Widely discussed in the mass media, Web 2.0, or social software, has also drawn the attention of researchers, developing into a whole new research area. With Web 2.0's further development, corporations aim to adopt its technologies and transfer its benefits, such as enhanced collaboration and knowledge sharing, to their organizations. Whether any of these benefits also apply in an organizational context and whether there are further, still uncovered, benefits remains unclear. Furthermore, research in this area is still in its early stages, thus hampering progress towards qualitative and quantitative models that could provide answers. In order to encourage further progress in this area, we reviewed the existing research on corporate blogging and identified 24 articles that investigate the topic. Using the framework by Ives et al. [18], we categorized the articles for further analysis. By means of process theory, we build a conceptual model and identify the antecedents and consequences of internal corporate weblog usage. Our findings suggest that usage is driven by organizational culture, as well as by attitudes towards blogging. In addition, the benefits of weblog usage are centered on community benefits.

While the rest of the world has generally accepted ISO 14001 as the environmental standard, US firms, in general, have lagged behind in their acceptance of this standard. Given the relative importance of US manufacturing and given that one of the objectives of the ISO 14001 standard was that it was to be a global standard, this apparent 'hesitancy' on the part of US firms is troublesome. This paper attempts to explain this hesitancy by casting the certification decision as a business decision. As such, certification will occur when the quantifiable benefits exceed the measured costs. The paper examines how US managers perceive the relative costs and benefits associated with this form of certification. The analysis is carried out at two levels: at an aggregated level using data generated from a large-scale mail survey and at the micro level using data provided by a detailed case study. The results indicate that environmental activities have a more positive impact on plant operations among those facilities that are certified. Moreover, these benefits tend to be primarily attitudinal in nature and strongly influenced by the industrial setting in which the plant is located.

Whereas the production of papers and books in classification grows rapidly, and the methods of clustering are numerous, sometimes making it difficult for practitioners to choose among available methods, one observes that there is only a small number of papers dealing with the validity of these methods and of the results (trees, partitions, clusters) provided by each. This is related to the lack of a statistical theory of classification, and of a rigourous mathematical definition of the concept of cluster.

When performing a classification study, it is sometimes a sound strategy not to classify all subjects but to leave a residue of unclassified entities to be analyzed separately. Starting from an interactional paradigm, theoretical reasons for this approach were given. A procedure, RESIDAN, for carrying out a classification analysis using a residue was presented and empirical results for two data sets were given, both of which indicated that belonging to a residue is a property of individuals that has a significant stability over time. Although RESIDAN can be applied to data from a single measurement period, test-retest data are useful for separating subjects with true deviant patterns from subjects with deviant patterns caused by errors of measurement. It was argued that the concept of antitype (= rare pattern) has theoretical significance and could be studied within the presented framework.

We investigated the nosology of endogenous depression by numerical taxonomy. Five hundred and sixty-nine patients diagnosed as having unipolar major depressive disorder in the NIMH Clinical Research Branch Program on the Psychobiology of Depression-Clinical were studied. Thirty-six symptoms which might distinguish endogenous from non-endogenous depressions were chosen from the literature. Patients' symptom profiles assessed by structured interview were grouped by two methods: a K-means improvement of Ward's method of cluster analysis, and a latent class algorithm. The methods produced very similar groups and several internal validity criteria suggested that the groups were not spurious. Cluster 1, 'nuclear depression,' included a nucleus of patients common to multiple definitions of endogenous depression. The non-nuclear group scored as less neurotic than the nuclear group on personality tests administered during the index episode. The groups do not differ in frequency, number or severity of reported life events prior to onset of the index episode. The nuclear group shows a poor prognosis on two-year prospective follow-up, greater disturbance on personality inventories, and increased heritability of depression in siblings.

We introduce a UPGMA based counterpart to the Lance and Williams (1966) p-flexible clustering strategy. This new method is compared with the original p-flexible technique based on WPGMA for a range of fl values from -0.8 to +0.1. Recovery of the known cluster structure from simulated data was evaluated using the Hubert and Arabie (1985) version of the Rand statistic. The algorithm used to generate the simulated clusters in the study included error-free data as well as three other error conditions. The simulated data also varied the number of clusters, the underlying dimensionality, and the density distribution of points to the clusters. Results showed that the flexible UPGMA gives the best recovery for all generated data configurations. Further, the flexible UPGMA method with a small negative β value performed better than the standard UPGMA method where β is set equal to zero. The flexible UPGMA strategy, employed with a narrow β range, is recommended based upon its ability to recover cluster structure over all error conditions.

