An emerging trend in classrooms is the use of networked visual argumentation tools that allow students to discuss, debate, and argue with one another in a synchronous fashion about topics presented by a teacher. These tools are aimed at teaching students how to discuss and argue, important skills not often taught in traditional classrooms. But how do teachers support students during these e-discussions, which happen at a rapid pace, with possibly many groups of students working simultaneously? Our approach is to pinpoint and summarize important aspects of the discussions (e.g., Are students staying on topic? Are students making reasoned claims and arguments that respond to the claims and arguments of their peers?) and alert the teachers who are moderating the discussions. The key research question raised in this work: Is it possible to automate the identification of salient contributions and patterns in student e-discussions? We present the systematic approach we have taken, based on artificial intelligence (AI) techniques and empirical evaluation, to grapple with this question. Our approach started with the generation of machine-learned classifiers of individual e-discussion contributions, moved to the creation of machine-learned classifiers of pairs of contributions, and, finally, led to the development of a novel AI-based graph-matching algorithm that classifies arbitrarily sized clusters of contributions. At each of these levels, we have run systematic empirical evaluations of the resultant classifiers using actual classroom data. Our evaluations have uncovered satisfactory or better results for many of the classifiers and have eliminated others. This work contributes to the fields of computer-supported collaborative learning and artificial intelligence in education by introducing sophisticated and empirically evaluated automated analysis techniques that combine structural, textual, and temporal data.