Generative recommendation reformulates item prediction as semantic identifier generation, but episodic content requires understanding narrative evolution rather than user preference. This setting introduces efficiency challenges due to redundant visual contexts and costly explicit reasoning generation.
NarraLite Framework
The authors propose NarraLite, an efficient multimodal generative recommendation framework that jointly compresses perception and reasoning. The system addresses the specific needs of short-form drama continuation across UGC, PGC, and OOD settings.
Technical Mechanisms
Progressive Spectral Compression selectively distills long visual contexts into compact narrative-relevant evidence. This preserves transition-critical information while reducing redundant visual computation.
Latent Narrative Reasoning introduces context-routed latent reasoning tokens. These align contextualized representations with future continuation semantics, enabling implicit narrative inference without autoregressively decoding textual rationales.
Evaluation and Results
The team established a user-agnostic multimodal benchmark for short-form drama continuation. Extensive experiments demonstrate that NarraLite consistently improves continuation accuracy, narrative coherence, and robustness over existing approaches while achieving a favorable accuracy-efficiency trade-off.
Source: https://arxiv.org/abs/2609.16070
