A New Generalized Similarity-Based Distillation Algorithm

摘要

The procedure of hypertext induced topic search basedon a semantic relation model is analyzed, and the reason for thetopic drift of HITS algorithm was found to prove that Web pagesare projected to a wrong latent semantic basis. A new concept-generalized similarity is introduced and, based on this, a newtopic distillation algorithm GSTDA(generalized similarity basedtopic distillation algorithm) was presented to improve the qualityof topic distillation. GSTDA was applied not only to avoid thetopic drift, but also to explore relative topics to user query. Theexperimental results on 10 queries show that GSTDA reducestopic drift rate by 10% to 58% compared to that of HITS(hypertextinduced topic search) algorithm, and discovers several relativetopics to queries that have multiple meanings.