Visualization can be very powerful in revealing cluster structures. However, directly using visualization techniques to verify the validity of clustering results is still a challenge. This is due to the fact that visual representation lacks precision in contrasting clustering results. To remedy this problem, in this paper we propose a novel approach, which employs a visualization technique called HOV (hypothesis oriented verification and validation by visualization) which offers a tunable measure mechanism to project clustered subsets and non-clustered subsets from a multidimensional space to a 2D plane. By comparing the data distributions of the subsets, users not only have an intuitive visual evaluation but also have a precise evaluation on the consistency of cluster structure by calculating geometrical information of their data distributions

Using Root Cause Analysis to Handle Intrusion Detection Alarms Klaus Julisch IBM Zurich Research Laboratory Säumerstrasse 4 8803 R̈uschlikon, Switzerland e-mail: kju@zurich.ibm.com In response to attacks against enterprise networks, administrators are increasingly deploying intrusion detection systems. These systems monitor hosts, networks, and other resources for signs of security violations. Unfortunately, the use of intrusion detection has given rise to another difficult problem, namely the handling of a generally large number of mostly false alarms. This dissertation presents a novel paradigm for handling intrusion detection alarms more efficiently. Central to this paradigm is the notion that each alarm occurs for a reason, which is referred to as the alarm’s root causes . This dissertation observes that a few dozens of root causes generally account for over 90% of the alarms in an alarm log. Moreover, these root causes are generally persistent, i.e. they keep triggering alarms until someone removes them. Based on these observations, we propose a new two-step paradigm for alarm handling: Step one identifies root causes that account for large numbers of alarms, and step two removes these root causes and thereby reduces the future alarm load. Alternatively, alarms originating from benign root causes can be filtered out. To support the discovery of root causes, we propose a novel data mining technique, called larm clustering. To lay the foundation for alarm clustering, we show that many root causes manifest themselves in alarm groups that have certain structural properties. We formalize these structural properties and propose alarm clustering as a method for extracting alarm groups that have these properties. Such alarm groups are generally indicative of root causes. We therefore present them to a human expert who is responsible for identifying the underlying root causes. Once identified, the root causes can be removed (or false positives can be filtered out) so as to reduce the

Two hundred persons arrested for driving under the influence (DUI), 30 social drinkers, 30 depressed patients, 30 incarcerated criminals, and 30 alcoholics completed the Hogan Personality Inventory (HPI) and Court Reporting Network (CRN) interview. A cluster analysis of HPI scores for the DUI group revealed 5 personality types: Impulsive-Extravert, Normal, Neurotic-Introvert, NeuroticHostile, and Unassertive-Conformist. The types differed predictably on demographic variables, drinking behavior, and driving records as assessed by the CRN. The Impulsive-Extravert and Normal types had HPI profiles similar to social drinkers. The Neurotic-Introvert type most resembled depressed patients, and the Neurotic-Hostile type most resembled incarcerated criminals. Results clarify previous findings on DUI personality types and establish a basis for tailoring therapeutic treatments to different types of DUI offenders. Recent advances in the field of personality structure have converged, providing a consensus regarding the superordinate personality dimensions underlying a comprehensive taxonomy of personality traits (Digman, 1990; McCrae & John, 1992). The Five-Factor Model (FFM) of personality hierarchically organizes personality traits into five basic dimensions: I. Extraversion, 11. Agreeableness, 111. Conscientiousness, IV. Emotional Stability (vs. Neuroticism), and V. Intellect or Openness to Experience. These dimensions broadly assess important individual differences in interpersonal, motivational, emotional, attitudinal, and experiential style. This convergence has provided a working model allowing personality psychologists to proceed with the conceptual and empirical tasks they have long set for

