Evaluation of ordinal attributes at value level

Abstract

We propose a novel context sensitive algorithm for evaluation of ordinal attributes which exploits the information hidden in ordering of attributes' and class' values and provides a separate score for each value of the attribute. Similar to feature selection algorithm ReliefF, the proposed algorithm exploits the contextual information via selection of nearest instances. The ordEval algorithm outputs probabilistic factors corresponding to the effect an increase/decrease of attribute's value has on the class value. While the ordEval algorithm is general and can be used for analysis of any survey with graded answers, we show its utility on an important marketing problem of customer (dis)satisfaction. We develop a visualization technique and show how we can use it to detect and confirm several findings from marketing theory.

Ordinal Evaluation: A New Perspective on Country Images

“…The direct interpretation and analysis of these numbers is of course possible, but visualization makes it easier. The slope visualization proposed by [11] (upward and downward reinforcement are represented with the steepness of the line segment between two consecutive feature values) is unusual for marketing research practitioners and, as we argue below, does not convey all the information necessary for this specific field. We therefore propose a marketing friendly visualization of the OrdEval results on Fig.…”

“…In particular, Prospect Theory and its assumptions of loss aversion and diminishing sensitivity has brought several marketing scholars to alter their views on classic decision making mechanisms such as upward and downward reinforcement. For instance, within the field of customer satisfaction, a number of studies were capable of demonstrating that the relationship between attribute levels and overall satisfaction is rather asymmetric and S-shaped [7,8,11]. In contrast with research on customer satisfaction, the COO-literature about asymmetric and non-linear mechanisms is rather scarce.…”

“…Only recently, Robnik-Šikonja and Vanhoof [11] introduced the Ordered Evaluation (OrdEval) algorithm. It can be used for analysis of any data where the dependent variable has ordered values, meaning that is suitable also for surveys where answers are given in the graded manner.…”

“…The OrdEval algorithm [11] is a novel analytical tool which emerged in data mining context with purpose to evaluate the importance and impact of various factors in the given data (e.g., survey). For example, in the analysis of customer satisfaction data for a particular product/service, OrdEval can determine the importance of each product's feature to the overall customer's satisfaction, and also indicate the thresholds where satisfaction with individual feature starts having strong positive or negative impact on the overall satisfaction.…”

Online random forests based on CorrFS and CorrBE

“…These methods evaluate features based on different criteria that include impurity decrease, Euclidean distance and information entropy respectively. For a valuable discussion on several measures for estimating attribute's quality in batch learning classification problem, the reader is directed to [5,8,9]. The remainder of this paper is organized as follows.…”