.png){kind=icon}
.png){kind=icon}
Return
Graph-based document-level relationship extraction for risk analysis: A transitive and dialog coherence approach
.png){kind=icon}
OA
AI
DOI：10.1016/j.eswa.2024.124990.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper proposes a solution to extracting relationships at the document level in the context of risk analysis. It addresses the problem of identifying the flow of hazard's impact in the system's description or the description of the consequences of its failure. The problem is challenging, as information on impact may form complex, interwoven relations distributed across many sentences within a description and across many sources. The proposed approach involves a stepwise decomposition of the descriptions: first into a Semantic Frames Graph (SFG) to detect risk-relevant relationships, then into the Intermediate Relationship Graph (IRG), which is built upon detected relations, and finally, the aggregation of risk interaction represented in Asset-Vulnerability-Hazard (A-V-H) graph. This approach allows for the modeling of risk interactions without needing a dedicated training set, as the authors present a method for relationship detection based on the verbalization of transitive relationships and dialog consistency validated through prompt engineering over the generative language model. Overall, this research provides insights into a novel approach to acquiring risk interactions using document-level relationship extraction. It demonstrates its potential in graph-based representations and transitive relationships to understand complex risk interactions.
Keywords:
Risk analysis
Relationship extraction
Graph representation
Language models
Knowledge acquisition
Semantic networks
Knowledge graphs
Entity recognition
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W
Organization
No organization information available