Why That Rating? : Explainable Data-Driven Opinion Score Distribution Models for Video QoE

Austin, Edward and Race, Nicholas and Lyko, Tomasz (2025) Why That Rating? : Explainable Data-Driven Opinion Score Distribution Models for Video QoE. In: 2025 17th International Conference on Quality of Multimedia Experience (QoMEX) :. IEEE.

Text (QoMEX_QoE_Paper-2)
QoMEX_QoE_Paper-2.pdf - Accepted Version
Available under License Creative Commons Attribution-NonCommercial-NoDerivs.
Download (249kB)

Abstract

Driven by the need to better reflect and understand audience quality of experience content providers and deliverers are no longer solely using models for the Mean Opinion Score but also the opinion score distribution. In tandem, motivated by the desire to understand how these models predict the QoE and which features contribute positively to it, there has been increased emphasis on explainable QoE modelling. Recently the advantages of directly explainable models - where model outputs can be explained using the inputs and the model’s inner workings have been championed. These models provide clear insights into how different features contribute positively, or negatively, to QoE. To date, research into directly explainable methods has focussed on MOS, rather than opinion score distribution, modelling. To bridge this gap we discuss the feasibility of using multinomial logistic regression for directly explainable MOS modelling, demonstrating our approach on a short form streamed video dataset and showing that it compares favourably to other explainable methods.