EXTRACT: MINING SOCIAL FEATURES FROM WLAN
TRACES: A GENDER-BASED CASE STUDY
By
Udayan Kumar
Ahmed Helmy
University of Florida
Presented by
Ahmed Alghamdi
Outline
2










Introduction
Motivations
Challenges and Research Questions
Contribution
Approach

Location Based Classification (LBC)

Group Behavior Based Filtering (GBF)

Hybrid filtering (HF)

Name Based Classification (NBC)
Validation of (LBC)

Temporal Consistency Validation

IBF vs. GBF

Cross Validation
User Behavior Analysis

User Spatial Distribution

Average Duration or Temporal Analysis
Device Preference
Application
Conclusion
Introduction
3

WLAN traces to understand mobile user
characteristics and behavior


This paper provide techniques to classify WLAN
users into social groups



Essential to network modeling and designing
By area
By users’ info
it presents general methodology with an example
case study of grouping by gender with investigation
of gender gaps in WLAN usage
Introduction
4

WLAN Traces

From 2 Universities (more than 50K users)

Over 3 Years





U1 - Feb 2006, Oct 2006, and Feb 2007
U2 - Nov 2007, Apr 2008
WLAN traces are logs of user association with a Wireless Access
Point (AP)
Traces generally contain

machine’s MAC address

associating time

duration

associated AP
WLAN traces are fed into a database for easy SQL retrieving
Motivations
5




Mobile devices becomes tightly coupled to users
Communication performance is bound to user
mobility and behavior
In AdHoc networks, any node can act as a router
It is imperative to understand the various aspects of
user behavior to design efficient protocols and
effective network models
Challenges and Research Questions
6



How can we meaningfully infer gender information
from such anonymous traces?
Does gender information influence user behavior
and preference in a significant and consistent
manner?
what is the impact of these finding on network
modeling, protocol and service design in the future?
Contributions
7



Class and gender inference methods based on
location, usage and name filtering from extensive
WLAN traces
Providing the first gender-based trace-driven analysis
in mobile societies, including study of majors and
device preferences
Identifying unique features in the studied grouping
that suggests consistent behavior and the design of
potential future applications
Approach
8

gender classification on campus

Location-based method


Analysis of WLAN traces

Cross validation with ground truth using Name based method





Based on individual and group network behavior
90% Accuracy
Usage patterns of males and females are different
Gender does affect user activity and vendor
preference
This contribution enhances the understanding of the
mobile society
It is essential to provide efficient network protocols
and services in the future
Approach
9

Gender-Based Grouping


Location Based Classification (LBC)
Name Based Classification (NBC)
Location Based Classification (LBC)
10





Sororities APs - female
Fraternities APs - males
CS Dept. APs - CS Students
Visitors Filtering
Visitor


Is a user with less number of sessions and smaller duration of
sessions than the average user in that location
(group behavior)
Or as user who has more sessions and larger online duration at
other locations (individual behavior)
Location Based Classification (LBC)
11

Individual Behavior Based filtering (IBF)



The probability of a user being male or female by counting
the number of sessions and measuring the duration he/she
spends in fraternities versus sororities
The probability of a user being male, considering only session
counts at fraternities and sororities
The probability of a user being male, considering only session
durations at fraternities and sororities
Location Based Classification (LBC)
12

Users visiting Fraternity and/or Sorority in decreasing order
of their Male probability (U1 feb2006)


1119 Users
425 Males
362 Females

P C M > 0.80

and PDM > 0.80
are males

PCM < 0.20
and P DM < 0.20
are females
Group Behavior Based Filtering (GBF)
13

filter a user based on where his usage pattern lies with
respect to all the users at a particular location



Find a Threshold
All users satisfy threshold are male or female due to the AP
location
All other users are visitors
Group Behavior Based Filtering (GBF)
14


Clustering: is dividing a set of users into several subsets
such that users in each subset are most similar based on
WLAN usage metrics (duration, session count, distinct
login days)
Metrics for user evaluation




Number of distinct days of login
Session count
Sum of session durations
By applying clustering technique to Sororities and
Fraternity user trace from both Universities U1 and U2



Best Cluster Size is 2 (Regular/Visitor)
Maximum width is 0.84
Minimum width is 0.65
Group Behavior Based Filtering (GBF)
15