Title of dissertation: GLOBAL OPTIMIZATION OF FINITE MIXTURE MODELS Jeffrey W. Heath Doctor of Philosophy, 2007 Dissertation directed by: Professor Michael Fu Robert H. Smith School of Business & Professor Wolfgang Jank Robert H. Smith School of Business The Expectation-Maximization (EM) algorithm is a popular and convenient tool for the estimation of Gaussian mixture models and its natural extension, modelbased clustering. However, while the algorithm is convenient to implement and numerically very stable, it only produces solutions that are locally optimal. Thus, EM may not achieve the globally optimal solution in Gaussian mixture analysis problems, which can have a large number of local optima. This dissertation introduces several new algorithms designed to produce globally optimal solutions for Gaussian mixture models. The building blocks for these algorithms are methods from the operations research literature, namely the Cross-Entropy (CE) method and Model Reference Adaptive Search (MRAS). The new algorithms we propose must efficiently simulate positive definite covariance matrices of the Gaussian mixture components. We propose several new solutions to this problem. One solution is to blend the updating procedure of CE and MRAS with the principles of Expectation-Maximization updating for the covariance matrices, leading to two new algorithms, CE-EM and MRAS-EM. We also propose two additional algorithms, CE-CD and MRAS-CD, which rely on the Cholesky decomposition to construct the random covariance matrices. Numerical experiments illustrate the effectiveness of the proposed algorithms in finding global optima where the classical EM fails to do so. We find that although a single run of the new algorithms may be slower than EM, they have the potential of producing significantly better global solutions to the model-based clustering problem. We also show that the global optimum matters in the sense that it significantly improves the clustering task. Furthermore, we provide a a theoretical proof of global convergence to the optimal solution of the likelihood function of Gaussian mixtures for one of the algorithms, namely MRAS-CD. This offers support that the algorithm is not merely an ad-hoc heuristic, but is systematically designed to produce global solutions to Gaussian mixture models. Finally, we investigate the fitness landscape of Gaussian mixture models and give evidence for why this is a difficult global optimization problem. We discuss different metrics that can be used to evaluate the difficulty of global optimization problems, and then apply them to the context of Gaussian mixture models. GLOBAL OPTIMIZATION OF FINITE MIXTURE MODELS

This study explored two methods that are used to assess the dimensionality of item response data. The paper begins with a discussion of the assessment dimensionality and the use of factor-analytic procedures. A number of problems associated with using linear factor analyses to assess dimensionality are also considered. A procedure is presented for hierarchical cluster analysis in combination with a new proximity measure. A simulation was performed to study how well the different cluster methods (group average, centroid, and Ward's cluster method) recovered unidimensional and multidimensional data and whether different cluster methods overor underestimated the number of dimensions in unidimensional or multidimensional data. In the simulation, only the centroid cluster method recovered the true dimensionality of simulated unidimensional data reasonably well and only in shorter tests. For all other conditions, the three cluster methods consistently overshadowed the true dimensionality of the simulated data. For three-dimensional data, Ward's cluster method was the best performing, and only the group average and Ward's cluster method recovered the multidimensional data well. Implications for practitioners are discussed. (Contains 6 tables and 46 references.) (SLD) Reproductions supplied by EDRS are the best that can be made from the original document. Assessing the dimensionality of constructed-response tests using hierarchical cluster analysis: A Monte Carlo study PERMISSION TO REPRODUCE AND DISSEMINATE THIS MATERIAL HAS BEEN GRANTED BY TO THE EDUCATIONAL RESOURCES INFORMATION CENTER (ERIC)

This study empirically develops a taxonomy that has implications for matching information technology (IT) and organizational structures. The taxonomy of IT structure is based on the degree of centralization of computer processing, capability to support communications, and the ability to share resources. By using a multistep cluster analysis, both the membership and number of groups are derived from the responses of 313 firms. Four IT structures are identified: centralized (centralized processing, low communication, low sharing), decentralized (decentralized processing, low communication, low sharing), centralized cooperative (centralized processing, high communication, high sharing), and distributed cooperative computing (decentralized processing, high communication, high sharing). Centralized computing is related to functional organizational forms with low integration and centralized decision making. Decentralized computing is related to product organizational forms with decentralized decision making. Centralized cooperative computing is related to functional organizational forms with high integration. Distributed cooperative computing is related to both matrix and product organizational forms with high integration. The ability to identify and understand the implications of IT structure is of critical importance to both academic and management practitioners.

This report details the development and trial of a method to assess the acceptability of seat belt reminder systems. The method is based on the simulation of social processes likely to occur with the introduction of the reminder systems and measurement of attitudes, motivations, and likely behaviours. The method was trialled using Australian participants recruited using a .telephone survey concerning seat belt use and related attitudes. The results were very positive with respect to the acceptability of seat belt reminders in Australia. Drivers with relatively low wearing rates were not opposed to seat belts but appeared not to have developed a seat belt wearing habit in some driving situations. Participants were generally positive about the likely introduction of seat belt reminder systems. Recommendations are made for the application of this method in Sweden and elsewhere, and for the application of the method to assess the likely acceptability of other ITS systems during their development. (a)

This paper surveys various ways in which probabilistic approaches can be useful in partitional (‘non-hierarchical’) cluster analysis. Four basic distribution models for ‘clustering structures’ are described in order to derive suitable clustering strategies. They are exemplified for various special distribution cases, including dissimilarity data and random similarity relations. A special section describes statistical tests for checking the relevance of a calculated classification (e.g., the max-F test, convex cluster tests) and comparing it to standard clustering situations (comparative assessment of classifications, CAC).

