Richard E. Orwig

Internet Categorization and Search: A Self-Organizing Approach

Abstract The problems of information overload and vocabulary differences have become more pressing with the emergence of increasingly popular Internet services. The main information retrieval mechanisms provided by the prevailing Internet WWW software are based on either keyword search (e.g., the Lycos server at CMU, the Yahoo server at Stanford) or hypertext browsing (e.g., Mosaic and Netscape). This research aims to provide an alternative concept-based categorization and search capability for WWW servers based on selected machine learning algorithms. Our proposed approach, which is grounded on automatic textual analysis of Internet documents (homepages), attempts to address the Internet search problem by first categorizing the content of Internet documents. We report results of our recent testing of a multilayered neural network clustering algorithm employing the Kohonen self-organizing feature map to categorize (classify) Internet homepages according to their content. The category hierarchies created could serve to partition the vast Internet services into subject-specific categories and databases and improve Internet keyword searching and/or browsing.

A graphical, self-organizing approach to classifying electronic meeting output

This article describes research in the application of a ( topics ) in an electronic meeting setting. Electronic meetKohonen Self-Organizing Map (SOM) to the problem of ing systems ( EMSs ) provide support for large groups classification of electronic brainstorming output and an interactively working on a single problem or collection evaluation of the results. Electronic brainstorming is one of problems ( Nunamaker, Dennis, Valacich, & Vogel, of the most productive tools in the Electronic Meeting 1991; Vogel, Nunamaker, Martz, Grohowski, & McGoff, System called GroupSystems. A major step in group problem solving involves the classification of electronic 1989 ) . Large groups of people are thereby enabled to use brainstorming output into a manageable list of concepts, a network of computers to discuss complex organizational topics, or issues that can be further evaluated by the problems electronically. These electronic discussions cregroup. This step is problematic due to information over- ate large quantities of text in a very short period of time. A load and the cognitive demand of processing a large major stage in the group problem solving process involves quantity of textual data. This research builds upon previous work in automating the meeting classification pro- classifying these large quantities of text into a manageable cess using a Hopfield neural network. Evaluation of the list or set of concepts / topics. Experience with this classiKohonen output comparing it with Hopfield and human fication process has shown that meeting convergence is expert output using the same set of data found that the problematic for participants and meeting facilitators. Kohonen SOM performed as well as a human expert in representing term association in the meeting output and The prevailing EMS provides only clerical classificaoutperformed the Hopfield neural network algorithm. In tion support for browsing the text and creating a list of addition, recall of consensus meeting concepts and top- topics for group members. However, it does not provide ics using the Kohonen algorithm was equivalent to that system support for managing or organizing the large volof the human expert. However, precision of the Kohonen results was poor. The graphical representation of textual ume of text that may be created as output from an elecdata produced by the Kohonen SOM suggests many op- tronic brainstorming session. The synthesis of electronic portunities for improving information organization of tex- brainstorming comments is a classification problem. It is tual information. Increasing uses of electronic mail, com- something that humans currently do well, but not willputer-based bulletin board systems, and world-wide ingly, a situation that suggests using an artificial intelliweb services present unique challenges and opportunities for a system-aided classification approach. This re- gence approach to understanding how humans classify search has shown that the Kohonen SOM may be used concepts and developing a system to test whether better to automatically create ‘‘a picture that can represent a classification support for groups can be provided. thousand (or more) words.’’ This research is a continuation of research previously reported in Chen, Hsu, Orwig, Hoopes, & Nunamaker

Technological support for group process modeling

During business analysis, business activities are modeled and analyzed. Redefined models become the blueprints for improved business activities. The cost to produce models of the organization is high and model accuracy is important. Involvement from knowledgeable participants and stakeholders is desirable during business modeling and analysis. Traditional modeling approaches limit direct participation to a small handful of participants. This paper discusses the development and evaluation of an electronic meeting system (EMS) based activity modeling tool. Modeling efforts supported by this new approach are compared with modeling efforts supported by analysts with a single-user tool. The results of this comparison reveal that the EMS-based modeling tool allows a greater number of individuals to participate efficiently in model development. Models are developed between 175 percent and 251 percent faster with the new approach than with the traditional approach. Specific features are discussed that help relatively novice modelers work with analysts to develop models of reasonable quality. Measures are set forth that can be used to assess modeling efficiency and quality.

Information visualization for collaborative computing

Information technology continues to generate increasing amounts of data, most of which is useless without scalable methods to collect, analyze, process, and understand it. Visualization is a promising approach to such systemization because it lets users see underlying processes and guide process simulations interactively. However, visualization must be combined with some way to make repositories of text documents more manageable, providing users with a flexible, interactive environment in which to access them. The article describes a prototype tool that addresses these problems for GroupSystems, an electronic meeting system developed at the University of Arizona and installed at more than 1,500 business, government, and university settings. The tool automatically categorizes information, statistically clusters similar documents, and displays the organized document set graphically, providing more at-a-glance information than a typical text based display. Users can thus more easily browse document collections. The tool uses text analysis techniques that aim to identify descriptors and develop an unambiguous internal representation of a document.

