Collaborative Problem Solving (CPS) is a critical skill in modern education, requiring students to engage in interactive collaboration to construct shared solutions. Traditional CPS assessment relies on human-coded frameworks, which are labour-intensive and challenging to scale. Recent advances in Natural Language Processing (NLP) enable automated CPS analysis, but existing models predominantly use single-label classification, oversimplifying CPS behaviours, and fail to fully incorporate conversational context. To address these limitations, we propose a context-aware, multi-label classification framework leveraging a sliding window mechanism and pre-trained language models to enhance CPS dialogue analysis. Our approach integrates local utterance semantics with broader conversational dependencies through structured feature fusion strategies. Experimental results on a real-world classroom dataset show that incorporating conversational context improves classification accuracy, with max-pooling and multiplication-based fusion achieving the best performance. These findings highlight the importance of contextual modelling in CPS assessment and provide a foundation for more scalable, automated educational analytics.