Measuring Constraint-Set Utility for Partitional Clustering Algorithms

Davidson, Ian; Wagstaff, Kiri L.; Basu, Sugato

doi:10.1007/11871637_15

Cited by 134 publications

(124 citation statements)

References 4 publications

Supporting

Mentioning

118

Contrasting

Order By: Relevance

“…Although previous work report on average sub-stantial improvement in the clustering purity, (Davidson et al, 2006) shows that even if the constraints are generated from gold-standard data, some constraint sets can decrease clustering purity. The results vary significantly depending on the specific set of constraints used.…”

Section: Constrained Clusteringmentioning

confidence: 75%

Iterative Constrained Clustering for Subjectivity Word Sense Disambiguation

Akkaya

Wiebe

Mihalcea

2014

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics

View full text Add to dashboard Cite

Subjectivity word sense disambiguation (SWSD) is a supervised and applicationspecific word sense disambiguation task disambiguating between subjective and objective senses of a word. Not surprisingly, SWSD suffers from the knowledge acquisition bottleneck. In this work, we use a "cluster and label" strategy to generate labeled data for SWSD semiautomatically. We define a new algorithm called Iterative Constrained Clustering (ICC) to improve the clustering purity and, as a result, the quality of the generated data. Our experiments show that the SWSD classifiers trained on the ICC generated data by requiring only 59% of the labels can achieve the same performance as the classifiers trained on the full dataset.

show abstract

Section: Constrained Clusteringmentioning

confidence: 75%

Iterative Constrained Clustering for Subjectivity Word Sense Disambiguation

Akkaya

Wiebe

Mihalcea

2014

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics

View full text Add to dashboard Cite

show abstract

“…Cannot-link constraints (like charges) increase the weights of affected edge. The situation is illustrated in Figure 1, where must-link constraints are {e (2,3), e (6,7), e (14,15)} and the cannot-link constraints are {e(9,10)}.…”

Section: Magnetically Affected Paths (Map)mentioning

confidence: 99%

“…In some experiments, we detect phenomena where the accuracy of the algorithm goes up and down slightly as we increase the number of constraints. As shown in [6], this is a general problem of randomly-chosen constraint sets, where some constraints reduce the clustering performance. Thus, a learning metric or an edge weight re-adjustment method, is not always reliable for a small number of constraints.…”

Section: Comparison With Other Techniquesmentioning

confidence: 99%

“…The use of labeled data is often important for the success of the clustering process and for the evaluation of the clustering accuracy. Consequently, learning approaches which use both labeled and unlabeled data have attracted the interest of the research community [3,4,6,12,20]. Such approaches incorporate user knowledge in the clustering technique, thus improving the clustering result.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Applying Electromagnetic Field Theory Concepts to Clustering with Constraints

Hakkoymaz

Chatzimilioudis

Gunopulos

et al. 2009

Machine Learning and Knowledge Discovery in Databases

View full text Add to dashboard Cite

Abstract. This work shows how concepts from the electromagnetic field theory can be efficiently used in clustering with constraints. The proposed framework transforms vector data into a fully connected graph, or just works straight on the given graph data. User constraints are represented by electromagnetic fields that affect the weight of the graph's edges. A clustering algorithm is then applied on the adjusted graph, using k-distinct shortest paths as the distance measure. Our framework provides better accuracy compared to MPCK-Means, SS-KernelKMeans and Kmeans+Diagonal Metric even when very few constraints are used, significantly improves clustering performance on some datasets that other methods fail to partition successfully, and can cluster both vector and graph datasets. All these advantages are demonstrated through thorough experimental evaluation.

show abstract

“…Recently, Wagstaff et al, and Davidson et al attempted to link the quality of constraint sets with clustering algorithm performance (Davidson, Wagstaff et al 2006;Wagsta, Basu et al 2006). Two properties of constraint setinconsistency and incoherence -were shown to be strongly negative correlated with clustering algorithm performance.…”

Section: Introductionmentioning

confidence: 99%

Clustering Genes Using Heterogeneous Data Sources

Zeng

Yang

et al. 2010

International Journal of Knowledge Discovery in Bioinformatics

View full text Add to dashboard Cite

Clustering of gene expression data is a standard exploratory technique used to identify closely related genes. Many other sources of data are also likely to be of great assistance in the analysis of gene expression data. Such sources include proteinprotein interaction data, transcription factor and regulatory elements data, comparative genomics data, protein expression data and much more. These data provide us with a means to begin elucidating the large-scale modular organization of the cell. Conclusions drawn from more than one data source is likely to lead to new insights. Data sources may be complete or incomplete depending on whether or not they provide information about every gene in the genome. With a view toward a combined analysis of heterogeneous sources of data, we consider the challenging task of developing exploratory analytical techniques to deal with multiple complete and incomplete information sources. The Multi-Source Clustering (MSC) algorithm we developed performs clustering with multiple, but complete, sources of data. To deal with incomplete data sources, we have adopted the MPCK-means clustering algorithm, which is a constrained clustering algorithm, to perform exploratory analysis on one complete source (such as gene expression data) and other potentially incomplete sources provided in the form of constraints. We have shown that the MSC algorithm produces clusters that are biologically more meaningful when integrating gene expression data and text data than those identified using only one source of data. For the constrained clustering algorithm, we have studied the effectiveness of various constraints sets. To address the problem of automatically generating constraints from biological text literature, we considered two methods (cluster-based and similarity-based). The novelty of research presented here is the development of a new clustering algorithm MSC to perform exploratory analysis using two or more diverse but complete data sources, and study of effectiveness of constraints sets and robustness of the constrained clustering algorithm using multiple sources of incomplete biological data, and incorporating such incomplete data into constrained clustering algorithm in form of constraints sets.

show abstract

Measuring Constraint-Set Utility for Partitional Clustering Algorithms

Cited by 134 publications

References 4 publications

Iterative Constrained Clustering for Subjectivity Word Sense Disambiguation

Iterative Constrained Clustering for Subjectivity Word Sense Disambiguation

Applying Electromagnetic Field Theory Concepts to Clustering with Constraints

Clustering Genes Using Heterogeneous Data Sources

Contact Info

Product

Resources

About