ITLMINE: Mining Frequent Itemsets More Efficiently

1
ITL-MINE Mining Frequent Itemsets More
Efficiently

• Raj P. Gopalan Yudho Giri Sucahyo
• School of Computing
• Curtin University of Technology
• Bentley, Western Australia 6102
• raj, sucahyoy_at_computing.edu.au

2
Introduction

• Association rule mining Finds interesting
  relationships among items in a data set.
• Two steps
• Find the frequent itemsets.
• Use the result of Step 1 to generate association
  rules.
• Step 1 computationally very expensive.
• So, focus of significant research effort.

3
Finding Frequent Itemsets

• Two general approaches
• Candidate generation-and-test
• Apriori and its variants
• Pattern Growth Approach
• FP-Growth, H-Mine

4
Contributions

• Present a new data structure called
  Item-TransLink (ITL).
• Propose a more efficient algorithm (called
  ITL-Mine) based on the pattern growth approach.
• Performance compared with Apriori and H-Mine
  algorithms.

5
Association Rules

• Given a database of transactions containing
  various items, statements of the form
• A ? B (10, 80)
• 80 of transactions that purchase A also purchase
  B and 10 of all transactions contain both of
  them.

6
Binary Representation of Transactions
Sample database
7
ITL Data Structure

• Based on these observations
• Item identifiers may be mapped to a range of
  integers.
• Transaction identifiers can be ignored provided
  the items of each transaction are linked
  together.

8
ITL Data Structure

• ItemTable
• Every item, with its support and a link to the
  first occurrence in TransLink.
• TransLink
• Every transaction in database, with items in
  sorted order.
• Each item has a link to the next occurrence.

9
ITL-MINE Algorithm

• Three steps
• Construct ItemTable and TransLink from
  transaction database.
• Prune any item below minimum support.
• Mine Frequent Itemsets of 2 or more items.
• Algorithm details in paper.

10
Example
11
Performance Study

• CHESS
• 3196 trans
• Max 37 items/trans

• MUSHROOM
• 8124 trans
• Max 23 items/trans

12
Performance Study
BMS-Web-View1 - 59602 trans - Max 267 items/trans

• T25I10D10K
• 10000 trans
• Max 25 items/trans

13
Comparing with H-Mine

• ITL-Mine traverses the database only once. H-Mine
  traverses twice.
• ITL remains unchanged during mining. H-struct
  in H-Mine continually re-adjusted.
• ITL-Mine builds a TempList and uses
  tid-intersection. H-Mine builds a series of
  header tables linked to H-struct.

14
Conclusion

• A generic data structure (ITL) and a new
  algorithm (ITL-Mine) were presented.
• ITL-Mine outperforms Apriori and H-Mine on
  typical data sets.

15
Further Work

• Extend ITL-Mine for very large databases.
• Integrate constraints into ITL-Mine.