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.