The generic term communityware (M. Chalmers and P. Chitson, 1992) has been proposed to represent systems that typically support the formation, activities, and organization of informal groups on open networks such as the Internet or mobile computing systems. Communityware is intended to provide support for a diverse and amorphous group of people who do not yet have a clearly defined goal. We explain the development of an example communityware system that graphically presents a potential community, i.e., people sharing common interests. The proposed system determines the degree of common interests for each pair of people, and it then locates the users in a plane where the distance between the users reflects the relevance between them. A major challenge for implementing this system is distinguishing an appropriate set of keywords to visualize potential communities. The system extracts keywords from the users' profiles and calculates the degree of relevance among the users using those keywords. We use a weight system for keywords as well as a learning mechanism that determines the desired weight from user feedback. It is shown that agents can speed up this learning process by cooperation.