News Recommendation in Real-Time - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 150

information is available about the user at hand, the harder it becomes to select relevant
readings.
A set of challenges arises to news recommender systems based on the specific
characteristics of news. What news item reflects a certain latent interest best? We
discuss strategies to deal with the dynamics of news. How to link interactions to users
profiles split over a variety of news portals? We present ways to construct user profiles
representing preferences that allow to provide relevant suggested readings. How to
handle the velocity, veracity, variety, and volume of large streams of interactions of
popular news portals? We elaborate on techniques to cope with big data requirements
in the context of news recommendation.
This chapter is structured as follows. Section 6.2 introduces previous research on
news article recommendations. Subsequently, we present specifics of our use case
in Sect. 6.3 . These characteristics include technicalities and requirements as well as
system particularities. In Sect. 6.4 , we show results of observing how users consume
news online. We cover essential aspects including sparsity, popularity bias, as well
as contextual factors. Section 6.5 illustrates recommendation algorithms which have
been applied to a variety of recommendation problems. We discuss how individual
methods suit news recommendation. Likewise, we highlight aspects impeding the
application of certain methods. Section 6.6 details design choices faced as we seek to
evaluate the performance of recommendation algorithms. Finally, we conclude and
give an outlook to future research directions in Sect. 6.7 .
6.2 Related Work
News portals have evidentially changed the way we consume news. This section
presents related research dedicated to support users consuming news. Billsus and
Pazzani [ 8 ] refer to four types of systems which have developed to support us con-
suming news. First, they introduce systemswhich enable personalized access to news.
The personalization manifests as news portals present varying news items depend-
ing on individual preferences. News recommender systems rank among this kind of
systems. Second, Billsus and Pazzani list adaptive news navigation systems. These
systems control how news stories link together. Ideally, they reduce users' efforts to
turn back to home pages before continuing reading. Third, Billsus and Pazzani men-
tion contextualized news systems. These systems present their contents depending on
users' current contexts. Context includes aspects such as location, time, and current
interests. Finally, they introduce news aggregation systems. These systems take col-
lections of news articles and automatically extract the very essential information. We
focus particularly on systems recommending news articles. These systems became
invaluable supportive to online news readers as more and more news became avail-
able. This growth induced an information overload problem. Recommender systems
represent a specific kind of information filter. Information retrieval systems filter
information contained in document collections having received a query [ 39 ]. In con-
trast, recommender systems attempt to learn preferences from previous interactions