Correlating Time-Related Data Sources with Co-clustering

A huge amount of data is circulated and collected every day on a regular time basis. Given a pair of such datasets, it might be possible to reveal hidden dependencies between them since the presence of the one dataset elements may influence the elements of the other dataset and vice versa. Furthermore, the impact of these relations may last during a period instead of the time point of their co-occurrence. Mining such relations under those assumptions is a challenging problem. In this paper, we study two time-related datasets whose elements are bilaterally affected over time. We employ a co-clustering approach to identify groups of similar elements on the basis of two distinct criteria: the direction and duration of their impact. The proposed approach is evaluated using time-related news and stock's market real datasets.