Agent-based Content Management System Hidekazu Kubota Graduate School of Informatics, Kyoto University Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501 Japan kubota@ii.ist.i.kyoto-u.ac.jp Jaewon Hur Graduate School of Information Science and Technology, The University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, Japan jwhur@kc.t.u-tokyo.ac.jp Toyoaki Nishida Graduate School of Informatics, Kyoto University Yoshida-Honmachi, Sakyo-ku, Kyoto 606-8501 Japan nishida@i.kyoto-u.ac.jp Abstract This paper describes the management of huge heterogeneous content using the agent-based content management system. The knowledge card that can wrap up heterogeneous content is proposed. The knowledge channel that is a model of content management system is also proposed and implemented by using conversational agent and dynamic program table. An empirical experiment conducted in three communities implies that the knowledge channel allows the management of huge heterogeneous content. Keywords: Personal content management system, Conversational agent 1 Introduction The purpose of this paper is computational management for personal content creation. The per- sonal content means here is a set of essays and papers that are created casually. The management of personal content is essential work for human intellectual life at present, for instance many peo- ple publish their personal journals on web pages. However, it is troublesome to manage huge personal content because the content can include heterogeneous parts such as text, photos, music and movieclips, its constituent parts are different sizes, and moreover boundaries of its topics are not clear. This paper is intended as a computational management of the unstructured heterogeneous content mentioned above. The essential idea is a knowledge card that is a well done piece of heterogeneous content. A knowledge card wraps up a few sentences of text and an image (or a movieclip) that represent one topic, so intra-card structure is too heterogeneous for a computer to understand, while inter-cards boundaries are very clear. Such distinction between cards is important clues for a computer to understand the whole structure of the content. In addition to that, these cards are easy to rearrange because their grain size is generally regular. To manage knowledge card formed content, a knowledge channel model has been proposed. This is a model of an agent-based content management system that integrates conversational process and editorial process for content management. These processes complementarily grow content. The conversational process manages time series of content through conversation, while the editorial process manages spatial relations among content. To support the conversational process, we have developed conversational agent that can talk with people about content, and