Event Classifier Training Tool for Data Stream Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for real-time detection of commercial blocks or content breaks in data streams, such as audio and video, are inefficient due to the need for extensive manual tagging and updating of advertisement databases, and existing machine learning algorithms are not optimized for real-time usage.
Innovation Solution
A computer-controlled method that detects trigger features from parameter variations in data streams, identifies associated separators, and generates balanced sets of positive and negative training samples to train an event classifier, allowing for real-time detection of events without relying on extensive fingerprint matching or manual annotation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fully automated machine learning detection algorithms are used for commercial block detection, then detection accuracy is improved, but training time and computational resources are excessively consumed
Solution Approach 1:
The patent segments the training process by dividing training samples into positive samples (from annotated data streams) and negative samples (synthetically generated). This segmentation allows the algorithm to learn from limited annotated data while using generated negative samples to improve detection accuracy without requiring extensive manual annotation of all training data.
Solution Approach 2:
The patent performs preliminary action by using a small set of annotated data streams to pre-train the detection algorithm, then generates additional negative training samples based on this preliminary model. This approach reduces the overall training time by establishing a baseline model quickly and then refining it with generated samples rather than requiring extensive manual annotation from the start.
2Adaptability or versatility
If extensive manual tagging and updating of advertisement databases is performed, then detection coverage is improved, but operational complexity and costs increase
Solution Approach 1:
The patent implements self-service by enabling the detection system to automatically generate negative training samples and update its own training data without requiring manual intervention. The system uses annotated data streams to generate synthetic negative samples, allowing it to expand its detection coverage automatically without human operators manually tagging additional content.
Solution Approach 2:
The patent applies parameter changes by transforming annotated data streams into both positive and negative training samples through parameter modifications. By changing parameters such as timing, content characteristics, and contextual features of the annotated data, the system generates diverse training samples that improve detection coverage across different scenarios without requiring manual creation of each sample.
3Measurement precision
If more training samples are used to train the event classifier, then classification accuracy is improved, but processing time and computational load increase
Solution Approach 1:
The patent applies partial action by using a balanced approach where only the necessary number of training samples are used - enough to achieve good classification accuracy but not so many that processing time becomes excessive. The system generates a balanced set of positive and negative samples that provides sufficient training without the diminishing returns of using excessively large datasets.
Solution Approach 2:
The patent uses parameter changes to optimize the number and composition of training samples. By adjusting parameters related to sample selection, balancing ratios, and generation criteria, the system achieves optimal classification accuracy with a manageable number of training samples, thereby maintaining high processing speed while improving accuracy.
Data Source
AI summary
Accurate real time automatic detection of events in content of a data stream, such as a transition to a commercial block in the content of a broadcast audio/video data stream, relies on a trainable event classifier that operates on a well-balanced training set input to the classifier. The present disclosure provides a computer controlled method of operating a training tool for classifying events annotated in the content of a data stream. The training tool presents training samples comprising separators and corresponding descriptors that relate to trigger features obtained from variations in parameters of the annotated data stream, and derived features restoring relationships between various separators and corresponding descriptors.