Graph Neural Networks for Missing Value Classification in a Task-driven Metric Space

Huang, Buliao; Zhu, Yunhui; Usman, Muhammad; Zhou, Xiren; Chen, Huanhuan

doi:10.1109/tkde.2022.3198689

Cited by 10 publications

(1 citation statement)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Event extraction, a pivotal task in Natural Language Processing (NLP), involves discerning and categorizing events, entities, and their interrelations within textual data [15]. The task encompasses entity extraction [16], event identification, argument role assignment, and relationship extraction [17,18,19], each presenting its unique challenges and complexities. Various methods have been explored to navigate through these complexities: 1) Rule-Based Methods: Utilize predefined linguistic rules and patterns, such as regular expressions and syntactic patterns, to identify and extract events and entities, often facing challenges in scalability and adaptability across different domains and contexts [20,21,22].…”

Section: Event Extraction: Task and Solutionmentioning

confidence: 99%

An Ensemble Event Extraction Method on News

Liu,

Chen,

et al. 2024

Preprint

View full text Add to dashboard Cite

Navigating through the complexities of news to extract pivotal events demands innovative methodologies to ensure accurate and reliable outcomes. This paper introduces a novel ensemble approach for event extraction, synergizing multiple models to enhance accuracy and reliability amidst the strategic and operational intricacies of news. A key metric, the Confidence Score (CS), quantifies the reliability and credibility of extracted events, serving as a robust indicator of a model's assurance in its predictions and providing a mechanism to navigate through ambiguities in the data. The paper offers a holistic representation and estimation of the confidence of individual model outputs. A comprehensive CS evaluation method is introduced, integrating consistency and prior confidence of multi-model output, and is augmented by an iterative update algorithm, enhancing CS evaluation accuracy by comparing consistency of pairwise outputs of identical arguments and updating the CS in a weighted manner based on prior scores.Experiments conducted on a dataset compiled from online news underscore the notable improvement of our method in comparison to both singular model and prevalent ensemble strategies, demonstrating its efficacy and potential applicability in practice.

show abstract