Local Nonnegative Matrix Factorization for Mining Typical User Session Profile

Abstract

Understanding the evolving user session profile is key to maintaining service performance levels. Clustering techniques have been used to automatically discover typical user profiles from Web access logs. But it is a challenging problem that many clustering algorithms yield poor results because the session vectors are usually high dimensional and sparse. Although standard non-negative matrix factorization (SNMF) can be used in reducing the dimensionality of the session-URL matrix, the clustering results is not precise, because the basis vectors SNMF gets are not orthogonal and usually redundancy. In this paper, we apply local nonnegative matrix factorization (LNMF), which get basis vectors as orthogonal as possible, to reduce the dimensionality of the session-URL matrix. The experiment results show that LNMF performs better than SNMF for mining typical user session profile.

This work was supported in part by the NSFC (60373066, 60303024), National Grand Fundamental Research 973 Program of China (2002CB312000), National Research Foundation for the Doctoral Program of Higher Education of China.]

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.