributions. The assumption is that each topic is characterized by a different multinomial distribution, i.e. diffe... |

60 | A mixture model for clustering ensembles
- Topchy, Jain, et al.
- 2004
(Show Context)
Citation Context ...imum number of hyperedges. This approach is linear with the number of data points, but requires fairly balanced data sets and all hyperedges having the same weight. A similar approach is presented in =-=[13]-=-, where each data point is represented with a set of meta-features. Each meta-feature is the cluster membership for each system, and the data points are clustered using a mixture model. An advantage o... |

31 |
Large Vocabulary Speech Recognition with Multispan Statistical Language Models
- Bellegarda
- 2000
(Show Context)
Citation Context ...stering systems. 1 Introduction Clustering has an important role in a number of diverse fields, such as genomics [1], lexical semantics [2], information retrieval [3] and automatic speech recognition =-=[4]-=-, to name a few. A number of different clustering approaches have been suggested [5] such as agglomerative clustering, mixture densities and graph partitioning. Most clustering methods focus on indivi... |

23 | An adaptive meta-clustering approach: combining the information from different clustering results - Zeng, Tang, et al. - 2002 |

14 |
A combination scheme for fuzzy clustering
- Dimitriadou, Weingessel, et al.
- 2002
(Show Context)
Citation Context ...gorithms. The same core idea can also be found in [10, 14– 16]. In [10], different clustering solutions are obtained by resampling and are aligned with the clusters estimated on all the data. In bot=-=h [14, 15]-=-, the different clustering solutions are obtained by multiple runs of the k-means algorithm with different initial conditions. An agglomerative pairwise cluster merging scheme is used, with a heuristi... |

7 |
A multi-clustering fusion algorithm
- Frossyniotis, Pertselakis, et al.
- 2002
(Show Context)
Citation Context ...gorithms. The same core idea can also be found in [10, 14– 16]. In [10], different clustering solutions are obtained by resampling and are aligned with the clusters estimated on all the data. In bot=-=h [14, 15]-=-, the different clustering solutions are obtained by multiple runs of the k-means algorithm with different initial conditions. An agglomerative pairwise cluster merging scheme is used, with a heuristi... |

3 |
Estimating the dimension of a model. The Annals of Statistics 6(2
- Schwartz
- 1978
(Show Context)
Citation Context ...o simplify our experiments, the number of topics is assumed to be known. This is an assumption that is not true in many practical cases, but standard techniques such as Bayesian Information Criterion =-=[17]-=- can be used to select the number of topics. It should be noted that the unconstrained and SVD combination schemes do not require the same number of clusters for all systems. On the other hand, the co... |