An expressive dissimilarity measure for relational clustering using neighbourhood trees

Clustering is an underspecified task: there are no universal criteria for what makes a good clustering. This is especially true for relational data, where similarity can be based on the features of individuals, the relationships between them, or a mix of both. Existing methods for relational clustering have strong and often implicit biases in this respect. In this paper, we introduce a novel dissimilarity measure for relational data. It is the first approach to incorporate a wide variety of types of similarity, including similarity of attributes, similarity of relational context, and proximity in a hypergraph. We experimentally evaluate the proposed dissimilarity measure on both clustering and classification tasks using data sets of very different types. Considering the quality of the obtained clustering, the experiments demonstrate that (a) using this dissimilarity in standard clustering methods consistently gives good results, whereas other measures work well only on data sets that match their bias; and (b) on most data sets, the novel dissimilarity outperforms even the best among the existing ones. On the classification tasks, the proposed method outperforms the competitors on the majority of data sets, often by a large margin. Moreover, we show that learning the appropriate bias in an unsupervised way is a very challenging task, and that the existing methods offer a marginal gain compared to the proposed similarity method, and can even hurt performance. Finally, we show that the asymptotic complexity of the proposed dissimilarity measure is similar to the existing state-of-the-art approaches. The results confirm that the proposed dissimilarity measure is indeed versatile enough to capture relevant information, regardless of whether that comes from the attributes of vertices, their proximity, or connectedness of vertices, even without parameter tuning.     

一种基于MST的自适应优化相异性度量的半监督聚类方法

针对混合属性空间中具有同一(或相近)分布特性的带类别标记的小样本集和无类别标记的大样本数据集,提出了一种基于MST的自适应优化相异性度量的半监督聚类方法。该方法首先采用决策树方法来获取小样本集的"规则聚类区域",然后根据"同一聚类的数据点更为接近"的原则自适应优化建构在该混合属性空间中的相异性度量,最后将优化后的相异性度量应用于基于MST的聚类算法中,以获得更为有效的聚类结果。仿真实验结果表明,该方法对有些数据集是有改进效果的。为进一步推广并在实际中发掘出该方法的应用价值,本文在最后给出了一个较有价值的研究展望。     

Determining the number of clusters using information entropy for mixed data

In cluster analysis, one of the most challenging and difficult problems is the determination of the number of clusters in a data set, which is a basic input parameter for most clustering algorithms. To solve this problem, many algorithms have been proposed for either numerical or categorical data sets. However, these algorithms are not very effective for a mixed data set containing both numerical attributes and categorical attributes. To overcome this deficiency, a generalized mechanism is presented in this paper by integrating Rényi entropy and complement entropy together. The mechanism is able to uniformly characterize within-cluster entropy and between-cluster entropy and to identify the worst cluster in a mixed data set. In order to evaluate the clustering results for mixed data, an effective cluster validity index is also defined in this paper. Furthermore, by introducing a new dissimilarity measure into the k-prototypes algorithm, we develop an algorithm to determine the number of clusters in a mixed data set. The performance of the algorithm has been studied on several synthetic and real world data sets. The comparisons with other clustering algorithms show that the proposed algorithm is more effective in detecting the optimal number of clusters and generates better clustering results.     

On clustering tree structured data with categorical nature

Clustering consists in partitioning a set of objects into disjoint and homogeneous clusters. For many years, clustering methods have been applied in a wide variety of disciplines and they also have been utilized in many scientific areas. Traditionally, clustering methods deal with numerical data, i.e. objects represented by a conjunction of numerical attribute values. However, nowadays commercial or scientific databases usually contain categorical data, i.e. objects represented by categorical attributes. In this paper we present a dissimilarity measure which is capable to deal with tree structured categorical data. Thus, it can be used for extending the various versions of the very popular k-means clustering algorithm to deal with such data. We discuss how such an extension can be achieved. Moreover, we empirically prove that the proposed dissimilarity measure is accurate, compared to other well-known (dis)similarity measures for categorical data.     

Combining partitional and hierarchical algorithms for robust and efficient data clustering with cohesion self-merging

Data clustering has attracted a lot of research attention in the field of computational statistics and data mining. In most related studies, the dissimilarity between two clusters is defined as the distance between their centroids or the distance between two closest (or farthest) data points However, all of these measures are vulnerable to outliers and removing the outliers precisely is yet another difficult task. In view of this, we propose a new similarity measure, referred to as cohesion, to measure the intercluster distances. By using this new measure of cohesion, we have designed a two-phase clustering algorithm, called cohesion-based self-merging (abbreviated as CSM), which runs in time linear to the size of input data set. Combining the features of partitional and hierarchical clustering methods, algorithm CSM partitions the input data set into several small subclusters in the first phase and then continuously merges the subclusters based on cohesion in a hierarchical manner in the second phase. The time and the space complexities of algorithm CSM are analyzed. As shown by our performance studies, the cohesion-based clustering is very robust and possesses excellent tolerance to outliers in various workloads. More importantly, algorithm CSM is shown to be able to cluster the data sets of arbitrary shapes very efficiently and provide better clustering results than those by prior methods.     

