2026/07/01 by Randa Cheima Bendjeddou, Giacomo Kahn, Aicha Sekhari Seklouli +1
Computer Science · #Bayesian Modeling and Causal Inference #Data Mining Algorithms and Applications #Rough Sets and Fuzzy Logic
paper · doi:10.1016/j.ijar.2026.109782
openalex publication_date 2026/07/01 · openalex created_date 2026/07/26 · openalex updated_date 2026/07/28
The continuous growth of complex and heterogeneous data in domains such as healthcare, ecology, and industry has increased the need for knowledge discovery methods capable of handling rich relational structures. Formal Concept Analysis (FCA) provides a mathematical framework for concept extraction and lattice-based data exploration, whilst Relational Concept Analysis (RCA) extends FCA to multi-relational datasets by modelling binary relations between multiple contexts; it nevertheless remains intrinsically limited to binary relations. When higher-arity relations arise, existing RCA-based approaches typically rely on encoding strategies or Cartesian-product constructions, which may lead to information loss, spurious associations, reduced interpretability, or combinatorial explosion. To address this limitation, we propose n-Relational Concept Analysis ( n -RCA), a novel extension of RCA that directly supports n -ary relations between multiple relational contexts. By combining the core principles of RCA with key concepts from Polyadic Concept Analysis, n -RCA directly models higher-arity relations whilst remaining faithful to the conceptual foundations of FCA. First, we introduce the formal framework of n -RCA and an iterative concept construction algorithm enabling the direct modelling and mining of n -ary relations across contexts. Then, we formally prove that n -RCA generalises classical RCA and preserves its behaviour in the binary case. Finally, we evaluate the effectiveness and practical relevance of n -RCA through detailed synthetic examples and a real-world case study on the Knomana dataset, involving both binary and ternary relations. The results demonstrate that the proposed method efficiently produces expressive and interpretable concepts whilst remaining comparable to existing RCA and graph-based approaches in terms of concept quality and runtime performance. Overall, n -RCA provides a semantically faithful and theoretically grounded framework for conceptual knowledge discovery in complex multi-relational datasets, bridging the gap between binary relational analysis and the direct modelling of n -ary relations.