An Aging Theory for Event LifeCycle Modeling

1
An Aging Theory for Event Life-Cycle Modeling

• From IEEE TRANSACTIONS ON SYSTEMS, MAN,
• AND CYBERNETICSPART A SYSTEMS AND
• HUMANS, VOL. 37, NO. 2, MARCH 2007
• Author Chien Chin Chen, Yao-Tsung Chen,
• and Meng Chang Chen
• 
  Tseng Chih-Ming 2007/07/24

2
Outline

• Introduction
• Related Works
• Aging Theory
• Event Detection and Tracking
• Empirical Evaluation
• Conclusion

3
Introduction (1)

• An event can be described by a sequence of
  chronological documents from several information
  sources.
• The goal of event detection and tracking is to
  automatically identify events and their
  associated documents during their life cycles
  (birth, growth, decay, and death).

4
Introduction (2)

• Conventional document clustering and
  classification techniques cannot effectively
  detect and track sequential events, as they
  ignore the temporal relationships among documents
  related to an event.
• Another difficulty of event detection is context
  shifting. During the life span of a sequential
  event, the themes of supporting documents may
  change frequently.

5
Related Works
6
Aging Theory(1) -- Definitions

• Xt the total support from its supporting
  documents in a time slot t.
• Yt g(x1, x2, .., xt, a, ß) the accumulative
  support at time t.
• a the support transfer factor, decides the
  influence of documents on the life of an event, 0
  ? a ? 1.
• ß the support decay factor, governs the pace of
  aging , 0 ? ß ? 1.

7
Aging Theory(2) -- Definitions

• F(y) energy function.
• 0 F(y) 1, F(y) is a strictly
  increasing function of y
• To adopt a sigmoid function as our energy
  function
• F(y) 10y/(1 10y), ygt0
• 0, otherwise.

Redefine F(r yT) s T is the number of time
slots yT is the accumulative support. s
energy value r percentages of an events
supporting documents appear Three aging schemes,
growth only, constant decay (CD), and recursive
decay (RD),
8
Aging Theory(3)

• Growth-only aging scheme
• yT aX1 aX2 aX3 .. aXT Si
  1,.,T(aXT)
• F(r yT) s ? ?
• Constant Decay (C.D.) aging scheme -- To emulate
  fading energy, the CD aging scheme subtracts a
  constant support ß from the accumulative support
  for every time slot.

9
Aging Theory(4)

• Recursive Decay (RD) aging scheme

Xt -1
.
X2
X1
Time
10
Event Detection and Tracking(1)

• Both documents and events are represented as a
  vector
• wt,d tft,d log(N / dft)
• where wt,d is the weight of term t in
  document d tft,d is the term frequency (TF) of
  term t in document d log(N/dft) is the inverted
  document frequency (IDF) of term t N is the
  number of documents in the systems corpus dft
  is the number of documents in the corpus where t
  occurs.
• To update the term weights of an event
  incrementally.
• wt,e (1 - ?) wt,e ? wt,d
• where wt,e is the weight of term t of event e, ?,
  in 0, 1, is a parameter that adjusts the
  contribution of document d to event e.

11
Event Detection and Tracking(2)

• For the CD scheme, the function e.EnergyUpdate()
  is defined
• e.eng F (F-1 (e.eng) a e.xt - ß)
• For the RD scheme, the function e.EnergyUpdate()
  is defined
• e.eng F (a (ß F-1(e.eng) (1 - ß) e.xt))

12
Empirical Evaluation (1)

• each incoming document is first classified into
  its most appropriate category according to 18

13
Empirical Evaluation (2)
14
Empirical Evaluation (3)
15
Conclusion

• Without a proper life-cycle model, an event may
  be unnecessarily prolonged by merging similar
  documents of different events or shortened by
  rejecting follow-up documents of the same
  events.
• In this paper, we have proposed an aging theory
  that models the life cycle of an event, and
  incorporate it into a traditional single-pass
  clustering algorithm to adaptively detect and
  track online sequential events.

16
The Major Components and Workflow in ACIRD
18 ACIRD Intelligent Internet Document
Organization and Retrieval (IEEE Trans. Knowl.
Data Eng. 2002)