Automating SSRL Detection in Asynchronous OCL via LLMs

Socially Shared Regulation of Learning (SSRL) in online collaborative learning (OCL) presents significant challenges due to the complexity of asynchronous interactions and the limitations of conventional content analysis methods. However, existing tools struggle to scale effectively with the heterogeneous nature of peer discussions, which are context-dependent and linguistically nuanced. This study introduces a novel, modular, intelligent coding system that integrates SSRL theory with large language models (LLMs), enabling automated content analysis. Using 1,380 peer-discussion entries from a 16-week undergraduate course, we evaluated three LLMs (ChatGPT-4o, ChatGPT-o1, and DeepSeek-R1) with a three-stage experimental pipeline: Zero-shot, Few-shot, and Few-shot Chain-of-Thought prompting. Results showed that Few-shot Chain-of-Thought prompting improved contextual reasoning, with DeepSeek-R1 achieving 83.3% accuracy in phase identification, leveraging attention-based modeling of implicit linguistic cohesion in collaborative discussions. The system innovatively encodes SSRL's pedagogical constructs into hierarchical prompt templates, allowing for automatic detection of regulatory patterns while mitigating semantic ambiguity. A case study further demonstrates its effectiveness in distinguishing between surface-level consensus and genuine co-construction behaviors. This work pioneers the integration of SSRL frameworks with generative AI, offering a scalable tool for SSRL regulation analysis in educational settings. Future research should focus on lightweight domain adaptation techniques, such as LoRA fine-tuning, and hybrid models combining cognitive diagnostics to enhance interpretability and bridge the gap between AI outputs and expert judgment in content analysis.