Towards structure-aware AI: modeling and analyzing directed balanced cliques in signed graphs

Author First name, Last name, Institution

• Muhammad Naeem, Ghulam Ishaq Khan Institute of Engineering Sciences and Technology
• Abdallah Tubaishat, Zayed University, Abu Dhabi CampusFollow
• Zahid Halim, National Yunlin University of Science and Technology
• Stefano Cirillo, Università degli Studi di Salerno
• Fawaz Alarfaj, King Faisal University
• Imad Rida, BMBI – Biomécanique et Bioingénierie
• Sajid Anwar, Institute of Management Sciences

Document Type

Article

Source of Publication

Information Sciences

Publication Date

12-5-2026

Abstract

Graphs are foundational to many Artificial Intelligence (AI) techniques, serving as powerful representations of relationships, dependencies, and interactions among entities. Within this paradigm, understanding dense cohesive subgroups provides deeper insights into the structural and functional organization of intelligent systems. Traditional studies have primarily focused on discovering balanced cliques in signed graphs to capture community cohesion. However, real-world AI-driven networks, such as social, informational, and technological systems, often involve directional and signed interactions, where both the edge polarity and direction convey essential semantic meaning. This work introduces a novel graph-based AI framework for modeling and enumerating directed balanced cliques. We formally define this concept to jointly consider the direction and sign of the edges, thus enabling richer structural reasoning in directed signed networks. Our approach incorporates vertex and edge reduction rules that exploit in- and out-neighbor relationships to efficiently prune the search space, overcoming the limitations of existing unsigned/undirected frameworks. Comprehensive experiments were conducted on real-world datasets spanning domains such as signed social networks and online dispute resolution. Results obtained demonstrate that our proposal outperforms traditional signed graph approaches that neglect edge directionality. The proposed framework achieves up to an order-of-magnitude speedup while producing cliques that more accurately reflect directed signed cohesion. This work contributes to graph-based artificial intelligence by advancing computational models capable of capturing both directionality and polarity in relational data. It opens new pathways for interpretable, structure-aware AI systems that leverage graph theory for deeper network understanding.

DOI Link

10.1016/j.ins.2026.123905

ISSN

0020-0255

Publisher

Elsevier BV

Volume

757

Disciplines

Computer Sciences

Keywords

Balanced cliques, Community detection, Directed signed graphs, Graph-based artificial intelligence, Network cohesion, Structural reasoning

Scopus ID

105044795802

Recommended Citation

Naeem, Muhammad; Tubaishat, Abdallah; Halim, Zahid; Cirillo, Stefano; Alarfaj, Fawaz; Rida, Imad; and Anwar, Sajid, "Towards structure-aware AI: modeling and analyzing directed balanced cliques in signed graphs" (2026). All Works. 8090.
https://zuscholars.zu.ac.ae/works/8090

Indexed in Scopus

yes

Open Access

no