1. DENSE ITEMSETS JOUNI K. SEPPANEN, HEIKKI MANNILA SIGKDD2004
Speaker: Li HueiJyun
Advisor: Koh JiaLing
Date:2007/10/2

2. OUTLINE
Introduction
Dense Itemsets
Algorithms
Experiments
Conclusion

3. INTRODUCTION
The usual way to find association rules is to seek first frequent itemsets.
For an itemset to be found frequent, all of its items must co-occur sufficiently often.
Problems:
Such situations are rare in real-world.
Connections between items may exist that are manifested by co-occurrence of not the full set of items but of varying subsets.

4. INTRODUCTION ε-approximate frequency f
An itemset have at least a proportion 1-ε of items present in at least a proportion f of database rows.
Problems:
Any frequent itemset will generate many approximately frequent itemsets that do not convey any meaningful information.
The usual kind of itemset mining algorithms like Apriori are not easily generated to the new task.

5. INTRODUCTION Example of problem1: Example of problem 2: ε = 0.5
ABCDE、ABCFGH、ABCDFGH、ABCDFGH
Example of problem 2:
The itemset ABCD has 0.5-approximate frequency 100%
50% for A, 83% for AB, 67% for ABC

6. DENSE ITEMSETS
Def1
Def2

7. DENSE ITEMSETS def3 Example 1. In the database <ABCDEFGH,ra>
wdens(ABCDE, ra) = 1
wdens(DEF, ra) = 2/3
wdens(FGH, ra) = 0

8. DENSE ITEMSETS
def4

9. DENSE ITEMSETS
def5

10. ALGORITHMS
algorithm1

11. ALGORITHMS
algorithm2

12. EXPERIMENTAL RESULTS
Table3
Table4

13. EXPERIMENTAL RESULTS
Table5

14. CONCLUSION Introduced the concept of dense itemsets.
Gave algorithms for finding all or the top-k dense itemsets.
Filtering and reordering techniques need to be developed for handling large collections of patterns.
External intersection counts: measuring not only how many items of a given set but also how many items of its complement appear in a transaction.