Ask and focus more: Question-prompt uncertainty allocation for dual-controllable video captioning

Video captioning aims to generate natural language descriptions from video content via hierarchical architectures that capture key visual elements. Although entities, predicates, and syntactic structures are critical for coherent descriptions, existing controllable methods often struggle to dynamically emphasize salient information due to reliance on predefined entity lists and static fusion strategies. Moreover, uncertainty in aligning multimodal features (e.g., visual actions and textual semantics) further weakens model robustness. To address these challenges, we propose QPDC, a Dual-Controllable video captioning framework that introduces a Question-Prompt mechanism to adaptively guide caption focus. Unlike traditional entity-driven approaches, QPDC adopts a unified question-prompt strategy to maintain semantic consistency while enabling flexible attention shifts. Specifically, the Question-Prompt (Q-Prompt) module dynamically steers the model's focus according to query intent, emphasizing salient content without predefined constraints. The Selective Fusion Module (SFM) hierarchically integrates Q-Prompt guidance with local action features and global sentence context to support multi-granularity control. In addition, a Dirichlet-based strategy is employed to model uncertainty in multimodal weight allocation, improving adaptability under varying data conditions. Extensive experiments on MSVD and MSR-VTT show that QPDC achieves competitive performance among controllable video captioning methods and produces more diverse and user-preferred descriptions. Code will be available at https://github.com/XiaotaiCHEN/QPDC.git.