Abstract: As a new-generation human-computer interaction interface, marketing digital humans' conversational intelligence systems have emerged as a crucial engine for driving consumption upgrades and cultivating new quality productive forces in the digital economy. Nevertheless, the mechanisms through which the conversational intelligence features of marketing digital humans influence consumer behavior remain unclear due to the complexity of multidimensional conversational intelligence characteristics, the dynamics of multi-turn interaction patterns, and the challenges in decoupling dual-trust effects, which hinders the healthy development of the marketing digital human industry. Guided by the cognitive-affective trust theory, this study primarily investigates: (1) the consumer behavior phenomena under the interactive influence of multidimensional conversational intelligence features and various external factors; (2) the dynamic cognitive process resulting from the impact of conversational intelligence features on dual trust; (3) the cognitive neuroscience mechanisms underlying dual trust in marketing digital human; and (4) the optimization of conversational intelligence features in marketing digital human and practical application validation. Based on these research findings, the study aims to explore effective pathways for leveraging marketing digital humans' conversational intelligence systems to enhance consumer experiences, optimize business costs, and improve efficiency.
In terms of behavioral phenomena, this research expands the current literature's holistic understanding of how marketing digital humans' conversational intelligence features influence consumer behavior. The advancement of large language model technology has created opportunities for digital humans to reshape human-computer interaction patterns in sales scenarios. However, studies on the multifaceted interactions between multidimensional conversational intelligence features and external factors affecting consumer behavior are still in their nascent stages, lacking systematic and comprehensive characterization of phenomena and identification of key elements. This study intends to employ methods such as panel vector autoregression models to characterize the phenomena of interactions among multidimensional heterogeneous elements and the differential impacts of various conversational intelligence features on consumer behavior, thereby contributing to a holistic understanding and isolating critical influencing factors.
In terms of psychological processes, this study innovatively proposes a dynamic trust processing framework for multi-turn conversations with marketing digital humans. Perceived trust is a crucial factor influencing the consumer-digital human interaction ecosystem and marketing effectiveness. Current research on perceived trust in digital human conversational contexts predominantly examines single-turn interactions or adopts static perspectives. However, real interactions between humans and digital humans are characterized by multi-turn, bidirectional exchanges, and the establishment of human-computer trust is a continuously dynamically adjusted calibration process. Trust levels evolve over time and eventually stabilize. Unlike previous structural analysis paradigms, this study adopts a process-tracing paradigm and employs Bayesian decision modeling to construct a psychological process coding model for trust levels. This approach helps explain the psychological processes underlying human-computer interaction behaviors, enhances the depth of theoretical construction, and provides richer evidence for comparing and validating divergent conclusions in the field of human-computer trust.
In terms of underlying mechanisms, this study systematically unveils the neural mechanisms of dual trust in marketing digital humans. According to cognitive-affective trust theory, consumer trust comprises two dimensions: cognitive trust and affective trust. Different conversational intelligence features exert differentiated effects on dual trust, which in turn diversely impact consumer behavior. However, due to the subjective and implicit nature of trust perception, distinguishing between cognitive trust and affective trust through consumer self-reports is challenging. Neuromarketing methods offer significant advantages in observing implicit variables. This study plans to utilize fMRI as a tool for characterizing the neural mechanisms of dual trust and researching consumer behavior. By disentangling the effects of dual trust, assessing differences in activation intensity, and distinguishing effective functional pathways, a brain network model of dual trust will be constructed. Furthermore, this study intends to use neural data, behavioral data, and historical consumption data as input variables, with trust levels and purchase intention as output variables. Based on a deep learning approach combining Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, a predictive model of consumer behavior in marketing digital human conversational contexts will be developed.

Key words: marketing digital humans, cognitive-affective trust theory, conversational intelligence systems, multidimensional intelligent features, consumer behavior