A Comparative Study of Counterfactual Explainers for Graph Neural Networks Enabling Multiple Types of Graph Edit
研究对比六种最新模型在图神经网络中的反事实解释方法,旨在通过添加或删除边来最小化修改以改变预测结果,评估其性能以指导未来研究。
研究对比六种最新模型在图神经网络中的反事实解释方法,旨在通过添加或删除边来最小化修改以改变预测结果,评估其性能以指导未来研究。
本文针对含有结构性零值的组合数据,提出了一种条件对数正态分布模型,并通过EM算法实现快速计算。
研究通过排除年龄因素,使用声学模型在COPD筛查中分离出非年龄相关的声学信号,以解决因年龄相关的声音变化对COPD筛查结果的影响。
This study addresses the scarcity of structured computational resources for non-standard Greek slang—a linguistic variety hindered by its dynamic and unregulated nature, which impedes both linguistic inquiry and NLP applications. To bridge this gap, the authors present slang.gr, the first large-scale computational resource for Greek slang, integrating lexical entries, user-generated tags, and interaction data. They introduce a novel multi-layer taxonomy that uniquely combines semantic and sociolinguistic metadata. Through folksonomy-based tag cleaning, ontology mapping, community behavior modeling, and a credibility-weighted scoring algorithm, the work reveals that Greek slang is highly oriented toward person- and evaluation-related expressions and exhibits strong morphological innovativeness. The proposed taxonomy significantly enhances analytical interpretability and establishes a foundational framework for computational modeling of non-standard language varieties.
This study addresses the lack of systematic, real-data-based evaluation among existing Non-negative Matrix Factorization (NMF) implementations in R, which hinders informed package selection by users. To resolve this gap, we introduce nnmf, a new R package for NMF, and present the first comprehensive benchmark comparing nnmf against two widely used NMF packages within a unified experimental framework. The evaluation leverages real-world datasets and assesses performance across multiple dimensions—including computational efficiency, convergence behavior, reconstruction accuracy, memory consumption, and numerical stability. Our results clearly delineate the relative strengths and weaknesses of each implementation and demonstrate that nnmf achieves superior overall performance, thereby offering practitioners a reliable basis for selecting an appropriate NMF tool for real-world applications.
研究对比六种最新模型在图神经网络中的反事实解释方法,旨在通过添加或删除边来最小化修改以改变预测结果,评估其性能以指导未来研究。
本文针对含有结构性零值的组合数据,提出了一种条件对数正态分布模型,并通过EM算法实现快速计算。
研究通过排除年龄因素,使用声学模型在COPD筛查中分离出非年龄相关的声学信号,以解决因年龄相关的声音变化对COPD筛查结果的影响。
This study addresses the scarcity of structured computational resources for non-standard Greek slang—a linguistic variety hindered by its dynamic and unregulated nature, which impedes both linguistic inquiry and NLP applications. To bridge this gap, the authors present slang.gr, the first large-scale computational resource for Greek slang, integrating lexical entries, user-generated tags, and interaction data. They introduce a novel multi-layer taxonomy that uniquely combines semantic and sociolinguistic metadata. Through folksonomy-based tag cleaning, ontology mapping, community behavior modeling, and a credibility-weighted scoring algorithm, the work reveals that Greek slang is highly oriented toward person- and evaluation-related expressions and exhibits strong morphological innovativeness. The proposed taxonomy significantly enhances analytical interpretability and establishes a foundational framework for computational modeling of non-standard language varieties.
This study addresses the lack of systematic, real-data-based evaluation among existing Non-negative Matrix Factorization (NMF) implementations in R, which hinders informed package selection by users. To resolve this gap, we introduce nnmf, a new R package for NMF, and present the first comprehensive benchmark comparing nnmf against two widely used NMF packages within a unified experimental framework. The evaluation leverages real-world datasets and assesses performance across multiple dimensions—including computational efficiency, convergence behavior, reconstruction accuracy, memory consumption, and numerical stability. Our results clearly delineate the relative strengths and weaknesses of each implementation and demonstrate that nnmf achieves superior overall performance, thereby offering practitioners a reliable basis for selecting an appropriate NMF tool for real-world applications.