Refining Preference-Based Recommendation with Associative Rules and Process Mining Using Correlation Distance

Abstract

Online services, ambient services, and recommendation systems take user preferences into data processing so that the services can be tailored to the customer’s preferences. Associative rules have been used to capture combinations of frequently preferred items. However, for some item sets X and Y, only the frequency of occurrences is taken into consideration, and most of the rules have weak correlations between item sets. In this paper, we proposed a method to extract associative rules with a high correlation between multivariate attributes based on intuitive preference settings, process mining, and correlation distance. The main contribution of this paper is the intuitive preference that is optimized to extract newly discovered preferences, i.e., implicit preferences. As a result, the rules output from the methods has around 70% of improvement in correlation value even if customers do not specify their preference at all.

Keywords

• process mining
• associative mining
• personalization
• recommendation
• decision support