social networks
security & privacy
Complex Networks
400M
930M
3B
3M
420M
300K
Labeling Data is Hard
Malicious Users Tend to Connect to Other Profiles Randomly
Link Prediction
Link Prediction
+
Crowd Wisdom
Topology Based
Feature Extraction
16 feautres
for directed
graphs
8 feautres for
undirected
graphs
◦ For undirected graphs:
- Common Friends
- Total Friends
- Jaccard’s-Coefficent
◦ For directed graphs:
- Transitive Friends
- Opposite Direction Friends
Meta Feature Exteraction
We extracted 7 features
- - the confidence that an edge is fake.
outline
Datasets
Network | Is Directed | Vertices Number | Links Number | Date | Labeled |
---|---|---|---|---|---|
Academia | Yes | 200,169 | 1,389,063 | 2011 | No |
Anybeat | Yes | 12,645 | 67,053 | 2011 | No |
ArXiv HEP-PH | No | 34,546 | 421,578 | 2003 | No |
CLASS OF 1880/81 | Yes | 53 | 179 | 1881 | Yes |
DBLP | No | 1,665,850 | 13,504,952 | 2016 | No |
Google+ | Yes | 107,614 | 13,673,453 | 2012 | No |
Orkut | No | 3,072,441 | 117,185,083 | 2012 | No |
Yes | 5,384,160 | 16,011,443 | 2012 | Yes | |
No | 1,053,754 | 2,161,968 | 2012 | No | |
Yelp | No | 249,443 | 3,563,818 | 2016 | No |
Fully Simulated Networks
AUC | TPR | FPR | Precision | |
---|---|---|---|---|
Simulation 1 (Arxiv HEP-PH) | 0.991 | 0.889 | 0.011 | 0.904 |
Simulation 2 (DBLP) | 0.997 | 0.994 | 0.064 | 0.993 |
Simulation 3 (Yelp) | 0.993 | 0.917 | 0.007 | 0.937 |
Semi Simulated Networks
AUC | TPR | FPR | Precision | |
---|---|---|---|---|
Academia | 0.999 | 0.998 | 0.000 | 0.997 |
Anybeat | 1.000 | 0.996 | 0.001 | 0.996 |
Arxiv HEP-PH | 0.997 | 0.953 | 0.004 | 0.965 |
DBLP | 0.997 | 0.940 | 0.005 | 0.995 |
Flixster | 0.992 | 0.990 | 0.092 | 0.990 |
Google+ | 1.000 | 0.999 | 0.000 | 0.999 |
0.999 | 0.955 | 0.005 | 0.951 | |
Yelp | 0.996 | 0.941 | 0.005 | 0.958 |
Real World Networks
Kids Friendship Network
Information gain
What About Communities?
Anomalous Communities detection
Shay Lapid, Dima Kagan, and Michael Fire. “Co-Membership-based Generic Anomalous Communities Detection”. In: Neural Processing Letters (2022).
Work in progress
-
Signed weighted networks
- Case Study: Unsupervised disease gene detection in protein-protein interaction (PPI) network
covid-19 changed everything
- 10.5% increase in active social media users.
- Instagram reported a 70% increase in viewers of live videos from February to March when lockdown measures began.
- TikTok user base increased the most, at 38%.
Zooming Into Video Conferencing Privacy
Video conferencing Privacy & Security
Information Leakage
Malware Attacks
Data Breach
Fake Avatars
Phishing Attacks
Zoombombing
Multiparty Privacy
Credit: Such, Jose M., and Natalia Criado. "Multiparty privacy in social media." Communications of the ACM 61.8 (2018): 74-81.
89,305 Zoom related tweets
90,395 Zoom related Instagram posts
16,133 video conferncing images
26,408 images from Twitter
78,435 images from Instagram
A video conferencing detection model was trained based on ResNet-50.
Accuracy = 0.969, TPR= 0.935, FPR = 0.016
89,305 Zoom related tweets
90,395 Zoom related Instagram posts
16,133 video conferncing images
26,408 images from Twitter
78,435 images from Instagram
A video Conferencing detection model was trained based on ResNet-50.
Accuracy = 0.969, TPR= 0.935, FPR = 0.016
Duplicates removed using dhash
Duplicates removed using the classifier last layer
15,709 unique images
- Face Detection - MTCNN + Azure Face API
- Face Embedding - dlib 128 dim vector
- Age detection = (Resnet-50 based model + Azure Face API)/2
- Gender Detection - Azure Face API
-
Username Recognition - EAST for detecting text location and MORAN for subsequently recognizing the composing characters.
Lemmatization was used to filter dictionary words that are not names.
Separate words were merged into names based on their distance. - Background detection - An Faster RCNN was trained to detect each participant screen
- A social graph was created by using face embedding.
- Each user was connected to all other participants in the same meeting and to all his other meetings.
- Also, a linkage can be done by the name and similar background.
What could possibly happen in zoom meeting
What could possibly happen in zoom meeting
Face
Detection
Face
Tracking
Feature
Extraction
Change
Detection
Group
Change
Detection
Shmuel Horowitz, Dima Kagan, Galit Fuhrmann Alpert, Michael Fire Interruptions detection in video conferences 2023
Face
Detection
Face
Tracking
Feature
Extraction
Change
Detection
Group
Change
Detection
Shmuel Horowitz, Dima Kagan, Galit Fuhrmann Alpert, Michael Fire Interruptions detection in video conferences 2023
Face
Detection
Face
Tracking
Feature
Extraction
Change
Detection
Group
Change
Detection
😃
🙂
🙂
🙂
Shmuel Horowitz, Dima Kagan, Galit Fuhrmann Alpert, Michael Fire Interruptions detection in video conferences 2023
Face
Detection
Face
Tracking
Feature
Extraction
Change
Detection
Group
Change
Detection
Single Parcticipent Change Detection
🙂
🙂
🙂
🙁
🙁
🙁
🙂
Shmuel Horowitz, Dima Kagan, Galit Fuhrmann Alpert, Michael Fire Interruptions detection in video conferences 2023
Face
Detection
Face
Tracking
Feature
Extraction
Change
Detection
Group
Change
Detection
Shmuel Horowitz, Dima Kagan, Galit Fuhrmann Alpert, Michael Fire Interruptions detection in video conferences 2023
Face
Detection
Face
Tracking
Feature
Extraction
Change
Detection
Group
Change
Detection
Feature Extaction | Change Detection | Recall | Precision | TNR | FPR |
---|---|---|---|---|---|
Embedding 128-dim | Arima Statistical Profiling |
50.1% 42.9% |
83.2% 75.0% |
98.1% 97.8% |
0.9% 1.4% |
Expressions 7-dim | Arima Statistical Profiling |
71.6% 72.5% |
90.5% 92.3% |
99.2% 99.1% |
0.5% 0.8% |
Deep-face expressions | Majority | 67.1% | 84.8% | 97.7% | 2.6% |
Method | Results |
---|
Thank you questions?
SOCIAL NETWORKS PRIVACY & SECURITY
By Dima Kagan
SOCIAL NETWORKS PRIVACY & SECURITY
- 64