Introduction - User-centric Social Multimedia Computing - page 7

1.4 Overview of the Topic
The topic comprises totally six chapters. Chapter 1 is this introduction. From
Chaps. 2 - 4 , we introduce our work on the three basic tasks of user-centric social
multimedia computing, respectively. Extensions under the cross-network circum-
stances are elaborated in Chap. 5 . Chapter 6 concludes the topic by summarizing the
major points and identifying the future works.
Chapter 2 : User-perceptive multimedia content analysis. While user serves as one
of themost fundamental elements in social multimedia, users' explicit interactions are
largely ignored in current social multimedia content analysis solutions. We propose
to model user information in the multimedia generation and consuming processes,
with the goal to better understand the observed social multimedia data and apply
the refined results into social multimedia applications. Specifically, a ranking based
multicorrelation factor analysis method is presented to jointly model the user, image,
and tag factors. The observed user-image-tag ternary relationships are represented as
three-order tensors. The improved user, image, and tag factors are extracted by regu-
larized tensor decomposition. Image tag refinement is then performed by exploiting
the associations between image and tag factors. So far as we know, this is the first
work to consider the user factor in social image tag analysis problems.
Chapter 3 : User modeling on social multimedia activity. Nowadays, more and
more people are engaged in social multimedia sharing websites to create profiles and
post messages. Such social multimedia activities indicate users' intents and prefer-
ences and can be utilized to infer multiple user attributes such as age, gender, and
personal interest. In this chapter, we address two issues in user demographic attribute
inference: user data sparsity and user attribution relation. For the first problem, mul-
timodal user activity as well as profile data are exploited for integrated attribute
inference. User-specific topic modeling is conducted on the expanded user collec-
tion to learn user preferences. For the second one, we exploit the relations between
user attributes via a relational latent model. The derived attribute relation is utilized
for accurate user attribute inference, as well as applied to structural attribute-based
user retrieval.
Chapter 4 : Personalized multimedia search. In social media, social relationship
has been recognized as significantly impacting social activities and user preferences,
which plays important role in personalized multimedia search. In this chapter, we
argue that social relationships between users are not simply binary 0/1, but topic
sensitive. From the perspective of generative models, we investigate into the pair-
wise topic-sensitive social influence by modeling the user annotation and contact
activities, and simultaneously obtain inter-user as well as user topic distributions.
After that, we apply the derived social relationship into personalized image search
problem. Under themechanismof riskminimization, we present a general framework
for personalized image search for the first time. It is capable of jointly modeling the
output of multimedia content analysis and user modeling.
As supplements to the three basic tasks and inspiration to future work, we also
introduce the very recent trends in cross-network social multimedia computing.