Algorithm of shielding noises in information filtering based on distribution of two-dimension similarity relation

This paper investigates the reasons for generation of noises in feedbacks of filtering system by analyzing the knowledge expressions and text structures, and builds a two-dimension similarity (DTS) model of information based on the opposed relation between the noises and user profiles. At the same time, by using the algorithm of AdaBoost based on the LMS (least mean square) classifier, this paper builds a classification curve between the noises and related information according to their distribution in two-dimension similarity space, which helps information filtering system detect and filter the noises in feedback. Experiments validated this algorithm substantially improved the capability of filtering system to rule out the noises.