Abstract
Team modeling remains a fundamental challenge at the intersection ofArtificial Intelligence and the Social Sciences. Social Science researchemphasizes the need to jointly model dynamics and relations, while practicalapplications demand unified models capable of inferring multiple teamconstructs simultaneously, providing interpretable insights and actionablerecommendations to enhance team performance. However, existing works do notmeet these practical demands. To bridge this gap, we present TRENN, a noveltempo-relational architecture that integrates: (i) an automatic temporal graphextractor, (ii) a tempo-relational encoder, (iii) a decoder for team constructprediction, and (iv) two complementary explainability modules. TRENN jointlycaptures relational and temporal team dynamics, providing a solid foundationfor MT-TRENN, which extends TReNN by replacing the decoder with a multi-taskhead, enabling the model to learn shared Social Embeddings and simultaneouslypredict multiple team constructs, including Emergent Leadership, LeadershipStyle, and Teamwork components. Experimental results demonstrate that ourapproach significantly outperforms approaches that rely exclusively on temporalor relational information. Additionally, experimental evaluation has shown thatthe explainability modules integrated in MT-TRENN yield interpretable insightsand actionable suggestions to support team improvement. These capabilities makeour approach particularly well-suited for Human-Centered AI applications, suchas intelligent decision-support systems in high-stakes collaborativeenvironments.