Research On Personalized Recommendation Algorithm Based On Natural Forgetting

With the rapid development of Internet and information technology, various types of data in the web grow quickly. The expansion of information resources causes the emergence of the so-called"information overload"and"information maze"phenomena. These kinds of phenomena make people difficult to find valuable information among the huge amount of information resources. Thus in order to make people more accurately and effectively to obtain useful information, recommender systems that could initiatively provide information for users come into being. Recommender system is based on users(?) browse histories, the preferences to items, or the relevance between items to build users(?) interest model. And then according to this model it could recommend items that may be interesting to users. For example, a movie web could base on the user(?)s view histories and the scores to films, and obtain the preferences information on films, and as a result recommend films to this user that may be interested in.Researchers have found that personalized recommendation issue exist the phenomenon of user interest drifting. That(?)s to say, people(?)s interests are not static, but will vary over time, while the traditional recommender methods do not take this into account. Therefore, based on the traditional collaborative filtering and memory and forgetting theory, we propose an algorithm that is based on natural forgetting to track the change rules of users(?) interests. The improved algorithm not only could reflect the change of interests over time, but also could recommend items according to the amount of forgotten.The main idea of the proposed users(?) drifting interest algorithm is to find appropriate time weight for items, namely, the ratings recently given by users could provide more impact on the current item. Intuitively, we think that the recent ratings are the user(?)s current main interest points; the recent ratings have a bigger time weight than others. The main work of this paper includes the following aspects: first, the related technologies of collaborative filtering are described; second, we introduce the evaluation metrics of recommender system; next, we review the modeling methods of users(?) drifting interests; finally, based on natural forgetting, a personalized recommender algorithm is proposed. This paper considering users(?) past ratings adopts a kind of time decay function that simulate natural forgetting law, track the change rule of users(?) interests, and then recommend items that is most likely to be interesting to users. Experimental results show that the proposed personalized recommender method, based on natural forgetting, could more accurately recommend items to users, and therefore improve the performance of the recommender systems.