Brought to you by EarthWeb
IT Library Logo

Click Here!
Click Here!

Search the site:
 
EXPERT SEARCH -----
Programming Languages
Databases
Security
Web Services
Network Services
Middleware
Components
Operating Systems
User Interfaces
Groupware & Collaboration
Content Management
Productivity Applications
Hardware
Fun & Games

EarthWeb Direct EarthWeb Direct Fatbrain Auctions Support Source Answers

EarthWeb sites
Crossnodes
Datamation
Developer.com
DICE
EarthWeb.com
EarthWeb Direct
ERP Hub
Gamelan
GoCertify.com
HTMLGoodies
Intranet Journal
IT Knowledge
IT Library
JavaGoodies
JARS
JavaScripts.com
open source IT
RoadCoders
Y2K Info

Previous Table of Contents Next


7.3. CBR IN THE CASE OF MULTIPLE MEDIA

An interesting application development framework that CBR might consider for adoption is the one used in the treatment of multimedia (image + text + structured data) applications such as CAFIIR (Wu, 1997). The concept of iconic indexing used in the development of CAFIIR resembles very closely to the needs of the CBR for maintaining cases and subcases. CAFIIR stands for Computer Assisted Face Identification and Information Retrieval system. It is primarily a mug-shot system that is used to retrieve from a composite of a suspect's face and then search a database using the composed face and other descriptions of the suspect and crime. The system has several features for input. Some of the features are derived from the image input and take on continuous values. Some other features are structured data about the suspect (such as height, weight, color of the eyes, hairstyle, etc.) and/or information from the scene of the crime, such as the type of tool used in the assault, etc. Other features are unstructured description about the scene of the crime and witness descriptions.

Iconic indexing uses the concept of subgroupings at different levels where each level corresponds to a feature. Subgroups in each level store pointers to the cases (or objects) that have similar values for that feature. There is also a candidate chosen to represent that group and such a prototypical candidate would have its feature value equicentric to all other cases in that subgroup.

Iconic indexing uses an icon to visually represent the prototypical candidate for each subgroup. The index tree starts with the most important feature and unfolds into successive layers of less important features. The index is built to handle very large collections and the scheme allows navigation from the root of the index tree to the different features, one feature at a time, as well as across different subgroups corresponding to a feature. See Figure 3, an iconic index in CAFIIR.

Where possible, the subgroups can be organized according to semantics. In other cases, it is possible to organize the subgroups according to some arbitrary criteria such as balanced distribution of cases across the subgroups or organizing the subgroups around the modes in the distribution of values for that feature.

Unstructured information is dealt with using traditional IR techniques. The results from these two are combined to present a consolidated rank list of cases in response to search for cases similar to the problem case.

7.4. CBR AND MODEL-BASED REASONING

Model-based reasoning emphasizes the use of large chunks of general knowledge, based on models that cover the normative situations. Model-based reasoning is usually used for well-understood domains that can be accurately represented in a formal language, and which tends to be static. Hence, building a model is time-consuming, and so is maintaining it. In combining model-based reasoning with CBR, the model-based reasoning handles the well-understood components, while the CBR component covers aspects of the domain that are "weak theory." In addition, by incorporating domain knowledge into the CBR system, it may also address the research challenges of case library organization (for more efficient search and retrieval) and case adaptation.

8. CONCLUSION

Case-Based Reasoning (CBR) has come a long way from a laboratory model to a methodology supported by commercial products. There is no doubt that CBR is becoming increasingly important, judging from the number of conferences and workshops focusing on CBR, CBR research papers being published in Europe and the U.S., the number of World Wide Web pages on CBR, and the CBR mailing lists (e.g., CBR-Newsletter, AI-CBR and CBR-MED).

One future challenge lies in the realm of CBR tools, which are currently inadequate for real-world applications. Better tools are needed for representing cases, indexing, designing and building case-based systems, adaptation, and integration with other reasoning paradigms and technologies (such as multimedia, information retrieval, databases, etc.). Methodologies for building and validating case-based systems, maintaining the case library, and for collecting cases are needed as currently, CBR systems are built in an ad-hoc manner, which is very much dependent on the application domain and the common-sense experience of the knowledge engineer. As CBR systems become part of an organization's information and decision-support system, the case libraries will becoming the "organizational memories" of the organization, capturing the experiences of its human resource. "It is the sum of everything everybody in your organization knows that gives you the competitive edge" (Fortune Magazine, July 1992). Protocols are then needed for organizational and control issues in a distributed system, such as, who can update the case library and how is it to be managed?

The case studies presented in this section highlight the exploitation of CBR in real situations and the resulting benefits. Widespread use of CBR will take place when CBR is able to leverage concepts and technologies from related disciplines. Database technologies can help in building larger case bases. Information retrieval technologies can help in the handling of unstructured information in the cases. Content-based retrieval can help broaden CBR's deployment to include multimedia applications. We have outlined several interesting research directions that are worth pursuing. We do hope that these suggestions will spur further research and development activities in making CBR a multimodal, robust, and scalable technology.


FIGURE 3 Ionic index in CAFIR.


Previous Table of Contents Next

footer nav
Use of this site is subject certain Terms & Conditions.
Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details.