Information Technology Reference
In-Depth Information
Another research line is on analyzing the diffusion patterns between social media
networks. Leskovec et al. conducted pioneer studies on the cite and influence corre-
lation within different social blogging networks [
17
]. Large-scale blog linking and
propagation graphs are constructed based on the observations. Around social blog-
ging networks, extensive work has examined the diffusion and evolution patterns
between news media [
11
], multimedia sharing networks [
6
], and Social Networking
Sites (SNS) [
2
]. Among these, in [
16
], cross-network diffusion is discussed together
with macro user activities, where the influence of user activity on information spread
in Twitter and Digg is examined. Recently, Kim et al. presented an interesting work
by measuring three metrics of activity, reactivity, and heterogeneity to understand
the diffusion patterns of social media networks on the same trending topics [
15
].
Actually, the above cross-network diffusion analysis can be viewed as one solution
to social multimedia data organization, by tracking the spread of news, events, and
topics. For social multimedia data association, the current macro-level solutions are
mainly based on semantic correlations. Tsagkias et al. [
26
] introduced the problem
of news retrieval in different social media networks. Three methods are proposed to
represent the query: internal news structural metadata, external social media utter-
ances in respective networks, and the selected keywords. In [
3
], with focus on social
events, the authors proposed to identify UGC across Twitter, YouTube, and Flickr.
Event descriptions and extracted frequent terms are utilized to construct the query
features for event retrieval. Recently, Suman et al. have conducted a series of work
to associate between social stream and video sharing networks. The basic idea is
to align the derived semantic video topics by integrating social stream knowledge.
Successful applications are reported on video query suggestion [
22
], video recom-
mendation [
23
], and video popularity prediction [
24
]. However, as discussed in the
introduction and demonstrated from the experiments, semantic-based social multi-
media data association fails in some complex cases. Given the social multimedia
data characteristics of “from user, for user”, taking the overlapped user into consid-
eration will provide insight at the micro-level and certainly provides a new clue to
heterogeneous social multimedia data organization and association.
5.2.2 Micro Cross-Network Analysis
With ubiquitous social media services, increasing people are voluntarily willing to
disclose their user accounts online, by filling in SNS registration information (such as
Facebook, Google+) or maintaining their social media aggregation profiles (such as
About.me, Friendfeed). Moreover, many companies share identical account among
their different social media networks, such as Google account for YouTube and
Google+, Tencent account for QQ, Weixin, and Tencent's Microblog. The accessi-
bility of overlapped user accounts opens up the possibility to conduct cross-network
analysis at micro-level and facilitating personalized social media services.
The first research line is user linkage mining, which aims to automatically iden-
tify the user accounts on different social media networks that correspond to the same
