Mining substitution rules: a knowledge-based approach using dynamic ontologies

Description

Association rule mining has traditionally focused on generating and pruning positive rules using various interestingness measures. However, relatively few studies have explored the mining of substitution rules. Existing approaches typically rely on limited definitions of substitution, either grounded in statistical relationships or based on a manager’s static knowledge. In this study, we propose a customer-centric model of substitution rule mining by adopting the lens of affordance. We employ a knowledge-based approach that incorporates a dynamic ontology, in which products are positioned based on the affordances they provide to consumers. This contrasts with traditional static ontology approaches that primarily reflect managerial perspectives. We develop an Expected–Actual Substitution Framework to evaluate the relatedness between items within both static and dynamic ontologies. Building on this framework, we introduce the Affordance-Based Substitution (ABS) algorithm to mine substitution rules in a manner that better captures consumer behavior. Additionally, we propose a novel interestingness measure that enhances the quality and relevance of the substitution rules, thereby facilitating more effective knowledge discovery. The effectiveness of the ABS algorithm is demonstrated through empirical analysis on a real-world supermarket dataset. We compare the substitution rules generated by our approach with those produced by existing algorithms in the literature. The results indicate that the ABS algorithm is more effective in capturing customer preferences and perceptions, which are often overlooked by traditional methods.

Publication Date

1-1-2018

ISBN

978-9897582752

Publisher

Scitepress

Keywords

Affordances, Dynamic Ontology, Interestingness, Market Baske, Substitution Rules

Conference

10th International Conference on Agents and Artificial Intelligence: 16-18 January, 2018, Funchal, Madeira, Portugal

This document is currently not available here.

Share

COinS