Information Technology Reference
In-Depth Information
4.6 Discussions
4.6.1 Understanding Influence in Different Fields
Influence has different explanations in different fields. In psychology, influence mea-
sures human dynamics for persuasion and stress. For example, howmuch a speaker's
confidence is influenced by the nodding action from the audiences [
8
]. In social sci-
ence, influence analysis involves with information flow and social network evolu-
tion. It explains the homophily—birds of feather phenomenon [
10
]. Therefore, we
can roughly conclude that influence in psychology is usually quantitative, in social
science is basically qualitative. In this chapter, we explain influence in multimedia
applications as affecting someone else on behavior, preference or decisions. Com-
bined with the toy example explained in Fig.
4.1
, we emphasize the topic-sensitive
characteristic for influence in multimedia applications.
Note that we are not claiming that influence in social science or psychology is not
topic-sensitive. The idea is that different areas have different focuses, and we high-
light the topic-sensitive feature from the perspective of multimedia applications. For
example, social science discusses the reasons behind correlation, namely whether
social relation come first or affinity action comes first. While what is useful for mul-
timedia applications like search or recommendation is whether there exist correlation
between social relations and affinity actions.
4.6.2 Methodological Contribution
In view of methodology, the proposed mmTIM influence model can be seen as a
multimodal version of the mTIHmodel presented in [
17
], but with important updates
to fit into the multimedia scenarios: (1) mTIH addresses the problem in text-based
network and did not consider visual information. While in our model, visual and
textual information affects to each other during the sampling and jointly contribute
to the derived influence. Also, the derived topic space spans over both textual and
visual information, making it easy to be used in multimedia applications. (2) mTIH
models influencers and influencees differently, while our model treats influencer the
same with ordinary users and simplify the model inference process. (3) mTIH tracks
document-level citation or follow relation, while we build the influencer set at user-
level and only use basic actions of tagging and annotating for modeling. This is more
intuitive for the user-centric applications.
4.6.3 Potential Extensions
Beyond the examined applications in the experiments, the proposed approach can be
easily extended to novel problems like topic-based social wall organization. In the
