Cognitive difference text classification in online knowledge collaboration based on SA-BiLSTM hybrid model

Abstract In the process of online knowledge collaboration, complex cognitive-difference texts are continually generated through collaborative editing processes. Effectively identifying and classifying these cognitive difference texts enables contributors to stay updated on the current status of knowledge editing, thereby enhancing group collaboration efficiency. However, accurate extraction of semantic features and contextual patterns from such texts remains challenging in multi-dimensional discourse contexts. To address this limitation, we developed a classification system based on the mapping relationship between conceptual relationships and cognitive differences. Based on this framework, we proposed a hybrid SA-BiLSTM architecture that integrates self-attention mechanisms with bidirectional long short-term memory networks for fine-grained cognitive difference text categorization. This study evaluated the SA-BiLSTM model through systematic experimentation with the Baidu Encyclopedia dataset, focusing on three aspects: architectural ablation studies comparing variant structures, comparative analyses with mainstream baseline models (FastText, TextCNN, RNN, BERT, and RoBERTa), and evaluations of the model’s generalization and robustness. Experimental results indicate that the proposed model achieves superior classification accuracy compared to conventional approaches, demonstrates effective mitigation of semantic ambiguity, and exhibits enhanced domain adaptation capabilities. The findings highlight the framework’s technical advantages in integrating attention mechanisms with sequential modeling, providing a viable solution for cognitive difference analysis in large-scale knowledge collaboration platforms.