This study explores the influence of integrating macro-built environment data, micro-level subjective and objective street environment perceptions, socioeconomic demographics of residents' burnout emotions and their spatial-temporal heterogeneity, with a particular focus on the changing patterns across three phases of Shanghai's pandemic lockdown. The research establishes a pipeline based on large language models to extract burnout emotions from social media content, and constructs a multi-dimensional spatial-temporal analytical framework integrating OLS, geographically weighted regression, and nonlinear interpretable machine learning modeling techniques. By collecting social media data, the study first validates the effectiveness of large language models in identifying burnout emotions from social media (with a consistency coefficient of 0.61 with human ratings), providing a scalable method for real-time psychological monitoring during crises. Analysis using spatial analysis reveals significant spatial clustering patterns of burnout emotions among Shanghai residents. Geographically weighted regression identifies five typical spatial patterns of influencing factors. Walkability consistently protected against burnout citywide, while green spaces lost restorative function in commercial centers. High-density housing areas with complex visual environments amplified psychological distress, and central districts formed "youth burnout peaks" where young, educated populations concentrated. Longitudinal results reveals significant nonlinear relationships and critical threshold effects between influencing factors and residents' burnout emotions, with these relationship patterns exhibiting various forms of transformation across the three phases. For example, population density's impact on burnout shifted from a negative correlation and "Reverse Excitement" pattern before lockdown to a "Basic Need" pattern during lockdown. Similarly, the young population ratio evolved from a single-threshold model with positive correlation ("Semi-Performance") to negative correlation during and after lockdown ("Performance" to "Excitement") with complex multi-threshold interval models. Green space transformed from weak positive correlation as an "Excitement Factor" before lockdown to strong positive correlation as a "Performance Factor" during lockdown, indicating that visually accessible but functionally inaccessible green spaces became sources of stress. Visual complexity reversed from positive correlation ("Basic Need") before lockdown to negative correlation ("Excitement Factor") highlighting how environmental monotony enhanced the value of perceptual stimulation. These findings challenge static urban health standards, suggesting environmental influences operate through dynamic systems that transform during crises, providing frameworks for resilient urban planning integrating spatial structure with psychological wellbeing.