Web-Based Architecture for Design Intelligent DSS

Size: px
Start display at page:

Download "Web-Based Architecture for Design Intelligent DSS"

Transcription

1 Web-Based Architecture for Design Intelligent DSS Guohua Bai Dept. of Computer and Systems Sciences Luleå University of Technology Sweden Abstract This paper delineates a Web-based architecture for design intelligent DSS (Decision Support Systems). After examining some rational decision theories, the paper first identifies the fundamental elements and functions (data warehouse system, modelling system, knowledge system, and interaction system) for building intelligent DSS. Based upon the identified functions and elements, a general architecture for design is outlined, and some general characteristics of each element in the architecture are discussed. In the end, an example demonstrates how agent technology could be applied in the context of intelligent DSS. The paper is based on a project proposal, and at this stage it is not able to provide with technical details for implementation. Keywords: Web-based architecture, intelligent DSS, agents, data warehouse, model base, Introduction The growth of the Web technology has created enormous opportunities for making more organisational information available to decision-makers. Client-server architecture and networks allow IS professionals to build up an enterprise-wide DSS to centralise information and yet easily to distribute it in a timely manner to decision makers. The so called intranet is providing an organisation with a flexible control over its own information resources, and meanwhile the intranet is linked to an extranet to get access to external information resources. Those days we are hearing about that intelligent agents will revolutionise the Internet, and every computer publication appears suddenly full articles about agent software. This rapid development in network-based technology and consumer-based applications explore new possibilities for developing DSS, and thus has motivated a project proposed at the Department of Computer and Systems Sciences, Luleå University of Technology, Sweden. 1. Decision process and decision support systems A special problem that confronts DSS builders today when technology is developing overwhelmingly is the emphasis on hardware and software rather than on decision-makers and decision processes. To paraphrase Keen (1997), DSS builders need to start their analyses by identifying decision makers and the decisions they need to make, and not to start with choosing a technology and readily available database. Only when a decision oriented perspective that focuses on decisions and context, rather than on technology, is it possible that future DSS will actually be improved as a result of the increasingly sophisticated technologies that are now available to DSS builder. So we start first to study how a decision is made and what information is needed, and then we can identify what functions DSS should provide to decision-makers. 1

2 1.1. Process of decision making To make a decision is to choose one of many possible actions or strategies in order to resolve a problematic situation. Different consequences resulted by each action or strategy will be evaluated according to some specified objectives. Decision making process can be approached from many different perspectives, such as individual vs. organisational, intuitive vs. analytical, rational vs. irrational (political, cultural, moral, religious, etc.). In the context of DSS, some kind of bounded rationality is needed in order to build some concrete data and models into the DSS. A DSS is applied most effectively in an interactive way to resolve a semi-structured problem. In this interaction, decision-makers set up criteria and objectives for a decision and evaluate consequences of each action, while DSS help searching all relevant information, generating all possible alternatives, and conducting sensitivity analysis. A widely accepted model of a rational-decision process has three genetic steps, i.e., intelligence, design, and choice (Simon, 1976). A conceptual picture of this decision process and tasks in each step in relation to DSS components is shown in figure 1. Intelligence Design Choice Identify Objectives Collect Data Specify problem Identify ownership Identify role-bears Design and select models Build up Criteria for choice Search for alternatives Predict and measure outcomes Analyse outcome sensitivity Select (good) alternative Evaluate of choice Update decision knowledge Plan for implementation Data Warehouse Modelling System Knowledge System DSS Functions DSS components Figure 1. Decision making process and tasks in relation to DSS components Based on the model of decision process in the above, one has to clarify at least the followings in order to make a rational decision: Information about the problem situation All available actions (strategies) and their related consequences Objectives (goals) that decide how to evaluate each action and their related consequences Principles of actions that link each action to its precondition (If-then rule) In many cases, the above are not all clear to decision-makers, and a pure rational decision is impossible. This is what we call a "semi-structured problem". Most decisions are semistructured, and the process of decision making is a heuristic and bounded-rational process Functions of a DSS based on decision process Based on the above decision analysis, we may briefly describe some main functions that a DSS should provide with to support a decision making. Support for clarifying objectives and information about the problem situation: Providing guide for where, what, and how to collect information which is concerned the decision 2

