Аңдатпа
Knowledge extraction from complex textual and semi-structured data requires methods that combine contextual language understanding with explicit constraint validation. This paper presents a neuro-symbolic approach to knowledge extraction using NP-complete subset sum formulations as a constraint-solving mechanism. The proposed framework integrates neural models for identifying and ranking relevant textual evidence with symbolic procedures that verify structural consistency and admissible combinations under predefined constraints. In this setting, candidate facts, attributes, and decision conditions are represented as constrained selection structures, allowing subset sum models to support interpretable and formally grounded validation within the extraction pipeline. The paper describes the conceptual basis of the framework, its architectural design, and its applicability to information extraction, constraint-aware text analysis, and structured decision support. The study shows that subset sum formulations can serve as an effective bridge between neural inference and symbolic reasoning, providing a promising foundation for more interpretable and robust neuro-symbolic knowledge extraction systems. In the context of sustainable development, the proposed framework can support more reliable digital decision-support systems for domains such as public administration, legal services, social policy analysis, environmental monitoring, and resource allocation. By combining neural extraction with explicit constraint validation, the approach helps reduce inconsistent automated decisions and improves transparency in data-driven systems used for sustainability-oriented tasks.