Scholarlink

<p>With the rapid expansion of communication technologies and the growing reliance on online services, accurate assessment of users’ Quality of Experience (QoE) without revealing their raw data has become increasingly critical. Although several studies have addressed QoE modeling, few have effectively captured its inherently subjective and person-centered nature while safeguarding user data. This paper introduces a Federated Learning (FL)-based framework that enables decentralized QoE assessment, allowing users to collaboratively train models without exposing their private data. To improve personalization and mitigate the challenges of non-Independent and Identically Distributed (non-IID) data, the proposed framework incorporates an initial clustering phase followed by Personalized Federated Learning (PFL). Two novel methods are introduced: (1) Clustered PFL for QoE Assessment (CPFLQA), which combines user clustering with personalized FL to reduce intra-cluster data diversity and enhance model accuracy; and (2) CPFLQAW, which extends CPFLQA by integrating a local data weighting mechanism to prioritize more relevant QoE instances, thus improving training stability and preventing model drift. Experimental results demonstrate the effectiveness of both methods. CPFLQA achieves a 4% average improvement in F1-score compared to the generalized FL baseline, while CPFLQAW provides an additional 21% gain over CPFLQA. Furthermore, the evaluations on another QoE dataset collected in this research showed a consistent performance pattern. CPFLQA delivered a slight improvement in F1-score over the baseline model, while CPFLQAW achieved a substantial additional gain of around 9.3% compared to CPFLQA. These findings highlight the potential of joint of clustering, personalization, and weighting in building accurate, user-centric QoE models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel clustering and personalized federated learning-based QoE assessment scheme with training data weighting

  • 期刊论文

  • 发表日期：

    2026年6月17日

查看全文
  • Skokufe Motaharipour,
  • Behrouz Shahgholi Ghahfarokhi,
  • Saeid Afshari,
  • Nahid Amani

摘要

With the rapid expansion of communication technologies and the growing reliance on online services, accurate assessment of users’ Quality of Experience (QoE) without revealing their raw data has become increasingly critical. Although several studies have addressed QoE modeling, few have effectively captured its inherently subjective and person-centered nature while safeguarding user data. This paper introduces a Federated Learning (FL)-based framework that enables decentralized QoE assessment, allowing users to collaboratively train models without exposing their private data. To improve personalization and mitigate the challenges of non-Independent and Identically Distributed (non-IID) data, the proposed framework incorporates an initial clustering phase followed by Personalized Federated Learning (PFL). Two novel methods are introduced: (1) Clustered PFL for QoE Assessment (CPFLQA), which combines user clustering with personalized FL to reduce intra-cluster data diversity and enhance model accuracy; and (2) CPFLQAW, which extends CPFLQA by integrating a local data weighting mechanism to prioritize more relevant QoE instances, thus improving training stability and preventing model drift. Experimental results demonstrate the effectiveness of both methods. CPFLQA achieves a 4% average improvement in F1-score compared to the generalized FL baseline, while CPFLQAW provides an additional 21% gain over CPFLQA. Furthermore, the evaluations on another QoE dataset collected in this research showed a consistent performance pattern. CPFLQA delivered a slight improvement in F1-score over the baseline model, while CPFLQAW achieved a substantial additional gain of around 9.3% compared to CPFLQA. These findings highlight the potential of joint of clustering, personalization, and weighting in building accurate, user-centric QoE models.