Average Width for Sorority and Fraternities from
University U1 and U2
Clustering results for University U1 Sororities (feb2006)
Hybrid filtering (HF)
16

classification validation

compare the results from IBF and GBF



methods mainly select same set of users, which should be the case
as both methods attempt to identify regular users
for high confidence, choose the users selected by both filtering
methods
more than 90% of the users selected by GBF are common to users
selected by IBF
Name Based Classification (NBC)
17





Usernames obtained on campuses that require
authorization mechanism to access WLAN
Traces coming from university U2 provide us with
usernames University U2 also host a directory that can be
searched using these usernames
By Searching the directory first names corresponding to
these usernames obtained
from the US Social Security administration, a list of top 1000
males and females first names is used and the names
present in both lists (neutral names) are removed
this list is compared to the list obtained from university U2
directory
Name Based Classification (NBC)
18




11,000 out of 27,000 users classified as males or females in
the trace period of Nov 2007
12,500 out of 30,000 users classified as males or females in
the trace period of Apr 2008
foreign national students
non-popular names
Validation of (LBC)
19


Validation of LBC is needed to raise confidence in the
results
Three statistical methods to validate filtering
mechanisms
1.
2.
3.
temporal consistency: this method finds out regular users in
the trace set belonging to adjacent months and compares
this list to see how many are common
IBF vs GBF: this method compares results from IBF and GBF
to check the similarities in the results
Cross Validation: this method takes the classification
achieved using NBC method and compares it with the
results of LBC
Temporal Consistency Validation
20



Multiple one-month traces from one semester
Apply IBF, GBF and HF to find out the common users in
all adjacent months before and after filtering
Because users living in fraternities and sororities do not
change from one month to another in the same
semester, after filtering, the percentage of common users
should increase
Temporal Consistency Validation
21

Similarity in the user population selected after filtering
fraternity users for U1
IBF vs. GBF
22

validation mechanism that compares the results of IBF
and GBF methods
Comparing users selected by IBF and GBF for U1
Cross Validation
23






NBC has a low error rate because of using statistics from real data coming
from the US Social Security Office
Using this property of NBC, we can find out the error bound for the LBC
To calculate the error bounds, the users classified by LBC as females and
males are put in sets FL and ML
Using NBC, we classify all users from Fraternities and Sororities and put them
in sets FN and MN and remove unclassified users
The error in female classification by LBC

Ef = (FL∩MN)/FL
The error in male classification by LBC

Em =(ML∩FN)/ML
Cross validation of LBC by NBC for U2
User Behavior Analysis
24


Group classification to understand usage differences
between groups
Gender based grouping





Male
Female
Unclassified
Groups evaluated on multiple metrics depending on the
application
This paper examines the existence of differences
between genders, they used the metrics


spatio-temporal distribution for wireless usage
vendor preference
User Spatial Distribution
25



This metrics can identify where users spend most of
their time
Difference in the number of users among the genders
can tell us about the building preferences of the genders
Existence of locations, which are consistently preferred
by one of the two genders, highlights the existence of
difference in WLAN usage by two genders
User Spatial Distribution
26
Comparison of user distribution across the
university U1 campus (in Percentage)
Comparison of user distribution across the
university U2 campus (in Percentage)
Average Duration or Temporal Analysis
27

Average duration of a session for males and females
gives us an understanding of the extent of WLAN usage
at different areas
Average Duration or Temporal Analysis
28
Average duration of male and females in different
Areas of university U1 campus
Average duration of male and females in different
Areas of the university U2 campus
Average Duration or Temporal Analysis
29

Some of these differences were found to be significant
and spatio-temporally consistent even across campuses;
females’ wireless activity is stronger in Social Science
and Sports areas, whereas males’ activity is stronger in
Engineering and Music
Device Preference
30



MAC address is used to find preferred vendors for the
groups
To test whether gender provides a bias towards specific
vendors, the Chi-Square statistical significance test is
used
The Chi-Square test shows with 90% confidence that
there is a bias between gender and vendor/brand
Device Preference
31
Device distribution by manufacturer at university U1
Device distribution by manufacturer at university U2
Applications
32


The results from these metrics ca be applied to an
application to make it context sensitive
Mobility Models


Protocol Design


Mobility models are important tools to understand user
movements and create models on which protocols can be
tested
Protocol and service design in Mobile Ad-Hoc networks can
take features of various groups to evaluate its performance
Privacy
Conclusion
33



This paper proposes novel methods, which use WLAN
traces to classify WLAN users in to social groups based
on features such as gender and study-major among
others
It presents a general framework that can be applied to
traces coming from multiple sources
there is a distinct difference in WLAN usage patterns for
different genders even with similar population sizes