‘Clustering the objective interestingness measures based on tendency of variation in statistical implications’ by Ngia Quoc Phan, Vinh Cong Phan, Hung Huu Huynh, Hiep Xuan Huynh
Second most downloaded paper from EUDL for May 2016, from Issue 9 of EAI Endorsed Transactions on Context-Aware Systems and Applications
You can read about the most downloaded paper for January and May 2016 here.
Objective interestingness is a crucial factor in all context-aware systems as is self-evident from its name. The success of recognizing important and relevant environmental cues hinges on the system’s ability to evaluate that relevance. This is especially important in the post-processing stage of data mining. The list of methods to measure objective interestingness is long and keeps growing, and plus, different methods work better in different situations. This paper proposes a method of clustering measures of objective interestingness based on tendency of variation in statistical implications.
This new approach uses a hierarchical structure of similarity tree to cluster the objective interestingness measures with agreed assymetrical properties. The results of tendency variation in statistical implications are based on partial derivatives of the calculated function of measures on each parameter to build a distance matrix of the measures. As a result, each cluster is a group of measures that have proximity or similarity to each other. This enables users to better choose appropriate measure for their application.
Want to know the details? Get the full paper for free on EUDL.