Catching Preference Drift With Initiators In Social Network

The change of users’preference over time, called preference drift in this paper,gets it difcult to make accurate recommendation. A recommender system ignoringthe fact always recommends similar items which were loved previously by users whileusers’preference have changed when new items appear in the system. It has beenproven empirically that algorithms handling the concept drift problem which simplytakes advantage of time weight is not appropriate in recommendation systems. A mod-el considering both the static and dynamic preference well is the key to catch users’preference drift. The focus of our work in this paper is catching the users’ preferencedrift in0-1recommender systems in which the rating of a user for an item is1when acertainbehaviorisperformedbytheuserandmissing,otherwise. Manysimilaritymea-surements used frequently in recommender systems are based on the assumption thatusers’ratingsforitemsareintegersfrom1to5anddon’tadapttothesituationsinwhichall the ratings are1. There are many systems which are instances of0-1recommendersystems, so we design several similarity measurements for0-1recommendation sys-tem. Then we put forward an original model which explores the behavior of infuentialpeople, the initiators who initiate trends in social network, when making recommen-dations, it combines the preference information and the trend information. Comparedto traditional collaborative fltering approaches and time weighted approaches, the em-pirical study on lastfm dataset has shown that our model improves the accuracy of therecommendation.