Facilitating CBR for Incompletely-Described Cases: Distance Metrics for Partial Problem Descriptions

Abstract

A fundamental problem for case-based reasoning systems is how to select relevant prior cases. Numerous strategies have been developed for determining the similarity of prior cases, given full descriptions of the problem at hand, and situation assessment methods have been developed for formulating appropriate initial case descriptions. However, in real-world applications, attempting to determine all relevant features of a new problem before retrieval may be impractical or impossible. Consequently, how to guide retrieval based on partial problem descriptions is an important question for CBR. This paper examines the problem of assessing similarity in partially-described cases. It proposes a set of similarity assessment strategies for handling missing information, evaluates their performance and efficiency on sample data sets, and discusses their tradeoffs.

P2P case storage and retrieval with an unspecified ontology

“…Alternatively, when the variability is high, feature values are distributed across the respective dimension, the organization of cases is sparse, and the retrieval capabilities are hampered. In order to show the correlation between the statistical properties of the data and the retrieval performance, we computed the values of pairs case , and two variability factors for the Free-Text and Numeric features 7 . var FT of a corpus was computed as a weighted average of the numbers of possible values of all the features that appear in a corpus.…”

“…This metric can be characterized as a pessimistic metric [7], as it requires exact matching of the values of the given feature. This pessimistic metric can be substituted, and softened metrics may be applied for non-matching values, e.g., assigning an average distance between the features, or a predefined value 0<x<1 to the distance between non-matching values within the given features [7].…”

“…This pessimistic metric can be substituted, and softened metrics may be applied for non-matching values, e.g., assigning an average distance between the features, or a predefined value 0<x<1 to the distance between non-matching values within the given features [7]. Although we admit that…”

“…Assume that the case c t is mapped to a location (5,6,7) in the hypercube. Then the retrieval will compare the c t with the candidate cases in the locations (*, 6, 7), (5, *, 7), and (5, 6, *), where * denotes any possible value in the respective dimension.…”

“…For example, to accomplish the comparison with all (*, 6, 7) cases, a node (5,6,7), storing the target case c t , queries his immediate logical neighbors, i.e., sends the description of c t to the nodes (4,6,7) and (6,6,7). Upon receiving this query, these nodes autonomously perform two operations: (i) forward the description of the target case c t to their next logical neighbors, i.e., nodes (3,6,7) and (7,6,7), and (ii) compute the similarity between the target case c t and the cases stored in their nodes, and back propagate the similarity value to the user storing the target case c t on the same route by which the query was received. As a result of the pure distributed P2P communication middleware of UNSO, the propagated similarity computation both parallelizes the retrieval of most similar cases and distributes the required computational overhead among the MLH nodes (i.e., the respective users managing these nodes), actually storing the cases stored in the case base.…”