Incorporating Temporally Decaying Source Profiles for Three-Dimensional Uncertainty-Aware Trustworthy Perception
Advances in Engineering Technology Research · 15.07.2026 · 17 (1) · 657
DOI: 10.56028/aetr.17.1.657.2026 ↗Kurz gefasst
Fachartikel in der Zeitschrift Advances in Engineering Technology Research, online erschienen am 15.07.2026. Verfasst von Hanyu Li und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Credibility-aware perception is a core process that jointly models source reliability, perceptual model limitations, and information timeliness in open information environments. It is of critical significance for enhancing the judgment accuracy and decision confidence of intelligence analysis systems operating in highly dynamic and adversarial scenarios. Addressing the limitations of existing approaches that treat source credibility as a static attribute independent of the temporal dimension and consequently fail to track the temporal evolution trajectories of sources' historical verification behaviors, this paper proposes a three-dimensional uncertainty credibility-aware method that incorporates temporal decay source profiling. We construct a Temporal Decay Source Credibility Profiling mechanism (TD-SCP), which, inspired by the Ebbinghaus forgetting curve, models the sequence of a source's historical verification events as a temporally accumulated process with exponential decay weights. And, we propose a Three-Dimensional Uncertainty Tensor (3D-UT), which orthogonally decomposes perception-layer uncertainty into three physically independent dimensions: source uncertainty , extraction uncertainty , and temporal uncertainty . These three dimensions are quantified via linear transformation of dynamic credibility scores, MC Dropout-based mutual information decomposition, and domain-sensitive exponential decay functions, respectively. By replacing conventional scalar confidence scores with a structured uncertainty triplet, the 3D-UT establishes a semantically well-defined credibility propagation interface. We further design a pipeline fusion mechanism that integrates TD-SCP and 3D-UT by directly converting dynamic credibility scores into the computational input for the source uncertainty dimension, thereby forming a directed unidirectional causal data flow. The mechanism also standardizes the interface protocol governing the coupling of the uncertainty triplet with differentiated strategies in the downstream Bayesian inference layer, achieving deep synergy between source prior modeling and perceptual uncertainty decomposition while maintaining constant-order online processing complexity. Experimental results demonstrate that TD-SCP achieves a Pearson correlation coefficient of 0.721 on the FakeNewsNet dataset, representing a 58.1% improvement over the static mean baseline. The optimal decay rate parameters for the political domain and the science-and-health domain are 0.080 (half-life ≈ 8.7 days) and 0.008 (half-life ≈ 86.6 days), respectively; this order-of-magnitude divergence in parameters is highly consistent with the information timeliness characteristics inherent to each domain. The proposed method provides a novel technical pathway for the dynamic modeling of source credibility and the multi-dimensional interpretable representation of perceptual uncertainty in intelligence analysis systems, with broad application prospects in domains such as online public opinion monitoring, multi-source intelligence verification, and security situational awareness.
Zitieren
APA
Li, H., Wang, M., & Tan, Z. (2026). Incorporating Temporally Decaying Source Profiles for Three-Dimensional Uncertainty-Aware Trustworthy Perception. Advances in Engineering Technology Research, 17(1), 657. https://doi.org/10.56028/aetr.17.1.657.2026
Vancouver
Li H, Wang M, Tan Z. Incorporating Temporally Decaying Source Profiles for Three-Dimensional Uncertainty-Aware Trustworthy Perception. Advances in Engineering Technology Research. 2026;17(1):657. doi:10.56028/aetr.17.1.657.2026
BibTeX
@article{Li2026,
author = {Li, Hanyu and Wang, Meng and Tan, Zhizhi},
title = {Incorporating Temporally Decaying Source Profiles for Three-Dimensional Uncertainty-Aware Trustworthy Perception},
journal = {Advances in Engineering Technology Research},
year = {2026},
volume = {17},
number = {1},
pages = {657},
doi = {10.56028/aetr.17.1.657.2026}
}
EUVIMED