Traditional football player similarity methods often rely on collapsed positional heatmaps or aggregate statistics, conflating spatial occupation with functional behavior. As a result, players operating in similar zones of the pitch may be incorrectly identified as similar despite performing fundamentally different actions. We propose an action-aware spatial similarity framework that models player behavior as a third-order spatial-action tensor, separating spatial distributions by action type. Channel-wise spatial association is computed using a bivariate spatial statistic, and similarities are aggregated through a profile-weighted fusion scheme. This design preserves functional context while incorporating spatial dependence. The framework is evaluated using cross-match self-retrieval on two independent datasets: the Impect Bundesliga 2023/24 season and the FIFA World Cup 2022. Results demonstrate consistent improvements over a collapsed baseline across event budgets, with performance gains increasing as event volume grows. Qualitative case studies further illustrate that the proposed method retrieves functionally aligned players rather than those sharing only positional overlap. These findings highlight the importance of action-specific spatial decomposition for robust and interpretable player similarity analysis in football analytics.