Enhancing Collaborative Learning Environments: A Multi-Feature Fusion Model for Disruptive Talk Detection

Effective detection and identification of disruptive talk is of significant practical value for improving collaborative learning quality and optimizing the online education environment. However, existing research on disruptive talk detection, which often relies on features from a single dimension, struggles to comprehensively capture the semantic information of talk and learner characteristics, thereby limiting the detection effectiveness. To address this, we propose a multi-feature fusion model. Based on social cognitive theory, this model captures learner features from multiple dimensions, including the week of discussion, talk length, sentiment polarity of talk, the learner’s demographic factor and pre-test knowledge level. The model integrates both Bidirectional Encoder Representations from Transformers (BERT) and Bidirectional Long Short-Term Memory (Bi-LSTM) models. This integration achieves deep fusion of talk semantics and time-series features, thereby enabling more accurate identification of disruptive talk. Experimental results on real classroom datasets show that our method outperforms existing baseline models across multiple metrics, fully demonstrating its effectiveness and practical value. This research not only advances the development of disruptive talk detection but also provides feasible insights for optimizing collaborative learning environments. Future work will explore the scalability of this model in different educational settings, and investigate its integration with real-time learning platforms.