posted on 2007-01-01, 00:00authored byRakheli Hever, Reuma De Groot, Maarten De Laat, Andreas Harrer, Ulrich Hoppe, Oliver Scheuer, Bruce M. McLaren
Moderation of e-discussions can be facilitated by online feedback promoting awareness and
understanding of the ongoing discussion. Such feedback may be based on indicators, which
combine structural and process-oriented elements (e.g., types of connectors, user actions) with
textual elements (discussion content). In the ARGUNAUT project (IST-2005027728, partially
funded by the EC, started 12/2005) we explore two main directions for generating such indicators,
in the context of a synchronous tool for graphical e-discussion. One direction is the training of
machine-learning classifiers to classify discussion units (shapes and paired-shapes) into predefined
theoretical categories, using structural and process-oriented attributes. The classifiers are
trained with examples categorized by humans, based on content and some contextual cues. A
second direction is the use of a pattern matching tool in conjunction with e-discussion XML log
files to generate "rules" that find "patterns" combining user actions (e.g., create shape, delete link)
and structural elements with content keywords.