DHCC: Divisive hierarchical clustering of categorical data

Clustering categorical data poses two challenges defining an inherently meaningful similarity measure, and effectively dealing with clusters which are often embedded in different subspaces. In this paper, we propose a novel divisive hierarchical clustering algorithm for categorical data, named DHCC. We view the task of clustering categorical data from an optimization perspective, and propose effective procedures to initialize and refine the splitting of clusters. The initialization of the splitting is based on multiple correspondence analysis (MCA). We also devise a strategy for deciding when to terminate the splitting process. The proposed algorithm has five merits. First, due to its hierarchical nature, our algorithm yields a dendrogram representing nested groupings of patterns and similarity levels at different granularities. Second, it is parameter-free, fully automatic and, in particular, requires no assumption regarding the number of clusters. Third, it is independent of the order in which the data is processed. Fourth, it is scalable to large data sets. And finally, our algorithm is capable of seamlessly discovering clusters embedded in subspaces, thanks to its use of a novel data representation and Chi-square dissimilarity measures. Experiments on both synthetic and real data demonstrate the superior performance of our algorithm.     

A semi-supervised regression model for mixed numerical and categorical variables

In this paper, we develop a semi-supervised regression algorithm to analyze data sets which contain both categorical and numerical attributes. This algorithm partitions the data sets into several clusters and at the same time fits a multivariate regression model to each cluster. This framework allows one to incorporate both multivariate regression models for numerical variables (supervised learning methods) and k-mode clustering algorithms for categorical variables (unsupervised learning methods). The estimates of regression models and k-mode parameters can be obtained simultaneously by minimizing a function which is the weighted sum of the least-square errors in the multivariate regression models and the dissimilarity measures among the categorical variables. Both synthetic and real data sets are presented to demonstrate the effectiveness of the proposed method.     

Spectral clustering with eigenvector selection

The task of discovering natural groupings of input patterns, or clustering, is an important aspect of machine learning and pattern analysis. In this paper, we study the widely used spectral clustering algorithm which clusters data using eigenvectors of a similarity/affinity matrix derived from a data set. In particular, we aim to solve two critical issues in spectral clustering: (1) how to automatically determine the number of clusters, and (2) how to perform effective clustering given noisy and sparse data. An analysis of the characteristics of eigenspace is carried out which shows that (a) not every eigenvectors of a data affinity matrix is informative and relevant for clustering; (b) eigenvector selection is critical because using uninformative/irrelevant eigenvectors could lead to poor clustering results; and (c) the corresponding eigenvalues cannot be used for relevant eigenvector selection given a realistic data set. Motivated by the analysis, a novel spectral clustering algorithm is proposed which differs from previous approaches in that only informative/relevant eigenvectors are employed for determining the number of clusters and performing clustering. The key element of the proposed algorithm is a simple but effective relevance learning method which measures the relevance of an eigenvector according to how well it can separate the data set into different clusters. Our algorithm was evaluated using synthetic data sets as well as real-world data sets generated from two challenging visual learning problems. The results demonstrated that our algorithm is able to estimate the cluster number correctly and reveal natural grouping of the input data/patterns even given sparse and noisy data.     

Overlapping correlation clustering

We introduce a new approach for finding overlapping clusters given pairwise similarities of objects. In particular, we relax the problem of correlation clustering by allowing an object to be assigned to more than one cluster. At the core of our approach is an optimization problem in which each data point is mapped to a small set of labels, representing membership in different clusters. The objective is to find a mapping so that the given similarities between objects agree as much as possible with similarities taken over their label sets. The number of labels can vary across objects. To define a similarity between label sets, we consider two measures: (i) a 0–1 function indicating whether the two label sets have non-zero intersection and (ii) the Jaccard coefficient between the two label sets. The algorithm we propose is an iterative local-search method. The definitions of label set similarity give rise to two non-trivial optimization problems, which, for the measures of set-intersection and Jaccard, we solve using a greedy strategy and non-negative least squares, respectively. We also develop a distributed version of our algorithm based on the BSP model and implement it using a Pregel framework. Our algorithm uses as input pairwise similarities of objects and can thus be applied when clustering structured objects for which feature vectors are not available. As a proof of concept, we apply our algorithms on three different and complex application domains: trajectories, amino-acid sequences, and textual documents.     

