Modeling Collaboration in Online Conversations Using Time Series Analysis and Dialogism

摘要

Computer Supported Collaborative Learning (CSCL) environments are frequently employed in various educational scenarios. At the same time, learning analytics tools are frequently used to quantify active learners' participation, collaboration, and evolution over time in CSCL environments. The aim of this paper is to introduce a novel method to cluster utterances from online conversations into zones based on different levels of collaboration. This method depends on time series analyses, grounded in dialogism and focuses on the underlying semantic chains that are encountered in adjacent contributions. Our approach uses Cross-Reference Patterns (CRP) applied on the convergence function between two utterances which captures their semantic relatedness. Two methods for clustering utterances into convergence regions are tested: clustering by uniformity and hierarchical clustering. We found that hierarchical clustering surpasses clustering by uniformity by considering only highly related contributions and providing a more straightforward unification mechanism. A validation analysis on the hierarchical clustering model was performed on a corpus of 10 chat conversation reporting variance in terms of F1 scores. The model and encountered problems are discussed in detail.