Abstract:  Recommender systems analyze patterns of user interest in items or products to provide personalized recommendations for items that are likely to be of interest to the user. Recommender systems are very important for
e ?commerce. One of the most promising recommendation technologies is collaborative filtering. However, the existing collaborative filtering methods do not consider the drifting of the user- s interests. For this reason, the
systems may recommend unsatisfactory items when the user- s interests changed. In order to produce high quality of recommendation, a novel collaborative filtering recommendation algorithm is proposed in this paper, which can study
and trace the user- s interests through studying Ebbinghaus forgetting curve. In order to adapt and trace the drifting of user-s interests, this paper proposes a collaborative filtering method
based on the forgetting curve. Thus, we divide the user- s interests into the long-term interests and the short-terminterests. For the short-term interest, because human memory is limited, if the item cannot be revisited very soon,
it will soon be forgotten. For the long -term interests, if the item is not revisited for a long time for some reason such as the changeable environment, the item will be gradually forgotten. The short term interests are changeable with
the special items such as electrical products, or the special time such as seasons. Otherwise, the long?term interests are stable. T herefore, we divide a period of visiting time into several chronological sub-time periods, each sub -time
period called a time -window. The interval between the two time -windows is different. The speed of forgetting for the items in the time window is supposed to be the same. It is reasonable that we suppose the change curve of the
size of time-windows is similar to the forgetting curve. T hat is, the time -window is more narrow if it more near to now! ( the current time), and the time-window is more broad if it more far away from now! ( the current time).
Furthermore, we find a special power function curve is much more fit to the forgetting curve by the mathematical
graphing calculator software ZGrapher. The goal of a collaborative filtering algorithm is to suggest new items or to predict the utility of a certain item
for a particular user based on the user-s previous likings and the opinions of other like -minded users. Therefore, the measure of similarity between items is very important to heighten the quality of recommendations. The attribute
similarity between the items is used to define the item-similarity as well as the similarity between the rating values used. Then, combing the weight function based on the time -window and the weight function based on the item-similarity, the interest degree of the user to item is defined. The concept of the neighbors of an item is also used here, and a collaborative filtering recommendation algorithm based on the forgetting curve is proposed in this paper, which is suit to the drifting of the user-s interests.
In order to show that the new method is effective, some experiments have done. First, a sequence of experiments are done to decide the coefficients used in the paper. Then, the comparative experimental results show
that the proposed method improves the precision of the recommendation.