Is VAT really single linkage in disguise?

This paper addresses the relationship between the Visual Assessment of cluster Tendency (VAT) algorithm and single linkage hierarchical clustering. We present an analytical comparison of the two algorithms in conjunction with numerical examples to show that VAT reordering of dissimilarity data is directly related to the clusters produced by single linkage hierarchical clustering. This analysis is important to understanding the underlying theory of VAT and, more generally, other algorithms that are based on VAT-ordered dissimilarity data.     

Locally linear metric adaptation with application to semi-supervised clustering and image retrieval

Many computer vision and pattern recognition algorithms are very sensitive to the choice of an appropriate distance metric. Some recent research sought to address a variant of the conventional clustering problem called semi-supervised clustering, which performs clustering in the presence of some background knowledge or supervisory information expressed as pairwise similarity or dissimilarity constraints. However, existing metric learning methods for semi-supervised clustering mostly perform global metric learning through a linear transformation. In this paper, we propose a new metric learning method that performs nonlinear transformation globally but linear transformation locally. In particular, we formulate the learning problem as an optimization problem and present three methods for solving it. Through some toy data sets, we show empirically that our locally linear metric adaptation (LLMA) method can handle some difficult cases that cannot be handled satisfactorily by previous methods. We also demonstrate the effectiveness of our method on some UCI data sets. Besides applying LLMA to semi-supervised clustering, we have also used it to improve the performance of content-based image retrieval systems through metric learning. Experimental results based on two real-world image databases show that LLMA significantly outperforms other methods in boosting the image retrieval performance.     

Cluster merging based on a decision threshold

Data clustering is an important unsupervised learning technique and has wide application in various fields including pattern recognition, data mining, image analysis and bioinformatics. A vast amount of clustering algorithms have been proposed in the past decades. However, existing algorithms still face many problems in practical applications. One typical problem is the parameter dependence, which means that user-specified parameters are required as input and the clustering results are influenced by these parameters. Another problem is that many algorithms are not able to generate clusters of non-spherical shapes. In this paper, a cluster merging method is proposed to solve the above-mentioned problems based on a decision threshold and the dominant sets algorithm. Firstly, the influence of similarity parameter on dominant sets clustering results is studied, and it is found that the obtained clusters become larger with the increase in similarity parameter. We analyze the reason behind this behavior and propose to generate small initial clusters in the first step and then merge the initial clusters to improve the clustering results. Specifically, we select a similarity parameter which generates small but not too small clusters. Then, we calculate pairwise merging decisions among the initial clusters and obtain a merging decision threshold. Based on this threshold, we merge the small clusters and obtain the final clustering results. Experiments on several datasets are used to validate the effectiveness of the proposed algorithm.

     

A novel incremental conceptual hierarchical text clustering method using CFu-tree

As a data mining method, clustering, which is one of the most important tools in information retrieval, organizes data based on unsupervised learning which means that it does not require any training data. But, some text clustering algorithms cannot update existing clusters incrementally and, instead, have to recompute a new clustering from scratch. In view of above, this paper presents a novel down-top incremental conceptual hierarchical text clustering approach using CFu-tree (ICHTC-CF) representation, which starts with each item as a separate cluster. Term-based feature extraction is used for summarizing a cluster in the process. The Comparison Variation measure criterion is also adopted for judging whether the closest pair of clusters can be merged or a previous cluster can be split. And, our incremental clustering method is not sensitive to the input data order. Experimental results show that the performance of our method outperforms k-means, CLIQUE, single linkage clustering and complete linkage clustering, which indicate our new technique is efficient and feasible.     

A dissimilarity function for geospatial polygons

Similarity plays an important role in many data mining tasks and information retrieval processes. Most of the supervised, semi-supervised, and unsupervised learning algorithms depend on using a dissimilarity function that measures the pair-wise similarity between the objects within the dataset. However, traditionally most of the similarity functions fail to adequately treat all the spatial attributes of the geospatial polygons due to the incomplete quantitative representation of structural and topological information contained within the polygonal datasets. In this paper, we propose a new dissimilarity function known as the polygonal dissimilarity function (PDF) that comprehensively integrates both the spatial and the non-spatial attributes of a polygon to specifically consider the density, distribution, and topological relationships that exist within the polygonal datasets. We represent a polygon as a set of intrinsic spatial attributes, extrinsic spatial attributes, and non-spatial attributes. Using this representation of the polygons, PDF is defined as a weighted function of the distance between two polygons in the different attribute spaces. In order to evaluate our dissimilarity function, we compare and contrast it with other distance functions proposed in the literature that work with both spatial and non-spatial attributes. In addition, we specifically investigate the effectiveness of our dissimilarity function in a clustering application using a partitional clustering technique (e.g. \(k\) -medoids) using two characteristically different sets of data: (a) Irregular geometric shapes determined by natural processes, i.e., watersheds and (b) semi-regular geometric shapes determined by human experts, i.e., counties.     

