A EXTRACTION AND PREDICTION METHOD OF LATENT USER'S PREFERENCE

摘要

PURPOSE: A method for extracting and estimating a latent user attractiveness is provided to enhance an accuracy of an estimation and reduce a performing time through a noise removal and a meaning dimension reduction of an input matrix when a latent user attractiveness is extracted and estimated using a graded evaluation matrix inputted by users. CONSTITUTION: A latent user attractiveness is extracted using an SVD(Singular Value Decomposition) technique, and a noise removal and a meaning dimension reduction of an input matrix are performed. Users are mapped to a meaning dimension space, and the users are expressed as one vector in a reduced dimension space. A distance for indicating a similarity among each user is calculated using the vector for expressing the users. Users who have a low similarity, respectively, are filtered using a predetermined sized critical value for selecting users similar to each user. A weight value is calculated for supplying much influence to similar users. Users' attractiveness with respect to items is calculated using an average of evaluation values including weight values of all similar users.