Upload
logan-bryant
View
216
Download
2
Tags:
Embed Size (px)
Citation preview
Gender based analysis
Udayan Kumar
Computer and Information Science and Engineering (CISE) Department, University Of Florida, Gainesville, FL
Inference of User Behavior
04/01/092
Problem Statement:1. How can we classify users into groups
based on gender, given publicly available traces ?
2. Can we analyze User behavior and preferences using the above classification ?
Our study is based on WLAN trace analysis*
We present a systematic technique to group WLAN users. As a case study we applied it to gender.
Information in WLAN traces
04/01/093
Trace we used :Number of Users ~4000Trace Period = Feb 2006, Oct 2006, Feb
2007
Do not provide :- User information like name, gender, major.
MAC (anonymized)
AP associations
Start time Duration
How can we classify users in groups like major and gender?
Methodology
04/01/094
visitorsvisitors
MalesFemales
University Campus
FraternitySorority
tracestraces
Fraternities house males
Sororites house females
Filtering
04/01/095
How can we distinguish visitors from residents?
Definition: Session of visitors would be significantly less than that of regular users.
This definition can be used in two ways based onIndividual behaviorGroup behavior
Group behavior
04/01/096
In this graph the “Knee” bend is a important feature
We classify those below the knee as visitors and exclude them from further studies. (order of magnitude difference in sessions)
Knee
Feb 2006 No. of Males
No. of Females
Before filtering
777 687
After filtering
452 463
Similar trend is seen in other trace samples as well
Name based UF traces have username (UFID)UF phonebook can be searched using
usernamesFinally SSN office provides a list of most
popular male and female baby names Can classify more than 25% of the users
Verification of filtering (Sororities)
04/01/099
Before Filtering
Month(a)
Month(b) # of Users(a)
# of users(b)
Common % users
Feb2006
Mar-Apr2006
991 1155 717 62.08
Oct2006
Nov2006 1264 1305 844 64.67
Feb2007
Mar-Apr2007
1169 1327 821 61.87
After Filtering
Month(a)
Month(b) Female(a) Female(b) Common % CommonFeb200
6Mar-Apr2006
463 474 429 90.51
Oct2006
Nov2006 493 456 432 87.63
Feb2007
Mar-Apr2007
439 458 405 88.43
Time Evolution*
04/01/0910
*Udayan Kumar, Nikhil Yadav and Ahmed Helmy, “Analyzing Gender-gaps in Mobile Student Societies”, CRAWDAD Workshop poster
3 days 4 days 5 days
6 days 7 days 14 days
Using this info
04/01/0911
Problem Statement: How can we classify users into groups
based on gender, given publicly available traces ?
Can we analyze User behavior and preferences using the above classification ?
We analyze the traces with the following metrics Major (Buildings most frequented) Online activity (Session duration) Device preference (Apple or PC)
WLAN User Distribution*
12
• Engineering and economics have more males, while social science has more females.
*Udayan Kumar, Nikhil Yadav and Ahmed Helmy, “Gender-based Grouping of Mobile Student Societies”, The International Workshop on Mobile Device and Urban Sensing (MODUS), IPSN 2008 Workshop
Average session duration
13
Are there trends in the average online times of users ?
Males have shorter sessions than females.
*Udayan Kumar, Nikhil Yadav and Ahmed Helmy, “Gender-based Grouping of Mobile Student Societies”, The International Workshop on Mobile Device and Urban Sensing (MODUS), IPSN 2008 Workshop
14
Device preferences: Which device vendors do different genders prefer?
Females seem to prefer Apple over Intel
Preference
Statistical significance test says with 90% confidence that there is bias within genders for brands/vendors
*Udayan Kumar, Nikhil Yadav and Ahmed Helmy, “Gender-based Grouping of Mobile Student Societies”, The International Workshop on Mobile Device and Urban Sensing (MODUS), IPSN 2008 Workshop
What can we do with it ? (Applications)
04/01/0915
• Profiling : identify gender, preference, major for a user
• Social Science: extent of WLAN adoption amongst genders can point out socio-economic and socio-cultural gaps between genders.
• Directed Ads: If an advertiser of male product want to target its users, he can find from the traces regions of campus frequented by men.
• Application Customization: Based on preference of the genders
• Designing network protocols for DTN (Delay Tolerant Networks)