Mining Sequential Patterns

1
Mining Sequential Patterns

• Authors Rakesh Agrawal and Ramakrishnan Srikant.
• Presenter Jeremy Dalmer.

2
Introduction

• What is a sequential pattern?
• Answers to final exam questions.

3
What is a sequential pattern?

• Requires set of attributes deciding each tuples
  class. Call this the class set.
• Exampleitems-purchasedclass set customer-id
• Tuples are sorted into classes.

4

• Requires set of attributes used for ordering
  tuples. Call this the order set.
• Exampleitems-purchasedorder set transaction-time
• Tuples within classes are sorted according to an
  order defined over order set codomain.

5

• Specifying a value for each attribute in (class
  set U order set) must be specifying at most one
  tuple. (class set U order set forms primary
  key.)
• Support and confidence measure classes now, not
  tuples.

6

• Exampleitems-purchasedorder set transaction-timecl
  ass set customer-idorange juice, beerbeerink, Band-Aids

7

• Classes Joe, Sarah

8

• Ordering within classes according to order set

9

• A large sequence (support 100) is
• (intuitive)
• I, , ,
• , , and
• are also large sequences.

10

• Exampleorder set
  year, monthclass set goldfish, lobsterlobstertiger

11

• Classes

12

• Ordering within classes (class!) according to
• order set

13

• Intuition suggests large sequence
• but this is not considered any larger than
• and
• because there is only one class.

14

• One more point about the previous example.
• Having recorded
• monkey as a large sequence, why
• record subsequences?
• and , though
• large sequences, are not informative.
• Maximal sequence.

15
final exam questions

• Root of each algorithm(1) Group into classes
  and order.(2) Find all large itemsets.(3) For
  each tuple, drop everything except a record of
  the large itemsets contained in that tuple.(4)
  Find all large sequences (of large itemsets).(5)
  Discard large sequences not maximal.

16

• Consider a previous example.

17

• Large itemsets (min-sup 100) knife, beer,
  knife, beer, Band-Aids.
• Set knife to 1, beer to 2, knife, beer to
  3, and Band-Aids to 4.
• Transform tuples to((1 2 3) (1 4))((1 2 3) (4))

18

• Large sequences (actually with 100 support)
  are((1)), ((2)), ((3)), ((4)),((1) (4)), ((2)
  (4)), and ((3) (4))
• But, since ((3) (4)) implies all the others, only
  ((3) (4)) is a maximal large sequence.

19

• Potentially large vs. Definitely large (candidate
  sequences vs. large sequences).
• Potentially large no counting, but many.
• Definitely large counting, but few.
• Algorithms similar to Apriori, but with sequences
  of large itemsets instead of large sets of items.

20

• AprioriAll Counts every large sequence,
  including those not maximal.
• AprioriSome Generates every candidate sequence,
  but skips counting some large sequences (Forward
  Phase). Then, discards candidates not maximal
  and counts remaining large sequences (Backward
  Phase).

21

• AprioriAll scans the database more, taking more
  time.
• AprioriSome keeps more potentially large
  sequences in memory, degenerating to AprioriAll
  when requests for memory fail.

22

• There were two types of algorithms presented to
  find sequential patterns, CountSome and CountAll.
  What was the main difference between the two
  algorithms?

23

• CountAll (AprioriAll) is careful with respect to
  minimum support, careless with respect to
  maximality.CountSome (AprioriSome) is careful
  with respect to maximality, careless with respect
  to minimum support.

24

• What was the greatest hardware concern regarding
  the algorithms contained in the paper?

25

• Main memory capacity. When there is little main
  memory, or many potentially large sequences, the
  benefits of AprioriSome vanish.

26

• How did the two best sequence mining algorithms
  (AprioriAll and AprioriSome) perform compared
  with each other? Take into consideration memory,
  speed, and usefulness of the data.

27

• MemoryIn terms of main memory usage, AprioriAll
  is better.In terms of secondary storage access,
  AprioriSome is better.

28

• SpeedWith sufficient memory, as minimum support
  decreases the difference between AprioriAll and
  AprioriSome increases. (AprioriSome is
  better.)More large sequences not maximal are
  generated.

29

• Usefulness of the dataFor the problem of
  finding maximal large sequences, the answer is
  Precisely the same..However, AprioriAll finds
  all large sequences, while AprioriSome discards
  some large sequences that arent maximal.
  AprioriAll, then, generates more useful
  data.The user may want to know the ratio of the
  number of people who bought the first k 1 items
  in a sequence to the number of people who bought
  the first k items.