Crowdsourced Media Skip Model Using ML Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ad skipping technologies face challenges in accurately identifying and skipping unwanted portions of media content, such as commercials and opening credits, especially in diverse content formats like news and sports, and are not scalable for low-power devices, posing DRM issues.
Innovation Solution
A crowdsourcing feedback model that collects user seek behavior data using machine-learning systems to determine the start and stop times of unwanted content portions, enabling automatic skipping and improving accuracy over time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional ad skipping technology analyzes media stream to identify commercials, then it can skip unwanted content, but it becomes increasingly difficult and inaccurate when content has multiple commercials with same differentiators or diverse formats like news and sports
Solution Approach 1:
The patent segments the content analysis task by dividing it into multiple independent features (audio features, video features, metadata features) that are analyzed separately and then combined. This segmentation allows the system to handle diverse content formats more effectively by focusing on multiple dimensions rather than attempting a single complex analysis approach.
Solution Approach 2:
The patent introduces an intermediary machine learning model that acts as a mediator between raw media content and commercial identification decisions. This intermediary processes multiple features and learns optimal patterns for identifying commercials across different content types, reducing the complexity of direct analysis while improving accuracy.
2Power
If server-side solutions are used for ad skipping, then processing power is available, but they are not scalable and have similar efficiency issues to client-side solutions
Solution Approach 1:
The patent performs preliminary action by pre-processing and extracting features from media content during ingestion or in advance of playback. This allows the heavy computational work to be done before the actual skipping decision is needed, enabling faster real-time processing and improving scalability without sacrificing accuracy.
Solution Approach 2:
The patent applies partial action by selectively analyzing only the most relevant features for commercial identification rather than processing all possible content attributes. This reduces computational overhead while maintaining sufficient accuracy, enabling better scalability across multiple devices and content streams.
3Ease of manufacture
If conventional methods identify commercials based on silent frames, audio levels, or content markers, then implementation is simple, but they are inefficient especially for news and sports programming with different commercial formats
Solution Approach 1:
The patent changes parameters by using multiple feature dimensions (audio characteristics, video characteristics, metadata) instead of relying on single parameters like silent frames or audio levels. This multi-parameter approach adapts to diverse commercial formats in news and sports programming while maintaining implementation feasibility through standardized feature extraction processes.
4Measurement precision
If client devices perform complex content processing to identify commercial breaks, then accurate skipping is possible, but many devices are unable to do so due to software or hardware restrictions and lack processing bandwidth
Solution Approach 1:
The patent extracts only the essential features needed for commercial identification from the full media content, separating these critical features from the rest of the content. This extraction approach reduces the processing bandwidth requirement on client devices while maintaining sufficient accuracy for effective skipping functionality.
Data Source
Figure 1A
Figure 1B
Figure 2
AI summary
The disclosure provides methods and systems for skipping unwanted portions of media content based on a crowdsourcing feedback model. The media content may include television programming content, video content, and audio content. The unwanted portions of media content include, for example, television commercials, opening and closing credits. An exemplary crowdsourcing feedback model involves receiving and processing a plurality of user seek information records. These records include time markers of the media content indicating when the users started and stopped fast forwarding or when the users started and stopped moving a scrub bar to jump from one portion of the content to another. When collected, the user seek information records are used to train a machine-learning system to calculate start and stop times of unwanted portions of media content. Once the start and stop times are calculated, user devices are enabled to automatically skip these unwanted portions.