A fuzzy relational clustering algorithm based on a dissimilarity measure extracted from data.

One of the critical aspects of clustering algorithms is the correct identification of the dissimilarity measure used to drive the partitioning of the data set. The dissimilarity measure induces the cluster shape and therefore determines the success of clustering algorithms. As cluster shapes change from a data set to another, dissimilarity measures should be extracted from data. To this aim, we exploit some pairs of points with known dissimilarity value to teach a dissimilarity relation to a feed-forward neural network. Then, we use the neural dissimilarity measure to guide an unsupervised relational clustering algorithm. Experiments on synthetic data sets and on the Iris data set show that the relational clustering algorithm based on the neural dissimilarity outperforms some popular clustering algorithms (with possible partial supervision) based on spatial dissimilarity.     

A redundancy-based measure of dissimilarity among probability distributions for hierarchical clustering criteria

We introduce novel dissimilarity into a probabilistic clustering task to properly measure dissimilarity among multiple clusters when each cluster is characterized by a subpopulation in the mixture model. This measure of dissimilarity is called redundancy-based dissimilarity among probability distributions. From aspects of both source coding and a statistical hypothesis test, we shed light on several of the theoretical reasons for the redundancy-based dissimilarity among probability distributions being a reasonable measure of dissimilarity among clusters. We also elucidate a principle in common for the measures of redundancy-based dissimilarity and Ward's method in terms of hierarchical clustering criteria. Moreover, we show several related theorems that are significant for clustering tasks. In the experiments, properties of the measure of redundancy-based dissimilarity are examined in comparison with several other measures.     

Min-max hyperellipsoidal clustering for anomaly detection in network security.

A novel hyperellipsoidal clustering technique is presented for an intrusion-detection system in network security. Hyperellipsoidal clusters toward maximum intracluster similarity and minimum intercluster similarity are generated from training data sets. The novelty of the technique lies in the fact that the parameters needed to construct higher order data models in general multivariate Gaussian functions are incrementally derived from the data sets using accretive processes. The technique is implemented in a feedforward neural network that uses a Gaussian radial basis function as the model generator. An evaluation based on the inclusiveness and exclusiveness of samples with respect to specific criteria is applied to accretively learn the output clusters of the neural network. One significant advantage of this is its ability to detect individual anomaly types that are hard to detect with other anomaly-detection schemes. Applying this technique, several feature subsets of the tcptrace network-connection records that give above 95% detection at false-positive rates below 5% were identified.     

An Algorithm for Clustering Using L1‐Norm Based on Hyperbolic Smoothing Technique

Cluster analysis deals with the problem of organization of a collection of objects into clusters based on a similarity measure, which can be defined using various distance functions. The use of different similarity measures allows one to find different cluster structures in a data set. In this article, an algorithm is developed to solve clustering problems where the similarity measure is defined using the L₁‐norm. The algorithm is designed using the nonsmooth optimization approach to the clustering problem. Smoothing techniques are applied to smooth both the clustering function and the L₁‐norm. The algorithm computes clusters sequentially and finds global or near global solutions to the clustering problem. Results of numerical experiments using 12 real‐world data sets are reported, and the proposed algorithm is compared with two other clustering algorithms.     

A multivariate fuzzy c-means method

Fuzzy c-means (FCMs) is an important and popular unsupervised partitioning algorithm used in several application domains such as pattern recognition, machine learning and data mining. Although the FCM has shown good performance in detecting clusters, the membership values for each individual computed to each of the clusters cannot indicate how well the individuals are classified. In this paper, a new approach to handle the memberships based on the inherent information in each feature is presented. The algorithm produces a membership matrix for each individual, the membership values are between zero and one and measure the similarity of this individual to the center of each cluster according to each feature. These values can change at each iteration of the algorithm and they are different from one feature to another and from one cluster to another in order to increase the performance of the fuzzy c-means clustering algorithm. To obtain a fuzzy partition by class of the input data set, a way to compute the class membership values is also proposed in this work. Experiments with synthetic and real data sets show that the proposed approach produces good quality of clustering.