Abstract:
Recommender systems facilitates to improve customers experience by providing them, with the suggestions they might need. These systems store user preferences in form of ratings to suggest other users with similar interest the similar items to help them get what they might desire i.e. collaborative recommender systems, or by analyzing the attributes of item
e.g. categories, properties of items to suggest similar items to that one is selected by the
user. These systems are widely incorporated in movies websites, e libraries and e-Commerce
widely incorporate users preference based recommender systems.
However recommender systems are experiencing noise issues due to multiple factors. One
can be due to immature and irregular behavior of users which effects the recommendation
results, and other can be malicious intent with which malicious users can temper the database
of recommender systems through providing false and biased ratings to promote or demote
certain items.
To overcome the issue of natural noise specifically, we will detect inconsistencies in ratings
and then implement correction to improve the predictions to prevent biased recommendations.
In this approach we will only consider rating values that are being provided by users
to the items. In order to do so we will define the tendencies of users and items, ratings deviating
from the tendencies will be characterized as noise so these ratings will be corrected
through prediction technique. The modifications in noisy ratings will depend on the degree
of noise present in the ratings. Fuzzification approach will be used to make flexible classes
instead of rigid classification of users and items. Finally we will compare our results with
the previous technique to show the improvement that our proposal has made.
