Semantic Networks as Clues: A Theoretical Foundation and Process Optimization for Semantic Network Construction
2026-08-03 • Social and Information Networks
Social and Information NetworksArtificial Intelligence
AI summaryⓘ
The authors study a special kind of Semantic Networks that represent clues about knowledge from texts, rather than exact facts. They explain why this approach makes it hard to have a perfect standard for accuracy, but still has scientific value through a reasoning method called abduction. The paper reviews key steps in building these networks automatically and suggests ways to evaluate how well they work. Finally, the authors propose a framework named ClueNetwork to rank and optimize these Semantic Networks based on combined evaluation criteria.
Semantic NetworksNon-propositional KnowledgeAbductionAutomatic Keyphrase ExtractionEdge WeightingCommunity DetectionProcess Optimization ProblemClueNetwork
Authors
JinWoo Ha, Dongsoo Kim
Abstract
The subject matter of this paper is twofold. One is to review the theoretical foundation of a specific type of Semantic Networks (SNs) representing textual non-propositional knowledge. The other involves proposing a framework (ClueNetwork) for ranking candidate SNs generated through various Semantic Network Construction (SNC) processes for the type. In the first fold, it is clarified that the type serves as clues, not surrogates, of reality, making gold standards elusive. Then, it is discussed why this type nevertheless holds scientific legitimacy in terms of abduction. Grounded in this legitimacy, the three main stages of SNC, comprising Automatic Keyphrase Extraction (AKE), Edge Weighting (EW), and Community Detection (CD), are reviewed alongside their objectives and operations. In the second fold, evaluation criteria (comprising two established and one reformulated) for achieving the objectives are first defined and justified, followed by illustrative experiments based on the criteria. Thereafter, SNC is reformulated as a Process Optimization Problem (POP), and its global objective function that integrates the local criteria is defined and justified. Based on these, ClueNetwork is ultimately proposed.