In context of data mining, what does it mean for an association rule measure to be maximal?
I cannot understand the term maximal in this context.
I know of maximal independent sets in algorithms but cannot make out this term.
$\begingroup$
$\endgroup$
3
-
$\begingroup$ can you be more precise ? where did you read this ? $\endgroup$– AJedCommented Jan 13, 2013 at 5:49
-
$\begingroup$ but usually, a maximal subset $S$ with certain charatestics is a subset such that there is no other subset $S'$ with the same characteristics and $S \subset S$. That is, cannot extended anymore. $\endgroup$– AJedCommented Jan 13, 2013 at 5:52
-
1$\begingroup$ I read it in the following paper infolab.stanford.edu/~sergey/dic.html . If you search for "conviction is truly a measure of implication because it is directional, it is maximal for perfect implications". This is the statement I cannot make out $\endgroup$– user1008Commented Jan 13, 2013 at 6:20
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
I believe that in the context of that paper (PS), you can read "maximal" as "maximized". That is to say, the conviction function $P(A)P(\neg B)/P(A, \neg B)$ attains its greatest value when $A \implies B$.
