This dissertation addresses four issues of pivotal importance in realizing the promises of adaptive argumentation learning systems: (1) User interface: How can argumentation user interfaces be designed to effectively structure and support problem solving, peer interaction, and learning? (2) Software architecture: How can software architectures of adaptive argumentation learning systems be designed to be employable across different argumentation domains and application scenarios in a flexible and cost-effective manner? (3) Diagnostics: How can user behavior be analyzed, automatically and accurately, to drive automated adaptations and help generation? (4) Adaptation: How can strategies for automated adaptation and support be designed to promote problem solving, peer interaction, and learning in an optimal fashion?Regarding issue (1), this dissertation investigates argument diagrams and structured discussion interfaces, two areas of focal interest in argumentation learning research during the past decades. The foundation for such structuring approaches is given by theories of learning and teaching with knowledge representations (theory of representational guidance) and collaboration scripts (script theory of guidance in computer-supported collaborative learning). This dissertation brings these two strands of research together and presents a computer-based learning environment that combines both approaches to support students in conducting high-quality discussions of controversial texts. An empirical study confirms that this combined approach has positive impact on the quality of discussions, thus, underpins the theoretical basis of the approach.Regarding issue (2), this dissertation presents a software framework for enhancing argumentation systems with adaptive support mechanisms. Adaptive support functionality of past argumentation systems has been tailored to particular domains and application scenarios. A novel software framework is presented that abstracts from the specific demands of different domains and application scenarios to provide a more general approach. The approach comprises an extensive configuration subsystem that allows the flexible definition of intelligent software agents, that is, software components able to reason and act autonomously to help students engage in fruitful learning activities. A graphical authoring tool has been conceptualized and implemented to simplify the process of defining and administering software agents beyond what has been achieved with the provided framework system. Among other things, the authoring tool allows, for the first time, specifying relevant patterns in argument diagrams using a graphical language. Empirical results indicate the high potential of the authoring approach but also challenges for future research. Regarding issue (3), the dissertation investigates two alternative approaches to automatically analyzing argumentation learning activities: the knowledge-driven and the data-driven analysis method. The knowledge-driven approach utilizes a pattern search component to identify relevant structures in argument diagrams based on declarative pattern specifications. The capabilities and appropriateness of this approach are demonstrated through three exemplary applications, for which pedagogically relevant patterns have been defined and implemented within the component. The approach proves particularly useful for patterns of limited complexity in scenarios with sufficient expert knowledge available. The data-driven approach is based on machine learning techniques, which have been employed to induce computational classifiers for important aspects of graphical online discussions, such as off-topic contributions, reasoned claims, and question-answer interactions. Validation results indicate that this approach can be realistically used even for complex classification tasks involving natural language. This research constitutes the first investigation on the use of machine learning techniques to analyze diagram-based educational discussions. The dissertation concludes with discussing the four addressed research challenges in the broader context of existing theories and empirical results. The pros and cons of different options in the design of argumentation learning systems are juxtaposed; areas for future research are identified. This final part of the dissertation gives researchers and practitioners a synopsis of the current state of the art in the design of argumentation learning systems and its theoretical and empirical underpinning. Special attention is paid to issue (4), with an in-depth discussion of existing adaptation approaches and corresponding empirical results.
展开▼
