Information Technology Reference
In-Depth Information
a residential news source. Mainstream news sources may not cover their local events
at all. Simultaneously, users may read sports-related news rather from mainstream
news sources as they can afford journalists to travel to these events. Hence, users
require services which online news portals provide in contrast to their analogous
counterparts.
As a consequence thereof, users increasingly face the information overload
problem. Recommender systems have established as the suited means to overcome
information overload. They filter available items thus reduce the decision problem
significantly. Users avoid to browse large sets of items. Instead, recommender sys-
tems provide a small fraction of items which they deem most relevant to the user at
hand. Research has focused on recommender systems in terms of preference elici-
tation methods [
55
]. In the context of news, users have rather preferences for latent
concepts than actual items. Recommendations of products such as movies, songs,
or books differ from news article suggestions in this aspect. In the following, we
present a recommendation method that allows dealing with requirements inherent
to news recommendations. These requirements include dynamic item collections,
incomplete user profiles, and differences between individual news portals.
Dynamic item collections refer to the rates at which items either enter or exit the
systems. Editors add novel news items as they emerge to provide readers with infor-
mation about recent events. On the other hand, news articles decrease in relevance
over time as more and more users become aware of them. News collections exhibit
much higher addition/deletion rates compared to collections of movies or songs.
Users may want to reconsume their favorite movies or songs. Contrarily, readers will
seldom read old news articles again.
Recommender systems' quality depends on how well their models reflect user
preferences. Typically, system operators require users to create explicit profiles by
design. Thus, they are able to feed preference directly linked to a specific user.
Contrarily, news portals do rarely require explicit profiles to be created. Supposedly,
readers are unwilling to spend time creating profiles. Privacy concerns represent
another reason keeping users from providing their personal information. News portal
operators tend to identify their users with session identifiers. However carefully they
monitor session identifiers, user profiles may contain errors. We mention three kinds
of such errors. First, readers may use several devices to consume news items. For
instance, they may read news on their tablets as well as their desktop computers.
News portal operators will struggle as they seek to merge these profiles based on
session keys. Second, readers may share their computers with other. For instance, a
couple which lives together might use the same computer for browsing news. Thus, a
profile emerges which captures not one but two preferences. Third, users may block
the session monitoring due to privacy concerns. Thus, the system operators monitor
various users which they cannot differentiate.
Having spent time and resources to build a user profile, users expect to benefit
of adequate recommendations. Conversely, users may consider continuing using the
system and not abort. On the other hand, news readers behave differently. Users
may choose to frequent several news portals. Consequently, users' profiles scatter
over various domains. Incomplete profiles impede creating suggestions. The less
Search WWH ::
Custom Search