Journal of Intelligence

Conversational interactions, rich in both linguistic and vocal cues, provide a natural context for studying these processes. In this work, we propose an explainable multimodal transformer framework that integrates textual semantics (via RoBERTa) and acoustic prosody (via WavLM) to advance emotion understanding. By projecting both modalities into a shared latent space, our model captures the complementary contributions of language and speech to affective communication, achieving an 0.83 accuracy value across five emotion categories. Crucially, we embed explainable AI (XAI) techniques including Integrated Gradients and Occlusion to attribute predictions to specific linguistic tokens and prosodic patterns, thereby aligning computational mechanisms with human cognitive processes of emotion perception. Beyond performance gains, this work demonstrates how multimodal AI systems can support transparent, human-centered emotion recognition.

This study introduces and validates the construct of Teacher Emotional Commitment (CED), understood as the conative–behavioral dimension that characterizes the emotional bond that teachers establish with their students. To this end, two complementary studies were conducted in the Autonomous Community of the Canary Islands (Spain), with the aim of: to empirically isolate the factorial structure of CED and differentiating it from related constructs, such as empathy; to analyze its presence in both active teachers and those in initial training; and to test the theoretical model’s validity by expanding the sample and enlarging the response scale. Study 1 involved 854 practicing teachers and 701 teachers in training, following a validation process that included exploratory and confirmatory factor analysis, as well as item response theory models. The results showed a four-factor structure: loving proactivity, teacher compassion, instructional commitment, and communicative affectivity, with adequate reliability and discriminant validity indices with respect to empathy. Study 2, with an expanded sample of 2096 participants, confirmed the robustness of the model. The findings allow us to consider CED as a psychological competence that can be trained, with relevant implications for improving the educational relationship, student learning, and the emotional well-being of teachers.

This study reviews 33 meta-analyses and systematic reviews on Computational Thinking (CT), focusing on research quality, intervention effectiveness, and content. Quality assessment of included studies was conducted using the AMSTAR 2 tool. The meta-analysis achieved an average score of 10.9 (a total of 16 points), while systematic reviews scored an average of 6.1 (a total of 11 points). The 15 meta-analyses showed diverse intervention strategies. Project-based learning, text-based programming, and game-based learning demonstrate more pronounced effects in terms of effect size and practical outcomes. Curricular integration, robotics programming, and unplugged strategies offered additional value in certain contexts. Gender and disciplinary background were stable moderators, while grade level and educational stage had more conditional effects. Intervention duration, sample size, instructional tools, and assessment methods were also significant moderators in several studies. The 18 systematic reviews used a five-layer framework based on ecological systems theory, covering educational context (microsystem), tools and strategies (mesosystem), social support (exosystem), macro-level characteristics (macrosystem), and CT development (chronosystem). Future research should focus on standardizing meta-analyses, unifying effect size indicators, and strengthening longitudinal studies with cognitive network analysis. Additionally, systematic reviews should improve evidence credibility by integrating textual synthesis and data-driven reasoning to reduce redundancy and homogeneity.