When Machine Learning Algorithms Meet User Engagement Parameters to Predict Video QoE

  • PDF / 1,797,157 Bytes
  • 19 Pages / 439.37 x 666.142 pts Page_size
  • 71 Downloads / 229 Views

DOWNLOAD

REPORT


When Machine Learning Algorithms Meet User Engagement Parameters to Predict Video QoE Fatima Laiche1   · Asma Ben Letaifa2 · Imene Elloumi1 · Taoufik Aguili1

© Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract In recent years, there has been a substantial increase in the distribution of videos over the Internet, and this has become one of the major activities that attract extensive attention. This means that users expect to watch videos of the highest quality. Quality of experience (QoE) describes the degree of satisfaction or annoyance of a user when they are using a multimedia service or application. Meeting users’ expectations requires understanding the factors that influence QoE and efficiently managing resources to optimize video quality. The current objective approaches that assess QoE mostly rely on the analysis of video traffic. However, recent research has demonstrated that this approach cannot sufficiently evaluate perceived QoE and that multiple factors, including media technical features, influence QoE. It is crucial for service providers to identify the effects of social context, in addition to those of user-related, content-related, and system factors, on perceived QoE of the end user. Recent studies have focused on understanding the characteristics of user behavior and engagement, as well as the effect of these factors on QoE. In this study, we use social context factors and user engagement as subjective factors to structure a user QoE evaluation model. First, we study social context factors and user engagement characteristics and investigate their correlation with QoE. Next, we build a metric that estimates the end-to-end QoE for a specific aspect of user actions. Then, by simulating mathematical metrics, we employ machine learning models to predict QoE; finally, we validate this approach using metrics for statistical evaluation of quality prediction models. Keywords  Video service · QoE · MOS · User engagement · Machine learning

* Fatima Laiche [email protected] Asma Ben Letaifa [email protected] 1

Communication Systems Laboratory, ENIT, University of Tunis El Manar, Tunis, Tunisia

2

MEDIATRON Laboratory, SUPCOM, University of Carthage, Tunis, Tunisia



13

Vol.:(0123456789)



F. Laiche et al.

1 Introduction As communication technologies have substantially developed and the Internet is being used extensively, service providers have begun providing end users with new multimedia streaming services such as short video sharing, internet television, video on demand, and live event video sharing. Out of these, online streaming videos are the most popular type of multimedia streaming service. Videos are generally hosted and shared on platforms such as YouTube, Facebook, and Netflix. Video streaming content can be displayed on different devices with diverse capabilities and transmitted through networks that have limited bandwidths and variable video viewing resolutions. Furthermore, video streaming is increasing and has become visibly a major contributor