Olcay Cirit - Consumer Profiling

Document Sample
Olcay Cirit - Consumer Profiling Powered By Docstoc
					     Consumer Profiling
using Fuzzy Query and Social
    Network Techniques


    Sema Alptekin, F. Olcay Cirit,
        Masoud Nikravesh
          Consumer Profiling


         Targeted Advertising


           Increased Revenue

Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
          Traditional Profiling

                     vs.

             Social Profiling


Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
         Motivating Research:

               Homophily

         Innovation Diffusion



Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
             Innovation Diffusion




    Implication: It is easiest to sell to those nearest the sold.


Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
              How and Why

           Web Communities

      Should Collect Social Data


                            images used without permission. sorry!



Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
    Unifying Different Modalities:

                  Chat
             Message Boards
                 Voice
                 Video

             7% 38% 55%

Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
   Why is online correspondence a
           good measure?
       – Immediacy Principle

       – Models ‘Buzz’

       – Easy to track, more Honest

Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
    The “Club” Profiling Model:

             The Chess Club

  The Acme Widget Buyer’s Club



Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
    BISC Decision Support System

 - Fuzzy Queries over a Database

 - G.A. for Query Refinement

 - Perfect for Traditional Profiling


Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
   Integrating Traditional Profiles
        with Social Profiles:

   1.Calculate a dynamic S-Score
     for each person

   2.Give this score to the DSS


Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
  S-Score: Significance & Derivation
 A                               B
           0.4       0.1
     0.6                                       - If the Social Network is thought of as a
                           0.2
                 C                             Fuzzy Relation, the S-Score is the direct
     0.5
                           0.2                 image of the Club on that Relation.
                     0.1
           0.5
                             D                 - Can be implemented as one Matrix
  F
                                               Multiplication to calculate everyone’s
                           0.3       0.3
                                               score.
                                 E
                                               - Conceptually, this score represents the
  A B C                D         E         F   social nearness of each person to the
A 1 0 .4               0         0         0   Club, or the amount of Buzz he/she might
B 0 1 .1               0         0         0   hear from/about the club.
C … …

 Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
                  Profiling Process
                                         New
                                    Customers
    “Blanket”          Targeted                 Refine
    Advertising       Advertising               Profile
    Initial
                     Potential
 Customer
                    Customers
      List


        Initial
                       DSS Query
        Profile



                         Other      S-Scores
                         Data


Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
     Other Applications/Extensions
- Customer Relationship Management: target more than just
ads

- Targeted Email Ads

- Predicting political outcomes in Thai Parliament (Dowpiset
1999)

- Anti-SPAM? Counter-terrorism?




Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
    Ethical/Legal Considerations
  • Does this constitute a privacy policy
  violation? Will people find it unpalatable?

  • Is maintaining this data a subpoena
  waiting to happen?

  • What if targeting tips off friends/family
  to embarrassing habits/conditions?



Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh
                 Conclusion
  • These ideas are hypothetical,
  representing a work in progress.

  • Work on this project will continue into
  the next year.

  • Any large web portals willing to take the
  plunge?



Sema Alptekin, F. Olcay Cirit, Masoud Nikravesh

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:2
posted:2/9/2012
language:English
pages:16