Making friends on Facebook Isaac Rahul Alex and

  • Slides: 12
Download presentation
Making friends on Facebook Isaac, Rahul, Alex and Kai

Making friends on Facebook Isaac, Rahul, Alex and Kai

Outline Introduction System framework Challenges Work in progress Conclusion and future work Demo

Outline Introduction System framework Challenges Work in progress Conclusion and future work Demo

Introduction Motivation Evaluate the difficulty of penetrating Facebook communities Methodology Investigate people’s preferences in

Introduction Motivation Evaluate the difficulty of penetrating Facebook communities Methodology Investigate people’s preferences in Facebook Make profiles with controlled sets of attributes Build an automated friend request sending system Analyze the responses to study people’s preference Byproducts include response time, geography Implications Advertising, spamming, phishing, etc Learn about Facebook’s current security policies Help direct future work in the detection of malicious profiles

System framework Build/debug an automated Facebook friending system Automated User ID crawler Automated friend

System framework Build/debug an automated Facebook friending system Automated User ID crawler Automated friend requester Automated profile scraper

Challenge during our research Automated profile creation Prevented by CAPTCHAs Source IP detection So,

Challenge during our research Automated profile creation Prevented by CAPTCHAs Source IP detection So, we manually create them Fewer needed than originally assumed Automated friend request sending Every friend request needs a CAPTCHA until the profile is cell -phone validated or the profile has an authorized e-mail address (e. g. , university email address) Facebook has a upper bound of 5, 000 friends per profile Facebook limits the number of pending friend requests We were temporarily blocked after send about 450 friend requests at once

Challenge (cont. ) Given such limitations Original approach: Friend request flooding New direction: Investigate

Challenge (cont. ) Given such limitations Original approach: Friend request flooding New direction: Investigate which kinds of people are most likely to accept friend requests from which kinds of profiles

Work in progress Gender effect Create a male and female profile Friend 1000 Facebook

Work in progress Gender effect Create a male and female profile Friend 1000 Facebook users Wait for the response 3 day timeout

Work in progress (cont. ) Preliminary results 2 profiles (using Kai and Yinzhi NU

Work in progress (cont. ) Preliminary results 2 profiles (using Kai and Yinzhi NU email) Using male profile, we sent 400 requests at 8 pm on 3/15/09, till 4 pm 3/16/09, we’ve harvested 67 acceptance Confirm with no reason Reply with questions, such as, “do I know you? ” Unfortunate, requests from our female profile got crashed because of the machine problems. (sent at a latter time, till 4 pm 3/16/09, we’ve harvested 38 acceptance)

Work in progress (cont. ) Preliminary results However, the comparison may lose statistical significance

Work in progress (cont. ) Preliminary results However, the comparison may lose statistical significance when time windows are different. (will redo this) Analysis we will do after the experiments has been redone Effect of gender (standard t-Test for this) Within each gender, who are the people accepting the requests (for example, male/female)? Temporal distribution of acceptance

Work in progress (cont. ) Limitations of our method Non-atomic evaluation of friend requests

Work in progress (cont. ) Limitations of our method Non-atomic evaluation of friend requests Computer errors can lead to friend requests being sent at different times

Future work Mutual friends Interests How facebook recommended users work? Age Image ……

Future work Mutual friends Interests How facebook recommended users work? Age Image ……

Conclusion The above-mentioned challenges forced us to change the direction of our research several

Conclusion The above-mentioned challenges forced us to change the direction of our research several times We currently do not have enough data to make any conclusion We will demonstrate some tools we develop in our friending system Automated user IDs crawler Automated friends requestor