developing a predictive model of quality of experience for internet video athula balachandran -cmu
TRANSCRIPT
Developing a Predictive Model ofQuality of Experience for Internet Video
Athula Balachandran
-CMU
QoE(Quality of Experience)
• Traditionally– Peak Signal-to-Noise Ratio (PSNR)
• Now– rate of buffering– bitrate– join time– viewing time– number of visits
Fraction of content viewed
MOTIVATION
Why to improve QoE(Quality of Experience)
• Advertisement and subscription based revenue
• ...• ...• ...
Money
Contributions
• Highlighting challenges in obtaining a robust video QoE model
Industry-standard quality metrics
• Average bitrate
• Join time
• Buffering ratio
• Rate of buffering
QoE
Challenges in developing QoE
• Complex relationships
• Interaction between metrics
• Confounding factors
Contributions
• Highlighting challenges in obtaining a robust video QoE model
• A roadmap for developing Internet video QoE that leverages machine learning
• A methodology for addressing confound-ing factors that affect engagement
Roadmap
• Tackling complex relationships and interdependencies
• Identifying the important confounding factors
• Refinement to account for confounding factors
Machine learning model
Confounding Factors
• Content attributes– type of video and the overall popularity
• User attributes– user’s location, device and connectivity
• Temporal attributes– time of day, day of week and time since relea
se
:
Information gain
Approach Overview
Summary of confounding factors
Refine the decision tree model
• Candidate approaches– Add as new feature– Split Data
Contributions
• Highlighting challenges in obtaining a robust video QoE model
• A roadmap for developing Internet video QoE that leverages machine learning
• A methodology for addressing confound-ing factors that affect engagement
• A practical demonstration of the utility of our QoE models to improve engagement
Evaluation
Evaluation
Dicussion