Information Technology Reference
In-Depth Information
Fig. 3.1
User attribute co-occurrence statistics onGoogle
+
.
a
Age versusmarriage status,
b
Gender
versus occupation
We call these personal information
user attributes
. Inferring such user attributes can
benefit many applications in user profiling, information retrieval, personalization,
and recommendation.
Inmost online social networks (e.g., Facebook, Twitter, LinkedIn, andSinaWeibo),
such user attributes are not always available. First, users are likely to provide the easy-
to-fill basic information such as name, gender, but seldom trouble to introduce their
interests and other detailed information. Second, due to the privacy issues, most
social network sites limit the access to some personal information. Based on our
preliminary statistics on the collected Google
dataset consisting of 19,624 popular
users,
3
nearly 90% of the user gender information is provided, while only 12.36% of
user birthday and 22.48% of user relationship are obtainable. User model distributes
in various aspects, which can be exploited toward corresponding services [
6
,
7
,
22
,
27
-
29
,
38
,
39
]. This demonstrates the potential of user modeling from social multi-
media activities.
Most existing work has treated user attributes separately and studied user attribute
prediction independently. Actually, different types of user attributes have significant
dependency relations, e.g., it is not very likely that a 18-year-old man is married and
his occupation is most probably student. Figure
3.1
shows our dependency analysis
results among user attributes of age and marriage status, gender, and occupation. The
statistics are derived based on a total of 105 million Google
+
4
users. From Fig.
3.1
a,
it is conceived that users under 24 years old are more likely to be single. Figure
3.1
b
illustrates that there exist obvious dependencies between the occupation and gender
attributes. We argue that the user attribute relation is an important characteristic and
will facilitate accurate user attribute inference. Moreover, exploration of attribute
relations will enable novel mining and application scenarios. This motivates us to
infer the relation between attributes as well as the user attributes.
In this chapter, we investigate the problem of relational user attribute inference by
exploiting the rich interaction with social multimedia content. Specially, we study
+
