In order to reduce the occurrence of accidents during construction in urban underground spaces, a combined methodology of content analysis and social network analysis was employed to deconstruct human factors in accidents during urban underground space construction for risk reduction. Firstly, based on PAR accident causation model, a tri-dimensional analytical framework of "Subject-Factor-State" was developed for human-induced accidents in urban underground construction. Secondly, authentic accident cases were utilized to encode and extract causal factors, while causal subjects and states were systematically identified. Finally, a structural model of human-induced causation was constructed through network co-occurrence, centrality analysis, and correlation analysis to reveal inherent causal mechanisms. Key findings reveal: Collapse, poisoning, and suffocation constitute the predominant accident types, with 24 high-frequency human factors identified, including 9 core contributing factors. The structural model demonstrates that decision-making managers bear critical safety responsibilities among causal subjects. The core factor "inadequate safety technical disclosure", functions as a pivotal risk hub significantly influencing the causal network. Causal states exhibit strong interdependencies, particularly highlighting the pronounced interaction between information deficiency and other states.