News media has become a prevalent information
spreading platform, where news sites can reprint news from
other sites. To better understand the mechanism of news
propagation, it is necessary to model reprint behavior and
predict whether a news site will reprint a piece of news. Most
existing works in news reprint analysis focus on analyzing the
semantic of news content, little work has been done on
integrating reprint relationship among sites and news content
for reprint prediction from the perspective of sites. The
challenge of improving prediction performance lies in how to
effectively incorporate these two kinds of information to learn a
more comprehensive reprint behavior model. In this paper, we
propose an Integrated Neural Reprint Prediction (INRP) model
that considers both reprint relationship and news content. It
models the reprint relationships as a directed weighted graph
and maps it into a latent space to learn sites representations.
During news content modeling process, sites representations are
embedded as attention guidance to build up more site-specific
content representations. Finally, sites and news representations
are jointly modeled to predict whether a piece of news will be
reprinted by a site. We empirically evaluate the performance of
the proposed model on a real world dataset. Experimental
results show that taking both the reprint relationship and news
content information into consideration could allow us make
more accurate analysis of reprint patterns. The mined patterns
could serve as a feedback channel for both corporations and
management departments.