Research and improvement on topic distillation algorithm in web mining

In this paper, the Hypertext Induced Topic Search algorithm which is the topic search algorithm based on hyperlinks is studied. Aiming at the problem of topic drift existing in this algorithm, the thoughts of the modified algorithms are studied. The methods of textual analysis such as weight conferral of links based on the relative values of ambient texts increase the complexity inevitably. And the methods of utilizing link-structural information such as root-set eigenvector projection method and similarity-based algorithm for topic distillation have large computation spending. So the possible reason for topic drift is basically analyzed, and then an improved algorithm using hyperlinks which is named as base-set shrinking method is presented. This algorithm not only avoids the problem of topic drift, but also consumedly reduces the computation spending, which is demonstrated in experiment and theory.