Abstract: Collaborative filtering is probably the most familiar and most widely
implemented recommendation algorithm. However, traditional collaborative
filtering methods focus only on rating data to generate recommendation; they
do not consider useful information like item genre and evaluation time, which affect the quality of
the system's recommendation seriously. In similarity computation,
traditional algorithms use all items; they do not introduce genre component
in correlation between user and item. Furthermore, they do not consider the
influence of time on user's interests; giving the same treatment to user's
score at different time. To address this issue, a new item-based
collaborative filtering algorithm is proposed to exploit genre information in each item and
reflect dynamic changes over time of user's preferences. The proposed
algorithm endows each score with a weight function which keeps user's
recent, long and periodic interest, and attenuate user's old short interest.
Experimental results from Movielens data set show that the new algorithm
outperforms the traditional item-based collaborative filtering algorithms.
Keywords: Recommender system, collaborative filtering, user dynamic preferences, item-based filtering
