Internet sellers hope to let the consumers find their commodities easily, but there is a divergence between commodities presentation requirements and the limitation of a web page. In order to pursuit the profit-maximizing, a web page must contain the most welcome commodities. In this study, we proposed a multiple candidate set method to discover those most welcome commodities from users' browsing data. Our approach utilizes frequent dependent relationships among the commodities to partition the unique candidate set, composed of all single frequent item, into multiple sub-candidate sets for computing the frequent browsing patterns, which avoids the problem of searching through a large space of candidate set. Our experiments illustrate the benefits of the proposed method against the single frequent item candidate set.