Year: 2025
Type: article
Abstract: Purpose. The purpose of this study is to evaluate the effectiveness of applying factorization-based abduction using a graph network while taking into account heuristic approaches, computational complexity, and efficiency in the context of solving content classification problems. Methodology. The methodology is based on the analysis and comparison of different approaches to abduction with cost coefficients and their application in the context of graph networks. Special attention is given to the theoretical foundations of abduction complexity and attempts to identify conditions under which the problem can be solved in polynomial time, meaning that the algorithm or process executes within a time estimated as a polynomial function of the input size. To achieve this, linear programming methods are employed, particularly transforming the original problem into a linear constraint equation to search for minimum-cost proofs. Conditions for the existence of polynomial solutions for abduction problems with cost coefficients are examined, and techniques for searching suboptimal solutions using appropriate algorithms are applied. Special emphasis is placed on logical inference in the “Prolog” style with an added factorization operation, which allows generating all non-contradictory explanations in the context of consistency with a knowledge base. Results. The study develops and implements conceptual frameworks for inductive and deductive approaches to classification tasks, particularly considering constraints and probabilistic hypothesis assessments. It focuses on algorithmic analysis aimed at finding minimal proofs through linear programming, inference using graph networks, rule clustering based on metrics and heuristics, and applying abductive logic for classification with incomplete and probabilistic data. As a result, the methodology enables the formulation of approaches to improve the accuracy and efficiency of automated classification methods under multi-constraint conditions in complex information systems. Scientific novelty. The scientific novelty of the proposed research lies in the development and verification of new abduction methods with cost coefficients, the implementation of graph networks to enhance classification processes, and the generation of multi-variant logical inference models for problems involving incomplete and probabilistic data. Practical significance. The practical significance of the proposed research is in increasing the efficiency, reliability, and accuracy of automated classification systems and decision-making processes across various industrial domains, especially in the fields of generative artificial intelligence, machine learning, big data analytics, and information systems. The implementation of factorization-based abduction methods through graph networks expands the ability to create and integrate more reliable and flexible models capable of interacting effectively with incomplete, imprecise, and uncertain data. Consequently, the approach enables the further development of more adaptive and intelligent systems applicable in advanced technological scenarios. Keywords: abduction, graph networks, model, optimization, clustering, machine learning.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7117139321
