Behavioral Entropy of a Cellular Phone User

1
Behavioral Entropy of a Cellular Phone User

• Santi Phithakkitnukoon
• Husain Husna
• Ram Dantu (Presenter)
• Computer Science Engineering
• University of North Texas

First International Workshop on Social Computing,
Behavioral Modeling, and Prediction, Phoenix, AZ,
USA, April 1-2, 2008
2
Introduction

• Mobile phone has become an integral part of many
  peoples social lives.
• This has had profound implications on both how
  people as individuals perceive communication as
  well as in the patterns of communication of
  humans as a society.

3
How to answer the following questions with
limited information like telephone call logs

• Who are my friends, family, opt-ins, opt-outs ?
• Who is going to call me next
• When are they going to call me
• Is John busy now ?
• Is Bob ready to answer my call ?
• Where is Mary now ?
• Are there any special events in your life

4
Contributions

• We analyzed the behavior of cellular phone users
  and identify behavior signatures based on their
  calling patterns.
• We quantify and infer the relationship of a
  persons information entropy based on the
  location, time of the call, inter-connected time,
  and duration of the call.

5
Randomness Level

• Information Entropy (Shannons Entropy)
• Based on H(X), we were able to quantify
  randomness level based on
• Location of user
• Time of call
• Inter-connected time
• Duration of call.

6
Results and Analysis

• Real-life Dataset
• MIT Reality Mining Project
• Call logs of 94 mobile users of 9 months
• Results
• Correlation Coefficient
  Factor Analysis

7
Results and Analysis

• Scatter plots showing relationships among H(L),
  H(C), H(I), and H(T) with the linear trend lines

8
Conclusion

• In this paper, we analyzed cellular phone user
  behavior in forms of randomness level using
  information entropy based on users location,
  time of call, inter-connected time, and duration
  of call. We are able to capture the randomness
  level based on the underlying parameters using
  the correlation coefficient and factor analysis.
• Based on our study, the users randomness level
  based on location has high correlation to time of
  making phone calls and vice versa. Our study also
  shows that the randomness level based on users
  inter-connected time has a high correlation to
  the time spent on phone calls.
• We believe that this work can also be extended to
  predict services suitable for the user.