A Short-term User Interest Model for personalized recommendation

Abstract

Personalized recommendation is a widely used application of Web personalized services which alleviate the burden of information overload by collecting information which meets user's needs. An essential of personalized recommendation is how to describe and obtain user's interests. Most existing approaches try to obtain interests from user's whole process of browsing. However, effective obtainment, storage and organization are challenging issues. In this paper, Short-term User Interest Model (SUIM) is presented to represent user's real-time interests based on his/her recent browsing content and behavior. On one hand, based on Semantic Link Network (SLN), user's real-time interests are semantically represented by Web pages he/she has browsed. The contribution of each page to representation of user's interests is weighted by information entropy based on associated degree from this Web page to other ones. On the other hand, memory capacity and recall probability from psychology are introduced to ensure the small scale and accuracy of SUIM. Experimental results show the validity of SUIM. The proposed method has a brilliant perspective in the applications of Web personalized recommendation.