Journal article
Horizontal case representation
2008
Abstract
This work investigates how an alternative case representation can benefit real world applications. Our motivation is that the traditional way of representing cases that requires a reasoning task and an indexing vocabulary−where values are to be assigned in each case−is unrealistically demanding and hinders the widespread use of case-based reasoning (CBR). In the real world, a single experience may include multiple reasoning tasks and/or missing values and may change with time. It is necessary that an alternative representation can perform reliably and reach levels of accuracy comparable to the traditional method. The proposed horizontal case representation requires less knowledge engineering effort and can be populated by users without knowledge of CBR; it has fewer restrictions as it is less demanding on individual feature-values. However, it provides accuracy comparable to a traditional case representation. Rather than parametrizing the distance function with weights, it requires one parameter that recommends the cardinality of values for new problems to be solved by the system. We show the performance of horizontal representations and how they compare to traditional case representation in four different case bases, how to compute the recommended cardinality factor, and demonstrate their use in the real world.
Metrics
11 File views/ downloads
8 Record Views
Details
- Title
- Horizontal case representation
- Creators
- Rosina O. Weber (Author) - Drexel UniversitySidath Deepal Gunawardena (Author) - Drexel UniversityCraig Matthew MacDonald (Author) - Drexel University
- Resource Type
- Journal article
- Language
- English
- Academic Unit
- College of Information Science and Technology (1995-2013); Information Science (Informatics)
- Identifiers
- 991014632838204721