Scholarlink

<p>Research on cognitive load and mental workload in human–computer interaction (HCI) spans learning technologies, healthcare interfaces, safety-critical operations, and biosignal-driven user-state assessment, yet the field lacks a consolidated, data-driven account of its thematic structure, its evolution, and the (in)dependence of its subfields. This study maps the thematic landscape of cognitive load–oriented HCI research published between 2006 and 2025 using a Structural Topic Model (STM) of titles, abstracts, and author keywords indexed in Scopus. Following a PRISMA-guided corpus-construction protocol, 1,474 documents (1,024 peer-reviewed conference papers and 450 journal articles) were retained; conference proceedings were deliberately included as a primary archival venue in HCI. The number of topics was selected by comparing solutions across K = 5–40 on held-out likelihood, semantic coherence, exclusivity, and residual dispersion, yielding an interpretable 18-topic model whose labels were validated by two independent expert coders (Cohen’s κ = 0.69). Topic prevalence, model-estimated temporal trajectories (estimateEffect with 95% confidence intervals), and a Mann–Kendall trend test with Benjamini–Hochberg correction were computed. The literature is measurement-centred — eye-tracking/pupillometry and EEG/fNIRS workload classification are the most prevalent themes — and thematically diversified across measurement, interaction-modality, and application axes. After correction, only three topics show a significant monotonic trend (health applications rising; multimedia/cognitive-load-theory and computational cognitive modeling declining), whereas the most pronounced change is the abrupt post-2022 emergence of generative-AI interaction (0.76% → 12.41%). Model-based topic correlations are near zero (mean |r| = 0.06), indicating largely independent rather than converging themes. Findings are bounded by single-database (Scopus) and English-only restrictions. The study provides a model-based synthesis that distinguishes measured from inferred patterns and identifies opportunities in multimodal sensing, adaptive interfaces, and workload-metric validation.</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

Mapping the Thematic Landscape of Cognitive Load Research in Human–Computer Interaction: A Structural Topic Modeling Study (2006–2025)

  • 期刊论文

  • 发表日期：

    2026年6月29日

查看全文
  • Betül Türkyılmaz,
  • Ahmet Ayaz,
  • Ekrem Bahçekapılı

摘要

Research on cognitive load and mental workload in human–computer interaction (HCI) spans learning technologies, healthcare interfaces, safety-critical operations, and biosignal-driven user-state assessment, yet the field lacks a consolidated, data-driven account of its thematic structure, its evolution, and the (in)dependence of its subfields. This study maps the thematic landscape of cognitive load–oriented HCI research published between 2006 and 2025 using a Structural Topic Model (STM) of titles, abstracts, and author keywords indexed in Scopus. Following a PRISMA-guided corpus-construction protocol, 1,474 documents (1,024 peer-reviewed conference papers and 450 journal articles) were retained; conference proceedings were deliberately included as a primary archival venue in HCI. The number of topics was selected by comparing solutions across K = 5–40 on held-out likelihood, semantic coherence, exclusivity, and residual dispersion, yielding an interpretable 18-topic model whose labels were validated by two independent expert coders (Cohen’s κ = 0.69). Topic prevalence, model-estimated temporal trajectories (estimateEffect with 95% confidence intervals), and a Mann–Kendall trend test with Benjamini–Hochberg correction were computed. The literature is measurement-centred — eye-tracking/pupillometry and EEG/fNIRS workload classification are the most prevalent themes — and thematically diversified across measurement, interaction-modality, and application axes. After correction, only three topics show a significant monotonic trend (health applications rising; multimedia/cognitive-load-theory and computational cognitive modeling declining), whereas the most pronounced change is the abrupt post-2022 emergence of generative-AI interaction (0.76% → 12.41%). Model-based topic correlations are near zero (mean |r| = 0.06), indicating largely independent rather than converging themes. Findings are bounded by single-database (Scopus) and English-only restrictions. The study provides a model-based synthesis that distinguishes measured from inferred patterns and identifies opportunities in multimodal sensing, adaptive interfaces, and workload-metric validation.