This document summarizes two presentations about community detection in social media networks. The first presentation discusses using edge content, like image tags, to help identify communities in networks. The second focuses on leveraging interaction intensities on Twitter to detect communities that form around certain events over time. Both aim to improve on traditional methods that only consider network structure.

1. * Community Detection with Edge Content in Social Media Networks
* Community Detection in Social Media by Leveraging Interactions and Intensities
Presented By: Mojtaba Rezaei & Reza Habibi Kerahroudi
University Of Tehran
Networked Systems Engineering
Community Detection in Social Media
Graph Algorithms Course Wikipedia
T w i t t e r
YouTube
Facebook

2. Community Detection with
Edge Content in Social Media Networks
Presented By: Mojtaba Rezaei
University Of Tehran
Networked Systems Engineering

3. Introduction
 Most community detection algorithms use the links between
the nodes in order to determine the dense regions in the
graph.
in many recent applications, edge content is available in order to provide better supervision to the community detection process

4. Introduction (cont.)
 An important problem in the area of social media is that of
community detection.
In the problem of community detection, the goal is to partition the network into dense regions of the graph.

5. Introduction (cont.)
a lot of rich information is encoded in the content of
the interactions among the actors in the network.
 E-mail networks
we will see that edge content provides a number of unique distinguishing characteristics of the communities which cannot be
modeled by node content.

6. Illustration of a social media network
The nodes represent users while the edges represent the favored images shared by the users

7. Introduction (cont.)
 Edge-based content is much more challenging, because
the different interests of the same actor node may be
reflected in different edges.
We will show that such an approach provides unique insights which are not possible with the use
of pure link-based or content-based methods.

8. Community Detection With Edge Content
 most community detection methods are focused on
partitioning the nodes based on linkage, and we are
interested in partitioning the edges based on both
linkage and content.

9. Community Detection With Edge Content (cont.)
 when there are no links, the problem defaults to the
pure content-based clustering problem.

10. Data Sets
 Enron Email Data Set
 200, 399 messages belonging to 158 members of senior management
 Flickr Social Network Data Set
 15 popular Flickr user groups, including “family”, “auto”, “concerts”, “pet portraits”, “kids and nature”, “street”,
“art”,“wide party,” “folk music“ , "magic city”, “party favors”, "British politics”, “youth basketball”, “fast food",
"fancy dress party” and “great sky.”
 This social media network has 4, 703 users in 15 groups

11. Community Detection in Social Media
by Leveraging Interactions and Intensities
Presented By: Reza Habibi Kerahroudi
University Of Tehran
Networked Systems Engineering

12. Introduction
 User interaction networks capture users’ associations derived from their
activities in social media such as: commenting on others’ posts, replying
to comments, referencing other users, etc.
 Communities can be generally defined as groups of users that are
"closely-knit”, in the sense that a group’s interconnections are more
dense compared to connections with the rest of the network.

13. Introduction (cont.)
 Our focus is on revealing the types of communities
generated with respect to certain events by
analyzing them in the dimensions of size, topic
diversity and time span.

14.  VERTEX STRUCTURE
 STRUCTURAL SIMILARITY
 ε – NEIGHBORHOOD
 CORE VERTEX
 DIRECT STRUCTURE REACHABILITY
 STRUCTURE REACHABILITY
 STRUCTURE CONNECTIVITY
 STRUCTURE-CONNECTED CLUSTER
 CLUSTERING
 HUB
 OUTLIER
SCAN algorithm

15. Getting from SCAN to WSCAN
 SCAN discovers cohesive network subclusters based on
parameters μ and , which control the minimum
community’s size and the minimum structural similarity
between two community’s nodes, respectively.
 To adapt SCAN for weighted interaction networks we
propose weighted structure reachability for (μ,
)-cores’ detection.

16. Real-World Networks
 For experimentation we have generated a network
based on Twitter user interactions, (i.e. mentions,
replies, retweets), extracted from data collected via
the Twitter Streaming API with topic-related
keywords.
Our selected topic refers to the official Euro group meetings (of Euro zone's finance ministers)

17. Real-World Networks
 Our EUROGROUP dataset
 (covering 8 meetings from 13/06/12 to 30/11/12) acts
as an exemplary case study of a series of events held
at different time instances, having the same
participants with a common generic context (i.e. the
Euro zone's monetary issues), but different focus
(depending on the agenda). The dataset spans 227
days and comprises: 29529 tweets, 10305 interactions
and 3015 different users.

18. EUROGROUP meetings, tweets, and communities

19. Classification of the most significant topics
based on interest intensity and diffusion