There There are two potential problems with the

There are two prominent filtering
methods that contribute in maximizing the user utility factor: Content based filtering/recommender
systems (CF) and collaborative filtering/recommender systems (CBF). One single
filtering technique or the static combination of two techniques does not meet
the challenges of today’s dynamic world. CF, being the prevalent personal
filtering technique, needs a considerable amount of interaction between the
users and the employed system before new items of interest can be suggested.
Furthermore, it is computationally expensive and therefore not suitable for
real-time interaction. Thus, it needs a different approach, until we have
sufficient information about the users, such as CBF which is independent from
the user. The systems should be able to adjust itself depending on the amount
of information available about a particular user.

In our approach, we have designed
MLARS (Multi Layered Architecture for Recommender Systems) framework which allows
making use of the combined strength of the different filtering methods. Moreover, the
framework should be portable to distributed scenarios, such as personal
recommender systems, in which case it allows only partial set of the whole
information space. Such a personal recommender system should provide its users
with recommendations wherever they are and whenever they request it. In this
scenario, each user is carrying their own personal recommender system and the
system should be updatable without the assumption of a centralized
infrastructure. In addition, such a recommender system must run fast and
efficiently on mobile devices which have limited memory and operate with slow
processors.

There are two potential problems
with the traditional recommender systems. One is the scalability, which
is how quickly a recommender system can generate recommendation, and the second
is to upgrade the quality of the recommendation for a customer. Pure CF
recommender systems produces high quality recommendation than those of pure
content-based.  

Moreover, traditional filtering techniques lack
performance due to following issues:

–         
The combination of the two techniques is static

–         
Does not adjust itself when the state of the problem
domain changes 5

–         
Hybrid methods can only combine two techniques in one
single framework

–         
The hybrid methods consume large amounts of memory and
require high processing power & time consuming

–         
First rater problem 2

–         
Sparsity problem 1

 

x

Hi!
I'm Mary!

Would you like to get a custom essay? How about receiving a customized one?

Check it out