Information Technology Reference
In-Depth Information
efforts have been undertaken by people and companies to improve their popularity,
growth, and power, the outcomes of which are all expressed as rankings. Conven-
tionally, these rankings are evaluated and ranked by values from statistical data and
attributes of actors such as income, education, personality, and social status.
In the theory of social network analysis (SNA), social networks are used to ana-
lyze the performance and valuation of social actors [13]. Network researchers have
argued that relational and structural embeddedness influence individuals' behavior
and performance, and that a successful person (or company) must therefore empha-
size relation management. Actually, several relations exist in the world with differ-
ent impacts; the actors might be tied together closely in one relational network, but
can differ greatly from one to another in a different relational network. The question
therefore arises:
Relations of what kind are important for entities?
Unfortunately,
the answers of important relations have been decided according to the judgments of
researchers themselves.
To identify the prominence or importance of an individual actor embedded in
a network, centrality measures have been used in social sciences: degree central-
ity, betweenness centrality, and closeness centrality. These measures often engender
distinct results with different perspectives of “actor location”, i.e. local (e.g. degree)
and global (e.g. eigenvector) locations, in a social network [13]. Another question
arises:
What kind of centrality indices are most appropriate for ranking actors?
That
question can be extended as
What kind of structural embeddedness of actors makes
them more powerful?
This chapter presents a description of an attempt to learn the ranking of named
entities from a social network that has been mined from the world-wide web. It
enables us to have a model to rank entities for various purposes: one might wish
to rank entities for search and recommendation, or might want to have the ranking
model for prediction. Given a list of entities, we first extract relations of different
types from the web based on our previous work [4, 8]. Subsequently, we rank the en-
tities on these networks using different network indices. In this chapter, we propose
a systematic algorithm that integrates features generated from networks (designated
as
network-based features
) for each and then use these features to learn and predict
rankings. We conducted experiments related to social networks among researchers
to learn and predict the ranking of researchers' productivity.
The contributions of this study can be summarized as follows. We provide an
example of advanced utilization of a social network mined from the web. The results
illustrate the usefulness of our approach, by which we can understand the important
relations as well as the important structural embeddedness to predict ranking of
entities. The model can be combined with a conventional attribute-based approach.
Results of this study will provide a bridge between relation extraction and rank
learning to facilitate advanced knowledge (web intelligence) acquisition.
The following section presents a description of an overview of the ranking learn-
ing model. Section 3 briefly introduces our previous work for extracting social net-
works from the web. Section 4 describes the proposed ranking learning models
based on extracted social networks. Section 5 explains the experimental settings
and results. Section 6 presents some related works before the chapter concludes.
