towards detecting social circles on twittercanf/cs533/hwspring2016/projects/final... · • create...
TRANSCRIPT
![Page 1: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/1.jpg)
Towards Detecting Social Circles on Twitter
Arda ÜnalHüseyin Celal Öner
![Page 2: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/2.jpg)
Outline• Problem Definition• Applied Methodology• Data Set• Evaluation Metric• Results
2
![Page 3: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/3.jpg)
How Do People Use Twitter?• Many Twitter users
follows friends, family, celebrities, sport teams or even brands.
• Barack Obama approximately follows 636,000 users on Twitter.
3
![Page 4: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/4.jpg)
Problem Definition• The personal social network of many
Twitter users is big and untidy.
• In 2009, Twitter released a feature, “lists”
• Designed to provide a better way to organize information on Twitter.
• Using lists to organize information on Twitter is a tedious task
4
![Page 5: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/5.jpg)
Project Goal• Automatically infer users' social circles using the set of connections between
users and their social network profiles.
5
![Page 6: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/6.jpg)
Applied Methodology 1. Smooth out inconsistencies in the basic adjacency matrix so that it gives a better notion of the
degree of connectedness of various node. (Exponential Adjacency Matrix)
2. Apply clustering. (Spectral Clustering)
3. Eliminate some circles found by the clustering algorithm if their density is smaller than a threshold
4. Add more people to the circles if they have number of friends over a threshold given in that circle
6
![Page 7: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/7.jpg)
Data Set• High quality hand-labeled data from major social networking sites (Facebook,
Google+, Twitter)
• All data are available on snap.stanford.edu/data/
7
# of ego-networks # of social circles # of nodes # of edges
Twitter 1000 4,869 81,306 1,768,149
![Page 8: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/8.jpg)
Evaluation MetricResults are evaluated using the edit distance between the predicted social circles and the ground truth. Each of the following operations costs one edit:
• Add a user to an existing circle
• Create a circle with one user
• Remove a user from a circle
• Delete a circle with one user
8
![Page 9: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/9.jpg)
Evaluation Metric: ExampleYour solution contains circles with user ID's => circle1: 3 1 2 circle2: 2 3 circle3: 5 4 6
And the correct solution is => circleA: 4 5 circleB: 1 3 2 4
Then the minimum number of required edits would be following 4 edits:
• Add user 4 to circle1
• Delete user 2 from circle2
• Delete user 3 from circle2
• Delete user 6 from circle3
9
![Page 10: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/10.jpg)
Results• Found the implementation for the “Learning to discover social circles in ego
networks” which is published in Neural Information Processing Systems (NIPS) on http://cseweb.ucsd.edu/~jmcauley/
• Our implementation outperforms their implementation based on Kaggle’sevaluation metric mentioned in the previous slide.
10
![Page 11: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/11.jpg)
Results
11
![Page 12: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/12.jpg)
Sample Graph• Colors indicate social circles obtained
from ground truth
• Pink nodes do not belong to anycircle.
12
![Page 13: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/13.jpg)
Sample Graph• Colors indicate social circles obtained
from our clustering technique
13
![Page 14: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/14.jpg)
References• Java, X. Song, T. Finin and B. Tseng, "Why we twitter", Proceedings of the 9th
WebKDD and 1st SNA-KDD 2007 workshop on Web mining and social network analysis - WebKDD/SNA-KDD '07, 2007.
• Julian Mcauley and Jure Leskovec. 2014. Discovering social circles in ego networks. ACM Trans. Knowl. Discov. Data 8, 1, Article 4 (February 2014), 28 pages. DOI=http://dx.doi.org/10.1145/2556612
• "Soon to Launch: Lists | Twitter Blogs", Blog.twitter.com, 2009. [Online]. Available: https://blog.twitter.com/2009/soon-to-launch-lists. [Accessed: 28- Mar-2016].
14
![Page 15: Towards Detecting Social Circles on Twittercanf/CS533/hwSpring2016/projects/final... · • Create a circle with one user • Remove a user from a circle • Delete a circle with](https://reader034.vdocuments.net/reader034/viewer/2022042318/5f071f747e708231d41b6b4f/html5/thumbnails/15.jpg)
Thanks for listening