Discovering Calendar-based Temporal Association Rules

Speaker: Sindy SHOU

Abstract

A temporal association rule is an association rule that holds during specific time intervals. An example can be that eggs and coffee are frequently sold together in morning hours. This paper studies temporal association rules during time intervals specified by user-given calendar schemas. Generally, the use of calendar schemas makes the discovered temporal association rules easier to understand. An example of calendar schema is (year, month, day), which yields a set of calendar-based patterns of the form <d3, d2, d1>, where each di is either an integer or the symbol *. For example, <2000, *, 16> is such a pattern, which corresponds to the time intervals consisting of all the 16th days of all months in year 2000. This paper defines two types of temporal association rules: precise-match association rules that require the association rule hold during every interval, and fuzzy-match ones that require the association rule hold during most of these intervals. Compared to the non-temporal association rule discovery, temporal association rules are more difficult to find due to the usually large number of possible temporal patterns for a given calendar schema. The paper extends the well-known Apriori algorithm, and also develops two optimization techniques to take advantage of the special properties of the calendar-based patterns. The paper then studies the performance of the algorithms by using a real-world data set as well as synthetic data sets. The performance data show that the algorithms and related optimization techniques are effective.

Referred Papers