QoE Prediction Training With User-Feedback Telemetry Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network systems rely on service level agreement (SLA) thresholds as a proxy for user quality of experience (QoE), which can be inaccurate, especially for complex impairments, and training a model to predict true QoE is challenging due to the curse of dimensionality, requiring extensive data.
Innovation Solution
A device generates realistic modifications to telemetry metrics using user feedback to augment the training dataset for a machine learning model, predicting QoE across multiple network layers, enabling predictive application-aware routing to optimize user experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If SLA thresholds are used as a proxy for QoE, then network monitoring is simplified, but measurement precision of QoE deteriorates
Solution Approach 1:
The patent introduces user feedback as an intermediary mechanism that bridges network metrics and actual user experience. Instead of directly measuring QoE through complex user behavior analysis, the system uses feedback signals from users to annotate and validate network metric data, thereby improving QoE measurement accuracy while maintaining operational simplicity.
Solution Approach 2:
The system implements a feedback loop where user experience information is collected and fed back into the network monitoring system. This feedback mechanism allows the system to continuously refine its understanding of QoE by comparing network metrics with actual user perceptions, resolving the contradiction between monitoring simplicity and measurement precision.
2Measurement precision
If a machine learning model is trained to predict true QoE, then measurement precision of QoE is improved, but device complexity and data requirements increase
Solution Approach 1:
The patent creates a simplified representation of user feedback by annotating existing network metric data with user experience labels. Instead of building a complex model from scratch to predict QoE, the system copies and reuses existing network data while enhancing it with feedback information, thereby reducing model training complexity while maintaining high prediction accuracy.
Solution Approach 2:
The system transforms the approach by changing the parameters of existing network metric data through user feedback annotation. Rather than introducing entirely new features that would increase dimensionality, the system modifies and enriches existing parameters with feedback-based labels, making the training data more informative without increasing the feature space complexity.
3Measurement precision
If extensive training data is collected to train QoE prediction model, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent performs preliminary action by pre-processing and annotating existing network data with user feedback labels before model training. This preliminary enrichment of the training dataset with relevant feedback information allows the system to achieve high model reliability without requiring extensive additional data collection, thereby reducing the time loss associated with gathering training data.
Data Source
AI summary
In one embodiment, a device obtains a training dataset to train a machine learning model to predict a quality of experience metric for an online application accessible via a computer network. The device identifies a range of modifications to a telemetry metric in the training dataset that would be realistic with respect to the computer network and the online application. The device receives feedback from a user interface regarding the range of modifications to the telemetry metric. The device augments the training dataset with additional values of the telemetry metric in accordance with the feedback from the user interface.