A Variability-Based Framework for Interpretable Naming in Formal and Relational Concept Analysis

Researchers propose a variability-based framework for automatically naming concepts generated by Formal Concept Analysis (FCA) and Relational Concept Analysis (RCA) using large language models. The framework addresses the challenge of translating formally-defined but opaque symbolic abstractions into human-readable names by controlling which information sources (intent, extent, implications, relations) are exposed during naming, making semantic choices explicit and interpretable.

This academic paper addresses a fundamental gap in knowledge extraction systems: formal structures generated by FCA and RCA are mathematically rigorous but often assigned meaningless technical labels that prevent domain experts from using them effectively. The researchers recognize that naming is not a trivial post-processing task but a critical interface between formal systems and human understanding, involving complex linguistic choices around ambiguity, discrimination, and consistency.

The proposed variability model represents an elegant solution by parameterizing which information sources influence LLM-generated names. Rather than treating naming as a black-box operation, the framework exposes semantic choices explicitly—allowing practitioners to understand how different configurations produce different interpretations of the same formal structure. This transparency is particularly valuable because it can surface modeling errors or unintended relational dependencies in the underlying data.

For the knowledge engineering and AI communities, this work bridges symbolic AI and modern LLM capabilities without abandoning interpretability. The pizzeria domain proof-of-concept demonstrates practical applicability, though the real impact will emerge when the framework scales to complex enterprise datasets. The ability to consistently generate meaningful names across related concepts has immediate applications in ontology engineering, data governance, and knowledge graph construction.

Future development should focus on evaluation metrics for name quality, integration with existing knowledge management systems, and testing across diverse domains. The framework's parametric design allows incremental adoption—organizations could begin with basic configurations and refine naming strategies as domain expertise accumulates.

• →A variability model framework enables controllable LLM-assisted concept naming by exposing different information sources during generation.
• →The approach makes explicit the semantic and interpretation choices involved in translating formal abstractions to human-readable names.
• →Naming variability can reveal modeling issues and unintended relational dependencies in symbolic datasets.
• →The framework bridges symbolic AI systems with modern LLMs while preserving interpretability and transparency.
• →This addresses a significant bottleneck in knowledge extraction adoption by domain experts who need meaningful, not technical, concept labels.