3 problem; providing functions to manage the information which includes not only quantitative data, but also qualitative data, auditory and visual data. Support for design alternatives and decision principles: providing what-if and if-then analysis; functions of model building; model validating; model updating; generating alternatives and simulating consequences. Support for system using: providing graphical user interface (GUI) and direct access to data, model; providing easy way to integrate data with models and generating decision reports. 2. Architecture of Web-based DSS The Web provides designers with a physical architecture for developing intelligent DSS. In many enterprises, a Web-based DSS is synonymous with an enterprise-wide DSS that is supporting large groups of managers in a networked client-server environment. By using Web browser like Netscape Navigator or internet Explorer decision makers in the enterprises can easily get access to decision information or decision tools from the server that is hosting the DSS components. Because of the Web infrastructure, enterprises-wide DSS can now be implemented in geographically dispersed enterprises and to link their suppliers and customers at a relatively low cost. Based on the analysis of decision making process and identified functions of a DSS, a Webbased architecture for developing an enterprise-wide and intelligent DSS is shown in figure 2. Those main components, data warehouse, model base, and intelligent agent will be discussed respectively in the followings. Intranet 1 Intern app. Intern data Internet Intranet 2 Public resources Intern app. Intern data Public resources Dept. Production Web Graphical User Interface Data warehouse process (Data mining, scan, modify) Data warehouse Modelling process (Component design, integration) (Internal, external, historical data) Dept. Marketing Intelligent Agent (Knowledge inquiring and presentation ) Models and methods base (Mathematics, operational, economic) Dept. Planning Knowledge base (Expertise, if-then rules, concepts) Web-based Enterprise-wide DSS Dept. Personal Figure 2. A Web-based Enterprise-wide DSS 3

4 2.1 Data warehouse and data mining A Data Warehouse is a database designed to support decision making in organisations. This means that a data warehouse has a different orientation of application from the traditional database systems. Database developers long understood that data could be used not only for transaction, accounting and reporting, but also for analysis, modelling, and decision making. Once these differences were understood, new databases were created specifically for analysis and decision making. These separate databases, given the name data warehouses, have the following characteristics (Gray, et al, 1999): Subject oriented: data are organised by how users refer to it Integrated: inconsistencies are removed in both nomenclature and conflicting information. Non-volatile: data do not change over time Time series: data are time series, not current status Summarised: operational data are mapped into decision usable form Larger: time series that implies much more data is retained Not normalised: DSS data can be redundant Metadata: data about the data. One specific application based on data warehouse is On-line Analytical Processing (OLAP) introduced by E.F. Codd in Codd, the father of relational databases, pointed out that relational databases are not adequate for answering decision questions. He therefore advocated the use of multi-dimensional databases that emphasised multiple analytical-function of databases. The basic idea in OLAP is that managers should be able to manipulate enterprise data models across many dimensions to understand changes that are occurring. For a software product to be considered an OLAP application it must contain three key features: multidimensional views of data, complex calculations, and time oriented processing capabilities (Gray, et al, 1999). Another important application based on data warehouse is data mining or knowledge data discovery (KDD). The mining terminology refers to finding answers from huge amount of data stored in a data warehouse. Decision makers can obtain knowledge from data warehouse by identifying 1) associations among business data, 2) logical sequences of business events over time, 3) pattern or trend over time, 4) predictions from time series, and 5) new groups or category of events. KDD applies techniques mostly from artificial intelligence to discover new knowledge. Those techniques include: statistical analysis of data neural networks, expert systems, intelligent agents multidimensional analysis; data visualisation decision trees Data mining, which is still in its early stage, has a very strong link to both modelling systems and intelligent systems. 2.2 Modelling System DSS can and should do more than to provide access to very large databases, to summarise data, to provide results quickly, and to facilitate drill-down analysis. Sparague and Carlson (1982) noted that this narrow reporting view of DSS reflects the accounting-oriented, structured reporting, information flow emphasis of much of MIS work. The information technologies to support modelling and model management have not improved as much in the past 15 years as the data base technologies. To build a model base will ask for the designers to have high abstract capabilities. Up to now the management and integration of models into enterprise-wide DSS is still very difficult. 4

