Semantic Analysis of Association Rules via Item Response Theory

This paper aims to install Latent trait on Association Rule Mining for the semantic analysis of consumer behavior patterns. We adapt Item Response Theory, a famous educational testing model, in order to derive interesting insights from rules by Latent trait. The primary contributions of this paper are fourfold. (1) Latent trait as an unified measure can measure interestingness of derived rules and specify the features of derived rules. Although the interestingness of rules is swayed by which measure could be applied, Latent trait that combines descriptive and predictive property can represent the unified interestingness of the rules. (2) Negative Association rules can be derived without domain knowledge. (3) Causal rules can be derived and analyzed by the Graded Response Theory which is extended model of Item Response Theory. (4) The features of consumer choice that is based on the concept of multinomial logit mode in Marketing Science could be extracted. Especially the effect of promotions and product prices based on Causal rules can be generated. Our framework has many important advances for accomplishing in mining and analyzing consumer behavior patterns with diversity.