Design of Multimodal Behavior Analysis and Evaluation System for Digital Leadership of University Teachers Integrated with Perceiver IO

University ideological and political education faces difficulties in quantitatively evaluating teachers’ digital leadership in complex digital teaching environments. This study proposes a multimodal evaluation framework based on the Perceiver IO architecture, using cross-modal attention and structured queries to model digital leadership behaviors across pedagogical contexts. Multimodal inputs, including speech, visual behavior, instructional content, and interaction logs, are mapped into a shared latent space and decoded into six leadership dimensions. Supervised learning enables structured interpretation of leadership patterns and dimension-specific scoring. Experimental results demonstrate strong model performance, with an F1-score of 0.88 for technology application, 0.80 for value guidance, and an AUC of 0.90 for digital influence. These results indicate that the proposed framework can accurately and robustly assess digital leadership in complex teaching environments. By integrating speech, visual, content, and log-based information through a unified latent representation, the method provides an engineering-oriented approach for multimodal behavior analysis, interpretable teaching evaluation, and cross-modal decision support.