On the Selection of Meaningful Association Rules
Rangsipan Marukatat · 2009
5.1 Semantic analysis Association rules can be classified by their meanings or semantics. A rule is meaningful if its whole content describes something which exists in the domain. In contrast, it is meaningless if its whole content describes something which does not exist. A partially meaningful rule is somewhere in the middle. It is comparable to a negative association rule in "A ~B" or "~A B" forms. As mentioned in Section 2, this type of association is useful in marketing research. It helps identify conflicting items that should not be promoted together, or a replacement item in case that the other is in short supply. Wu et al. (2004) gave another example. Suppose that A normally triggers an alert of event B. Rule "A ~B" suggests that the alert can be postponed because B has not yet happened. This chapter has also demonstrated how such rules were used to gain a better understanding about the domain. However, not all of them are useful for the analysis. Consider the following: 1. {V3 = 1, C5 = 1} {S2 = 2, V1 = 1, V2 = nil, V4 = nil, V5 = nil, V6 = 1} 2. {V3 = 1, C5 = 1} {S2 = 2, C1 = nil, C3 = nil, C4 = nil, C8 = nil} Both rules say that accidents involving bus and being caused by swerving in close distance is likely to occur at the intersection. They give further detail about vehicles involved and causes of accidents, respectively. The extra detail regarding vehicles involved is interesting because it is normal that an accident involves more than one vehicles. On the other hand, it is unlikely (but sometimes possible) that an accident is caused by so many reasons. Hence, adding that there is no other cause of accident is unnecessary. The above example shows different natures of the subjects. In some subjects, multiple or all the items can exist at the same time. But in the others, one or only a couple of items can coexist. Further refinement should be made to the semantic analysis to handle this. A few works have mentioned negative association rules in "~A ~B" form (Antonie & Zaiane, 2004; Wu et al., 2004; Yuan et al., 2002). None explained how to exploit such rules. Only Yuan et al. (2002) suggested that "~A ~B" is equivalent to "B A", but did not elaborate any further. To make sense of this, "~A ~B" is interpreted as that the absence of A causes the absence of B. Therefore, the presence of B would imply the presence of A. It is probable if the association () is perceived as cause-and-effect relationship.