Information Technology Reference
In-Depth Information
Fig. 3.7 Examples of attribute-based user retrieval for each type of query by relational LSVM
model
except the fifth user in the second row and the third user in the third row. However,
we note that the incorrect users also match partial queries, e.g., Larry Page in the
third row failed in “elder” but correct with “IT person” and “positive”.
3.7 Discussions
Relational user attribute inference . Inferring user attributes from interaction with
multimedia information is an important and interesting problem. We systematically
study how to infer different user attributes from online multimedia interaction such
as profiles, posted texts and images. Our experimental results on Google
+
set have
justified that user attributes are predictable from their generated online multimedia
content, which is consistent with the conclusion in previous work. In this work,
we highlight the importance of relations in attribute inference. The results in user
attribute inference by our proposed relational LSVMmodel and attribute-based user
retrieval have demonstrated the effectiveness of modeling the attribute relations. It is
significantly important to emphasize that our proposed model for relational attribute
inference is effective for general users with online generated content. We conduct the
study of relational user attribute inference on the Google
user dataset. The users
in this dataset are popular users with a number of followers. A fraction of users are
celebrity users. The only reason of using such users is that the profiles of these users
can be largely accessible from other social media platforms (Facebook, Wikipedia,
Twitter, etc.), which can help annotators build the groundtruth for evaluation. The
+
 
Search WWH ::




Custom Search