Article Preview
TopIntroduction
The intelligent interaction systems of the future use emotion recognition and behavioral modeling to enhance adaptive and human-friendly interactions (Kim & Seo, 2022; Shafik, 2025; Tozadore & Romero, 2025). As digital technologies penetrate everyday life, interest in systems that can sense a user’s emotions and take designed to guide behavioral responses has grown considerably (Haddock et al., 2022; Seiferth et al., 2023). Applications of such affect-sensitive mechanisms vary greatly, but they can be found in intelligent assistants (Nafees et al., 2025), distance learning technologies (Fernández-Herrero, 2024; Luo et al., 2024), the mental health and wellness field (Kumar et al., 2025; Nawaz et al., 2025), and personalized recommendation systems (Mahmud et al., 2025). In addition, real-world interactive services such as artificial intelligence (AI)–driven customer service systems increasingly emphasize user-centered response quality and adaptive decision making, in which emotion-aware guidance is critical for improving interaction outcomes (Wu et al., 2024). Moreover, behavior modeling has also demonstrated practical value in high-stakes decision scenarios, such as fraud detection and risk management, highlighting the broader importance of accurate behavioral inference in intelligent systems (Tsai, 2025). Integration of emotional and behavioral modeling is therefore one of the most important steps toward the achievement of truly empathetic and responsive AI (Cinar & Bilodeau, 2024; Singh et al., 2024).
Although there have been significant advancements in affective computing due to deep learning, all the currently existing studies are centered on emotion recognition and behavior guidance considered independently (García-Hernández et al., 2024). Emotion recognition models generally recover user affect through text, audio, or visual modalities (Hosseini et al., 2024), whereas behavior forecasting models are developed to predict user intentions and choice trends (Harish & Benitta, 2024). Crucially, existing strategies tend to ignore the inherent interdependence of emotion and behavior, according to which emotions influence behaviors and vice versa (Uchida et al., 2022). From an organizational and system-level perspective, such a separation may further limit the effectiveness of AI capability in supporting creativity, decision quality, and performance outcomes, which rely on coherent and human-aligned interaction intelligence (Ali & Simmou, 2025). The consequence of ignoring such a two-way relationship is a diminishment int these systems' ability to make emotionally coherent decisions and provide essentially adaptive responses.
To address this gap, the current study explores a single deep-learning architecture, which simultaneously addresses user emotion perception and behavior control. The guiding principle is that emotion-based dynamics should be used as a contextual stimulus to model user behavior; on the other hand, emotion knowledge is enhanced with the help of behavioral feedback. In particular, a time-conscious emotion monitoring system is proposed to record the trends of emotional conditions throughout interactions. Meanwhile, a multi-task learning (MTL) approach is used to match emotional and behavioral representations in a common latent space. The combination of these components aims to allow two-way improvement of affect recognition and behavior prediction activities, which will eventually result in more predictable and human-similar interaction modes.