Universal Poor-QoE Assessment Model for Network Data Streams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing poor-QoE assessment methods are limited to specific applications and cannot be universally applied to assess quality of experience across similar types of applications, hindering network operators' ability to manage application traffic effectively.
Innovation Solution
A method and device for performing poor-QoE assessment using a trained model that can assess data streams corresponding to a target application type, allowing for the identification and screening of data streams based on preset conditions, followed by feature extraction and model training to obtain a trained assessment model that can evaluate multiple data streams of the same type, enabling universal application across similar types of applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If existing poor-QoE assessment methods are used, then assessment can be performed on a specific application, but it cannot be universally applied to assess quality of experience across similar types of applications
Solution Approach 1:
The patent creates a universal poor-QoE assessment model that can evaluate multiple types of applications (video streaming, online gaming, file downloading, etc.) using the same model structure and feature extraction process. The model is trained on diverse data streams from different applications and then applied universally to assess QoE across all similar application types without requiring separate models for each application.
Solution Approach 2:
The patent transforms the assessment approach by changing from application-specific parameters to generalized flow-level parameters. Instead of using application-specific metrics, the system extracts general features from data streams (such as packet inter-arrival times, packet sizes, throughput variations) that can be applied across different application types, enabling universal assessment while maintaining accuracy.
2Measurement precision
If a trained assessment model is obtained by training with screened data streams, then the model can better perform poor-QoE assessment on target application types, but the training process becomes more complex
Solution Approach 1:
The patent applies preliminary screening to data streams before training the assessment model. The system pre-processes and selects representative data streams from various application types that meet specific criteria (such as sufficient data volume, diverse application coverage, and quality thresholds). This preliminary action ensures that the model is trained on high-quality, representative data, improving assessment accuracy while making the training process more systematic and manageable.
Solution Approach 2:
The patent segments the training data into different categories based on application types and flow characteristics. By dividing the large dataset into manageable segments (video streaming flows, gaming flows, file transfer flows, etc.), the system can selectively train on representative samples from each segment, improving model accuracy for target applications while reducing overall training complexity through structured data organization.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Embodiments of this application provide a poor-QoE assessment method and a related device. The method includes: obtaining m data streams, where the m data streams are data streams transmitted by a network device; screening the m data streams, and determining that p data streams that are in the m data streams and that meet a first preset condition are data streams corresponding to a first type; performing feature extraction on the data streams corresponding to the first type, to obtain known feature vectors corresponding to the first type; training, by using the known feature vectors corresponding to the first type, an untrained first poor-QoE assessment model, to obtain a trained first poor-QoE assessment model; and performing, by using the trained first poor-QoE assessment model, poor-QoE assessment on n data streams corresponding to the first type, to obtain a first poor-QoE assessment result. This method can be used to perform poor-QoE assessment on a same type of applications instead of a specific application, and can be universally used.