Conversation Thumbnails for Large-Scale Discussions Martin M. Wattenberg IBM Research One Rogers Street Cambridge, MA 02142 +1 617.693.5650 mwatten@us.ibm.com David R. Millen IBM Research One Rogers Street Cambridge, MA 02142 +1 617.693.7490 David_R_Millen@us.ibm.com ABSTRACT We present a new interface for large-scale online conversations. Our technique, the Conversation Thumbnail, differs from existing discussion interfaces in two respects. First, it employs a focus+context visualization technique that exploits message-level metadata to provide an easily navigable overview of a discussion. Second, it helps reduce conversational redundancy and improve coherence via a fast automatic search mechanism that highlights related messages during message composition. The Conversation Thumbnail Viewer is currently implemented as a Java applet that can be applied to a variety of discussion data sources. Keywords Social Navigation and Visualization, User Interface, Persistent Conversations, Collaboration INTRODUCTION Large discussions are a key part of the internet. Usenet newsgroups, web sites like Slashdot.com, and intranet- based bulletin boards all present conversations that are so huge—often containing hundreds of messages adding up to the size of a full novel—that they are simply too vast to absorb in their entirety. Readers of such large conversations face two critical challenges. Can they identify and navigate to interesting messages? And when composing a message, how can they ensure they are not fragmenting the conversation by duplicating a previous message or ignoring a relevant thread they have not yet read? Many researchers have tackled these problems by providing a visual overview of the conversation. Timeline representations of member activity have been used in the Babble interface [2]. Conversation Maps [8] display social and discourse structure, as well as highlighting key topic words. Loom [1] displays thread structure and emotional content. Smith and Fiore [9] explore interpersonal connections and thread structure. All of these visualizations, however, are demanding for the user. In each case it is difficult to skim rapidly through many messages since individual messages can only be viewed via an additional explicit user action, such as a button click. Moreover they rely heavily on thread structure, which in a very large conversation may not be sufficient to pick out interesting messages. Finally, they do nothing at the time of message composition to encourage or support conversational coherence. THE CONVERSATION THUMBNAIL VIEWER In this paper we introduce Conversation Thumbnails, a prototype discussion interface that provides skimming- friendly navigation, exploits available metadata to help spotlight important sections of the discussion, and uses a special composition-time search mechanism to help maintain conversational coherence. Figure 1 shows a portion of the interface for an illustrative discussion. The interface has three parts: a navigable overview of the entire conversation, a detail window displaying individual messages, and a composition area for new messages. Figure 1. Conversation Thumbnail. Copyright © 2003, IBM. All rights reserved.