Incorporating generative conversational agents into collaborative argumentation: Do different customization strategies matter?

Abstract

Collaborative argumentation allows groups to express, criticize, and integrate arguments to achieve the co-construction of collective knowledge. However, students often face challenges when proposing diversified arguments, gathering evidence, and rebutting others reasonably. Incorporating generative conversational agents (GCAs) into collaborative argumentation has been demonstrated to effectively broaden the perspective of the argument and to stimulate the generation of new ideas. For this study, we designed rhetorical argumentation customized strategies (RACS), dialectical argumentation customized strategies (DACS) for collaborative argumentation, and a mixed-strategy (RACS+DACS), and compared their effects on the quality of argumentation mappings and argumentation discourse patterns. A total of 121 first-year postgraduate students were enrolled: 33 for the control group, 33 for the RACS group, 27 for the DACS group, and 28 for the mixed-strategy group. Results found that: (1) Regarding the quality of argumentation mappings, DACS could help students select high-quality evidence and learn the logical skills from evidence reasoning to claims. In addition, the mixed-strategy could help students search for multiple types of evidence to support their position; (2) Regarding argumentation discourse patterns, in the characteristics of the structural dimension, DACS could help students use evidence to support higher-order claims during argumentation. The mixed-strategy could help students use evidence to rebut others’ arguments in group discourses. However, no significant differences were detected among the four groups in the characteristics of the social dimension.

Keywords

• chatgpt
• collaborative argumentation
• generative conversational agents (gcas)
• agent customization
• ordered network analysis (ona)