5 A recent review of 11 commercially available DSS products (DSS generators) in the market (Bbargava, et al, 1999) shows that DSS products today have very little model-base functions, such as model generating, model integration, model updating, and model communication. Most products focus on one or two specific models, e.g., decision trees, influence diagram, hierarchical analysis. A complete support to all steps of a decision making process (intelligence, design and choice by Simon, 1976) is not seen in the current market. To reach full effectiveness, the functions designed in the DSS must cover the whole process of decision making. This will need theoretical understanding of general decision making process as well as empirical observation of practical decision making. Some general functions for model bases: Model generation: a flexible mechanism for building or generating models. Model modification: a mechanism for rebuilding, changing, and updating models in response to the changes of modelled situations. Model report: the output of running models should be intuitive, and easy to understand. There is a strong link between data warehouse and model bases. Many models are derived from data processing, such as data series analysis, regression and time series analysis. Results of running a model are usually in the form of serial data that will be stored and managed by data warehouse systems. Finally, models and their parameters have to be dynamically updated according to the changed data in data warehouse. 2.3 Intelligent Agent systems There is so much talk these days about software agent, such as interface agent, intelligent agent, decision-making agent. People are talking about agents that sort mails, recommend web pages, assist with scheduling, find people, and search books. The main catalysts behind the agents movement have been two rapidly accelerating trends in computing - consumer-based computing and the spread of the internet. However, there is not a clear definition to what is an agent, though we can have one person s own version of the concept. What we can do at this stage is to group some general properties, which can be used to guide the development and application of agent technology. Some properties about agent systems are summarised by (Shoham, 1999): Ongoing execution: unlike software routines that are invoked to achieve particular tasks and then disappear, agents function continuously for a lengthy period of time Autonomy: agents do not require constant human control or supervision. Environment awareness: agents model the environment in which they operate, track it, and react to change in it. Adaptiveness: over time, agents adapt their behaviour to suit the preferences and behaviour of individual user. Intelligence: agent embody sophisticated techniques for example, one based on probabilistic reasoning, machine learning, or automated planning. Agent awareness: agents model other agents, reason about them, and interact with them (for example, communicate or negotiate) using specialised protocols. Mobility: agent can migrate in a network. Anthropomorphism: agents exhibit human-like mental quality; they respond to queries about their belief or obligations convey understanding or confusion through icons depicting facial expression, or display an animation that connotes fear or friendliness. (p29) Figure 3 shows an example of an agent that helps you search an apartment. A task is specified first before the agent is sent out. It is important to clearly specify the task (even the searching paths) so the agent knows exactly where and how to find the objects. 5

6 Enter desired location: Luleå Centre Press to send searching agent Number of rooms: 5 2 Minimum Living area(m ): 120 Process going on Searching LuleåBo 20 found Searching Bovisa 50 found Process completed Choose a category: Rent x Press to show founded Purchase Invest Maximum price(skr): 5000 Figure 3. An agent searching for apartments After the agent is sent out to the Web, it will communicate with other agents (servers) which have been specified. The agent has been taught ahead of time to know those sites, it does not go to an unknown site or do a broad search and bring back trash to the user. The agent interacts with the sites, retrieves information according to the task specification, integrates the information, and finally sends results back to the users. The process as such is conceptually described in Figure 4. Data warehouse (agent LuleåBo) Result (1) Issue subtask(search 1) Data warehouse (agent BoVision) Issue subtask(search 2) Task agent Fed back result Result (2) Task specification World-Wide Web zone Figure 4. A conceptual model of the searching agent Conclusions Web technology is and will continue to change and shape the way that an organisation conducts its activities and manages information resources. By applying the Web technology and related set of technologies, such as data warehouses, data mining, agent technology, organisations will be able to construct an enterprise-wide DSS and to improve decision quality. To implement the architecture proposed in the paper is a gigantic project for any enterprise. The development process of such a system will be an iterative and stepwise process. For keeping the space, the 6

7 paper has not been able to discuss the interfaces and design perspectives, though the author is aware the incompleteness of the discussion. The author is seeking for international participants for implementing the proposed architecture. Reference Bbargava H.K, S. Sridbar and C. Herrick (1999) "Beyond Spreadsheets: Tools for Building Decision Support Systems" IEEE Computer, Vol32, No.3, pp Gray P. H.J. Watson (1999) The New DSS: data warehouses, OLAP, MDD, and KDD. AIS Americas Conference, Phoenix, AZ, (URL Keen P.G.W. (1997) let s focus on action not info. ComputerWorld, Nov. 17. Shoham Y. (1999) What we talk about when we talk about software agent, IEEE Intelligent Systems & Applications, Vol.14, No. 2, pp Simon, H. (1976) Administrative Behavior, The Free Press, N.Y. Sparague and Carlson (1982) Building an effective Decision Support Systems, Prentice-Hall, Inc, Englewood Cliffs. 7