Collaborative Recommendation Method Reflecting Temporal Trends

Abstract

Automated collaborative recommendation has been a popular method that predicts a user's affinity for each item indirectly in order to complement content-based recommendation. Especially, this method has been widely used for a variety of web services due to its well-formulated mathematical background and fair performance. However, it cannot effectively reflect temporal trend of popularity on each item so that it often fails to give useful recommendation in practice. In many cases, item popularity depends on "time" so that it may be differently assessed by the users as time goes, because trendy or hot item is likely to be popular first but not any more later. In this paper, we propose a new collaborative recommendation method reflecting temporal trends (called "temporal trend prediction"). Our method predicts temporal trend using linear regression analysis and combines temporal trend into conventional collaborative recommendation effectively. We also present some experimental results in comparison with conventional collaborative method. ©2013 International Information Institute.