Abstract: A system of software components is constructed which includes multiple recommendation agents, each having: a list of recommended items for each user; a numerical predicted rating representing the predicted extent of each user's preference for the item recommended and numerical value information representing confidence; and a user profile for holding basic information on each user and influential power information of each recommendation agent with respect to each user. A recommendation manager is constructed to: select one or more recommendation agents using the basic user information and the influential power information; generate a final list of recommended items; and adjust the influential power of each recommendation agent stored in said user agent according to the predicted score and the belief provided by each recommendation agent, and the extent of a user's actual preference for the recommendation.