Emotion-Based Scene Recommendation to Reduce OTT Seeking Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional OTT services face inefficiencies in server-side data processing and network bandwidth usage due to manual user searches for specific video scenes, leading to poor user experience and increased computational load, while lacking the ability to recommend content that resonates with users' granular emotional responses.
Innovation Solution
A method and apparatus that record and analyze users' emotional responses to content scenes in real-time, generating additional content based on emotion information, and providing personalized recommendations using learning models to predict and deliver scenes of interest without manual seeking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If users manually search for specific scenes by repeatedly using seeking functions, then users can find desired content, but server computational load increases and network bandwidth is wasted
Solution Approach 1:
The system pre-generates and stores thumbnail images at multiple time points of video content before users need to search. When users want to find specific scenes, they can directly select from pre-generated thumbnails without triggering repeated seeking operations, thus reducing server computational load and network bandwidth consumption while maintaining ease of content access
Solution Approach 2:
Instead of transmitting the actual video data multiple times during seeking operations, the system creates and transmits only thumbnail image copies at key time points. These thumbnails serve as lightweight representations that allow users to navigate content without requiring the server to repeatedly stream full video segments, significantly reducing network bandwidth usage
2Device complexity
If conventional recommendation engines use coarse data like viewing history, then implementation is simple, but they cannot understand granular emotional responses to specific scenes
Solution Approach 1:
The video content is divided into multiple discrete time points with corresponding thumbnail images generated at each segment. Emotion recognition is performed independently on each thumbnail-image-timepoint triplet, allowing granular emotional analysis at scene level rather than requiring analysis of entire videos or complex temporal sequences, thus achieving high measurement precision with manageable system complexity
Solution Approach 2:
Thumbnail images serve as intermediaries between the video content and emotion recognition algorithms. Instead of directly analyzing raw video frames or audio, the system uses pre-processed thumbnail images as mediators that capture essential visual information at key moments, simplifying the emotion detection process while maintaining accuracy in understanding user emotional responses to specific scenes
Data Source
AI summary
Provided is an operating method of a server, comprising obtaining emotion information of a user about at least one scene constituting content played on an electronic device; producing an additional content including the emotion information of the user, and providing an additional service based on the content into the electronic device, based on at least one of factor information representing a factor by which the emotion information of the user about the at least one scene is obtained and the additional content.