This paper proposes a new approach for discriminative clustering. The intuition is, for a good clustering, one should be able to learn a classifier from the clustering labels with high generalization accuracy. Thus we define a novel metric to evaluate the quality of a clustering labeling, named Minimum Separation Probability (MSP), which is a lower bound of the generalization accuracy of a classifier learnt from the clustering labeling. We take MSP as the objective to maximize and propose our approach Maximin Separation Probability Clustering (MSPC), which has several attractive properties, such as invariance to anisotropic feature scaling and intuitive probabilistic explanation for clustering quality. We present three efficient optimization strategies for MSPC, and analyze their interesting connections to existing clustering approaches, such as maximum margin clustering (MMC) and discriminative κ-means. Empirical results on real world data sets verify that MSP is a robust and effective clustering quality measure. It is also shown that the proposed algorithms compare favorably to state-of-the-art clustering algorithms in both accuracy and efficiency.

This paper examines bus ride times of exceptional school children in North Carolina. It uses levels of service data, which show excessive ride times for exceptional school children, to form clusters of local education agencies. From these clusters short term objectives or standards for maximum ride times are determined for each cluster. Among the findings are three distinct clusters of local education agencies (clusters 1, 2 and 3) whose maximum ride times are respectively 170 minutes, 105 minutes and 62 minutes. The maximum ride times show that only one cluster (cluster 3) provides a service whose maximum ride time is comparable to the ideal maximum ride time (long term objective) of one hour revealed by the literature and also by responses of school superintendents and transportation directors. Methods of achieving these standards are examined. A comparison of the levels of service of regular and exceptional children transportation is also provided. This study has been conducted under subcontract to Ernst and Young Inc. It is part of a larger study on exceptional children transportation funded by the North Carolina Department of Public Instruction. The views expressed herein are those of the author and not of the North Carolina Department of Public Instruction nor Ernst and Young Inc.

This paper describes a set of four BASIC computer programs for computing the Rand (1971) index of cluster similarity. The routines calculate the unadjusted Rand index; the separate routines are designed to handle problems of different sizes and alternate methods of data storage.

This paper considers the classifications produced by application of the single linkage, complete linkage, group average and Ward clustering methods to the Keen and Cranfield document test collections. Experiments were carried out to study the structure of the hierarchies produced by the different methods, the extent to which the methods distort the input similarity matrices during the generation of a classification, and the retrieval effectiveness obtainable in cluster based retrieval. The results would suggest that the single linkage method, which has been used extensively in previous work on document clustering, is not the most effective procedure of those tested, although it should be emphasized that the experiments have used only small document test collections.

This dissertation takes a relationship-based approach to cluster analysis of high (1000 and more) dimensional data that side-steps the ‘curse of dimensionality’ issue by working in a suitable similarity space instead of the original feature space. We propose two frameworks that leverage graph algorithms to achieve relationship-based clustering and visualization, respectively. In the visualization framework, the output from the clustering algorithm is used to reorder the data points so that the resulting permuted similarity matrix can be readily visualized in 2 dimensions, with clusters showing up as bands. Results on retail transaction, document (bag-of-words), and web-log data show that our approach can yield superior results while also taking additional balance constraints into account. The choice of similarity is a critical step in relationship-based clustering and this motivates our systematic comparative study of the impact of similarity measures on the quality of document clusters . The key findings of our experimental study are: (i) Cosine, correlation, and extended Jaccard similarities perform comparably; (ii) Euclidean distances do not work well; (iii) graph partitioning tends to be superior to k-means and SOMs especially when balanced clusters are desired; and (iv) performance curves generally do not cross. We also propose a cluster quality evaluation measure based on normalized mutual information and find an analytical relation between similarity measures. It is widely recognized that combining multiple classification or regression models typically provides superior results compared to using a single, well-tuned model. However, there are no well known approaches to combining multiple clusterings. The idea of combining cluster labelings without accessing the original features leads to a general knowledge reuse framework that we call cluster ensembles. We propose a formal definition of the cluster ensemble as an optimization problem. Taking a relationship-based approach we propose three effective and efficient combining algorithms for solving it heuristically based on a hypergraph model. Results on synthetic as well as real data-sets show that cluster ensembles can (i) improve quality and robustness, and (ii) enable distributed clustering, and (iii) speed up processing significantly with little loss in quality.

This chapter provides a review of the field of clustering and classification as applied in the behavioral sciences. The intent is to provide a set of recommendations or best practices in the analysis of empirical data. Clustering and classification methods are presented within a context of exploratory data analysis, as opposed to theory development or confirmation. Some methods or strategies useful for theory confirmation are discussed as appropriate.