We propose a web page classification method for creating a high quality collection of researchers’ homepages. A method to reduce manual assessment required for assuring given precision/recall using a recall-assured and a precision-assured classifier is presented. Each classifier is built with SVM using textual features obtained from each page and its surrounding pages and tuning parameters. These pages are grouped based on connection types and relative URL hierarchy levels, and independent features are extracted from each group. Experiment results show the proposed features evidently improve classification performance and the manual assessment is significantly reduced.