Facilitation Methods for Collaborative Modeling Tools

This paper presents results of an ongoing research effort to support effective user involvement during modeling and analysis meetings. Productivity and user participation of traditional group meetings have been limitations imposed by chauffeured facilitation and single-user tools. These tools have been designed for analysts rather than for direct use by non-analyst users. Recently, electronic meeting systems (EMS) modeling tools that allow users to work in parallel to contribute directly during meetings have been developed. Such tools allow more domain experts to participate directly and productively during model development meetings than is possible using the traditional approach. Although previous research has demonstrated that EMS modeling tools may be used to develop some model content, little research had been done on collaborative facilitation methods that employ these tools. This paper presents a comparison of modeling approaches for use with EMS modeling tools and proposes an approach that overcomes significant problems inherent in other approaches. It leverages the productivity enhancement afforded by direct group access and still results in production of complete, integrated, high quality models. This approach allows models to be developed two to four times faster than with traditional modeling support and yet avoids model ambiguities and inconsistencies.

A Multi-Agent View of Strategic Planning Using Group Support Systems and Artificial Intelligence

The strategic planning process is dynamic and complex. Including a Group Support System (GSS) in the problem-solving process can improve the content quality of the strategic plan by allowing increased participation by more members of the organization. However, it can also add to the complexity of the problem by increasing the quantity of textual information that can result from group activity. Added complexity increases cognitive overload and frustrations of those participants negotiating the contents of the strategic plan. This article takes a multi-agent view of the strategic planning process. It considers group participants as multiple agents concerned with the content quality of the strategic plan. The facilitator agent is responsible for guiding groups in the strategic plan construction process as well as for solving process problems such as cognitive overload. We introduce an AI Concept Categorizer agent, a software tool that supports the facilitator in addressing the process problem of cognitive overload associated with convergent group activities by synthesizing group textual output into conceptual clusters. The implementation of this tool reduces frustrations which groups encounter in the process of classifying textual output and provides more time for discussion of the concepts themselves. Because of the large amount of convergent activity necessary for strategic planning, the addition of the AI Concept Categorizer to the strategic planning process should increase the quality of the strategic plan and the buy-in of the participants in the strategic planning process.

Facilitation methods for use with EMS tools to enable rapid development of high quality business process models

An electronic meeting system (EMS) designed to support construction of IDEFO models can effectively support groups for rapid development of models. Such an EMS tool enables more domain experts to directly and productively participate during model development than traditional approaches supported by single-user tools. Although previous research has demonstrated that EMS-IDEFO tools can used to rapidly develop some model content, the corresponding facilitation methods produced models that included far too many inputs, controls, outputs, and mechanisms arrows with unspecified sources or destinations. Since such dangling connections signify incompleteness, ambiguity, and inconsistency in the models, a better approach was sought. The paper reports research on modeling approaches for use with EMS that produced different levels of productivity and model quality. Although some approaches produced suboptimal quality, a more successful approach was derived. This approach, the top-down, integrated (TDI) approach, overcomes significant problems inherent in the other approaches. The TDI approach allows models to be developed two to three times faster than traditional modeling support and yet avoids model ambiguities and inconsistencies.

Reengineering with enterprise analyzer

A flexible set of electronic meeting systems (EMS) tools and methodologies which have the potential to improve the efficiency and effectiveness of cross-functional teams involved in business process redesign was developed. Collectively, this set of tools and methodologies, and the supporting automated infrastructure, are referred to as enterprise analyzer (EA). The literature on reengineering is reviewed, and the role of EMS technology in providing support to business process redesign is discussed. The EA architecture and methodology are described. Past and ongoing reengineering applications are identified. A framework for future research is proposed.<<ETX>>

Quantitative Measures for Evaluating Knowledge Network Node Clusters: Preliminary Results

One viewpoint of a knowledge network is a knowledge map that clusters similar knowledge sources into knowledge domains. What is needed is an automatic mapping tool that 1) takes the knowledge sources, 2) creates a conceptual map of the domain space, 3) clusters like sources, and 4) places them together on the map. This research (in progress) is an attempt to determine the value of the Kohonen Self-Organizing Map for use as an interactive textual knowledge mapping tool for categorization of large sets of textual knowledge sources. Initial results have shown the algorithm to be promising in the area of creating a conceptual map of the document space, but it has been less successful at the task of clustering and assigning documents within categories. The purpose of this paper is to quantify the Kohonen algorithms ability to cluster similar documents and to explore possible improvements to it.

A method for evaluating information systems from workflow models: Results from a case study

In this dynamic age of corporate acquisitions, mergers, and downsizing, decisions concerning optimization of workflow require workflow modeling and, potentially, evaluation of legacy technology that supports the workflow as represented in the model. This paper describes a case study of workflow modeling and the method of evaluation used by a Department of Defense (DoD) to determine the optimal information system to be used in supporting the workflow. The workflow that was modeled was the business of hazardous material and hazardous waste management at any typical DoD military installation. The workflow model consisted of 65 business activities organized in a hierarchical manner as identified and defined by DoD hazardous substance and hazardous waste subject matter experts. These activities served as the foundation for the creation of a questionnaire of 165 questions that was used to select three information systems out of the over three hundred known systems that supported one or more of the 65 workflow activities within the DoD. In effect, this DoD experience demonstrated the effectiveness of an EMS-supported modeling methodology and a method for the creation of an instrument for the evaluation of legacy systems based upon the workflow model. We argue that this method is generalizable to any organization that needs to develop criteria for evaluation of information systems that support workflow.