Paul D. Whitney

Visualizing association rules for text mining

An association rule in data mining is an implication of the form X/spl rarr/Y where X is a set of antecedent items and Y is the consequent item. For years researchers have developed many tools to visualize association rules. However, few of these tools can handle more than dozens of rules, and none of them can effectively manage rules with multiple antecedents. Thus, it is extremely difficult to visualize and understand the association information of a large data set even when all the rules are available. This paper presents a novel visualization technique to tackle many of these problems. We apply the technology to a text mining study on large corpora. The results indicate that our design can easily handle hundreds of multiple antecedent association rules in a three-dimensional display with minimum human interaction, low occlusion percentage, and no screen swapping.

Multi-faceted insight through interoperable visual information analysis paradigms

To gain insight and understanding of complex information collections, users must be able to visualize and explore many facets of the information. The paper presents several novel visual methods from an information analysts perspective. The authors present a sample scenario, using the various methods to gain a variety of insights from a large information collection. They conclude that no single paradigm or visual method is sufficient for many analytical tasks. Often a suite of integrated methods offers a better analytic environment in todays emerging culture of information overload and rapidly changing issues. They also conclude that the interactions among these visual paradigms are equally as important as, if not more important than, the paradigms themselves.

User-directed Sentiment Analysis: Visualizing the Affective Content of Documents

Recent advances in text analysis have led to finer-grained semantic analysis, including automatic sentiment analysis---the task of measuring documents, or chunks of text, based on emotive categories, such as positive or negative. However, considerably less progress has been made on efficient ways of exploring these measurements. This paper discusses approaches for visualizing the affective content of documents and describes an interactive capability for exploring emotion in a large document collection.

PNNL: A Supervised Maximum Entropy Approach to Word Sense Disambiguation

In this paper, we described the PNNL Word Sense Disambiguation system as applied to the English all-word task in SemEval 2007. We use a supervised learning approach, employing a large number of features and using Information Gain for dimension reduction. The rich feature set combined with a Maximum Entropy classifier produces results that are significantly better than baseline and are the highest F-score for the fined-grained English all-words subtask of SemEval.

Minimum Hellinger distance estimation of mixture proportions

Beran (1977) showed that, under certain restrictive conditions, the minimum distance estimator based on the Hellinger distance (MHDE) between a projection model density and a nonparametric sample density is an exception to the usual perception that a robust estimator cannot achieve full efficiency under the true model. We examine the MHDE in the case of estimation of the mixing proportion in the mixture of two normals. We discuss the practical feasibility of employing the MHDE in this setting and examine empirically its robustness properties. Our results indicate that the MHDE obtains full efficiency at the true model while performing comparably with the minimum distance estimator based on Cramer-von Mises distance under the symmetric departures from component normality considered.

Human—Computer Interaction with Global Information Spaces — Beyond Data Mining

This chapter describes a vision and progress towards a fundamentally new approach for dealing with the massive information overload situation of the emerging global information age. Today we use techniques such as data mining, through a WIMP interface, for searching or for analysis. Yet the human mind can deal and interact simultaneously with millions of information items, e.g. documents. The challenge is to find visual paradigms, interaction techniques and physical devices that encourage a new human information discourse between humans and their massive global and corporate information resources. After the vision and the current progress towards some core technology development, we present the grand challenges to bring this vision to reality.

Graph-based comparisons of scenarios in intelligence analysis

Intelligence analysis relies heavily on extracting relevant information from history and using this information as models and context for interpreting evolving situations. Yet, intelligence analysts are severely limited in their capacity to identify and recall relevant past situations and cases, and the analyses that were performed on them. The typical intelligence case is a folder of disjoint memos, reports, photographs, and audio recordings. It may also contain analysis summaries captured in report form that highlight key results and findings from analyses. Unfortunately, the details of the analyses that could provide rich, contextual sources of comparison are commonly not captured nor saved for future reuse. The goal of the scenario and knowledge framework for Intelligence Analysis project at Pacific Northwest National Laboratory is to develop an effective, computable representation for capturing intelligence analyses in their full contexts and an analytical framework from which these analysis representations may be explored and compared.

Calibrating bayesian network representations of social-behavioral models

While human behavior has long been studied, recent and ongoing advances in computational modeling present opportunities for recasting research outcomes in human behavior. In this paper we describe how Bayesian networks can represent outcomes of human behavior research. We demonstrate a Bayesian network that represents political radicalization research – and show a corresponding visual representation of aspects of this research outcome. Since Bayesian networks can be quantitatively compared with external observations, the representation can also be used for empirical assessments of the research which the network summarizes. For a political radicalization model based on published research, we show this empirical comparison with data taken from the Minorities at Risk Organizational Behaviors database.

A Layered Dempster-Shafer Approach to Scenario Construction and Analysis

The ability to support creation and parallel analysis of competing scenarios is perhaps the greatest single challenge for todays intelligence analysis systems. Dempster-Shafer theory provides an evidentiary reasoning methodology for scenario construction and analysis that offers potential advantages when compared to other approaches such as Bayesian nets as it places less conceptual load on the analyst by not requiring the complete specification of joint probability distributions. This paper presents a method that can further reduce the conceptual load by taking advantage of hierarchically structured indicators. We present a novel interface for this layered, Dempster-Shafer evidentiary reasoning approach and demonstrate the utility of this interface with reference to analysis problems focusing on comparing distinct hypotheses.

A Graphical Approach to Diagnosing the Validity of the Conditional Independence Assumptions of a Bayesian Network Given Data

Bayesian networks (BNs) have attained widespread use in data analysis and decision making. Well-studied topics include efficient inference, evidence propagation, parameter learning from data for complete and incomplete data scenarios, expert elicitation for calibrating BN probabilities, and structure learning. It is common for the researcher to assume the structure of the BN or to glean the structure from expert elicitation or domain knowledge. In this scenario, the model may be calibrated through learning the parameters from relevant data. There is a lack of work on model diagnostics for fitted BNs; this is the contribution of this article. We key on the definition of (conditional) independence to develop a graphical diagnostic that indicates whether the conditional independence assumptions imposed, when one assumes the structure of the BN, are supported by the data. We develop the approach theoretically and describe a Monte Carlo method to generate uncertainty measures for the consistency of the data with conditional independence assumptions under the model structure. We describe how this theoretical information and the data are presented in a graphical diagnostic tool. We demonstrate the approach through data simulated from BNs under different conditional independence assumptions. We also apply the diagnostic to a real-world dataset. The results presented in this article show that this approach is most feasible for smaller BNs—this is not peculiar to the proposed diagnostic graphic, but rather is related to the general difficulty of combining large BNs with data in any manner (such as through parameter estimation). It is the authors’ hope that this article helps highlight the need for more research into BN model diagnostics. This article has supplementary materials online.