Abstract
<title>Abstract</title> <p>The rise of Generative Artificial Intelligence (GAI) in higher education has instigated paradigmatic shifts in teaching and research for faculty. However, existing research predominantly relies on faculty self-reported data, lacking objective analysis of actual usage behaviors, which results in a cognitive gap between "intention" and "action." To address this gap, this study utilizes retrospective log analysis and Latent Profile Analysis (LPA) based on usage log data from a university's smart learning platform spanning January to June 2025, to conduct a multidimensional quantitative characterization of GAI usage behaviors among 358 faculty members. The research constructs four categories of behavioral indicators: usage intensity, functional breadth and depth, temporal patterns, and evolutionary trends. Cluster analysis identified two distinct behavioral profiles: "Teaching-Assistance Type" and "Research-Support Type." The former primarily uses the platform on weekdays, focusing on teaching-related functions such as lesson plan generation and assignment grading, with a predominantly beginner-level technical background. The latter is more active during non-working hours, engaging deeply with research assistance and knowledge management functions, and largely possesses an expert-level technical background. The contributions of this study are threefold: First, it proposes an analytical framework for faculty GAI usage behaviors based on log data. Second, it infers and validates influencing factors of technology adoption from objective behavioral data. Third, it provides empirical evidence for universities to implement differentiated faculty training and support policies, promoting the precise integration of GAI in higher education.</p>