Thoughts on:

	 "Mining Temporal Features in Association Rules",
	 Xiaodong Chen, Ilias Petrounias
	 Department of Computing and Mathematics, Manchester
	 Metropolitan University, Department of Computation, UMIST,
	 PKDD 1999, Prague, Czech Republic pp 295-300

This paper is concerned with when to use a regular association
rule. The rules dealt with contain no temporal information
themselves. The main premise is that a rule is more applicable to
certain times than others. Supermarket transactions from 20 years ago
might not be applicable today. Transactions about summer sale items
might not be valid during the winter. This works finds all periods
during which a specific rule is valid. This work also finds when and
how often a rule is repeated. A calendar algebra describing
transaction time is used.

The authors try to give the impression that their algorithm is very
efficient, increasing linearly with the size of the database. They
fail to point out the significance that this is for a single
association rule. The association rule must be formed/found
first. This is not a quick process, taking as long as Apriori or more
to complete. I say more since you may want to consider using lower
support since one would expect to have valid, interesting rules with
less support if the transaction times which are considered valid for a
rule are less than the total number of transactions.