C-TREND: A New Technique for Identifying and Visualizing Trends in Transactional Data

C-TREND: A New Technique for Identifying and Visualizing Trends in Transactional Data Gediminas Adomavicius, Jesse Bockstedt Department of Information and Decision Sciences Carlson School of Management University of Minnesota { [email protected], [email protected] } Last Modified: February 1, 2007 Abstract. We introduce C-TREND, a new temporal data analysis and visualization technique for representing trends in multi-attribute transactional data. C-TREND performs meta-analysis on standard agglomerative hierarchical clustering solutions for multiple time period data partitions to generate trend graphs. C-TREND uses nodes to represent commonly recurring transaction types in each period and edges to represent cross-period relationships between the common types. In this paper, we provide an overview of the C-TREND technique, discuss the procedures for node and edge discovery from temporal data, discuss the implementation and performance of C-TREND, and demonstrate an application of the technique by analyzing the change in technical characteristics of wireless networking technologies over five years.