.png){kind=icon}
.png){kind=icon}
Return
Context-aware visual scene modeling for cinematic media using a minimum-attribute-difference ranking algorithm
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-026-21850-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Context-aware multimedia systems require structured representations of visual content to support intelligent scene understanding and recommendation. However, most existing movie recommender systems primarily rely on high-level metadata and do not exploit fine-grained cinematic attributes that contribute to visual storytelling. To address this limitation, this paper presents CarMovie, a context-aware, content-based recommendation framework that models cinematic scenes as structured contextual instances defined by location, time, and mood, and associates them with visual attributes including color schemes and shot types. A context-annotated cinematic dataset is constructed from real movies using a structured annotation protocol to support contextual scene analysis and recommendation. To enable efficient and interpretable recommendation, the framework incorporates a Minimum-Attribute-Difference (MAD) ranking mechanism that performs reference-based, attribute-level comparison using binary feature encoding. Unlike conventional similarity measures based on global vector comparison, MAD prioritizes contextual consistency through structured attribute matching while maintaining low computational complexity. A mobile prototype system is implemented to demonstrate the practical applicability of the proposed framework. The approach is evaluated using a two-stage experimental setting involving general and personalized recommendation scenarios with real cinematic data. Results demonstrate consistently high precision under the controlled evaluation setting, while recall, F-measure, and accuracy improve from 24.67% to 38.10%, 39.57% to 55.17%, and 24.67% to 91.33%, respectively, following contextual personalization. These findings indicate that structured contextual modeling can effectively enhance recommendation quality, although recall remains influenced by dataset sparsity and contextual coverage. The proposed framework integrates contextual scene modeling with interpretable attribute-level ranking, establishing a connection between recommender systems and cinematography and supporting fine-grained recommendation of cinematic visual attributes within multimedia environments.
Keywords:
Context-aware recommendation system
Visual story
Movie
Context
Visual approach
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W