A hybrid recommendation system for forming balanced tearms in multi-user online environments

DOI: 10.31673/2412-9070.2026.318106

Abstract

The rapid growth of competitive e-sports environments increases the need for personalized systems capable of forming balanced teams. Traditional recommendation approaches relying on single-criterion optimization fail to capture complex interactions between user characteristics, role preferences, and equipment parameters. Furthermore, existing solutions lack mechanisms to handle distributed access to shared digital resources, such as collective gaming accounts, leading to resource collisions and reduced team performance. To address these challenges, the proposed hybrid system combines content-based analysis, collaborative filtering, and metaheuristic optimization. The logical model is integrated into a microservice architecture built with Java and Spring Boot for clan management in War Thunder. The system's core is a mathematical model based on the Assignment Problem, extended with strict mutual exclusion (Mutex) constraints. This ensures the prevention of duplicated player assignments and eliminates concurrent access conflicts to shared accounts. Team composition optimization is executed by maximizing a hybrid fitness function using the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm. The function evaluates an objective vector (combat effectiveness, team synergy) and a subjective vector (individual player skills, role preferences). Simulation modeling on a dataset of 30 players, 100 accounts, and defined role quotas demonstrated the superiority of the proposed approach. Compared to a Greedy Search algorithm, which converged to a lo-cal extremum at 62% efficiency, the MOPSO-based model reached a stable Pareto-optimal front with 89% efficiency within 45 iterations. Mutex constraints completely eliminated resource access collisions, while multi-criteria analysis provided a 34% higher match rate between players and preferred roles. With an execution time of 140 ms, the system confirms its scalability and suitability for real-time matchmaking.

Keywords: recommender systems, hybrid models, machine learning, team formation, personalization, multi-criteria optimization, content-based analysis, collaborative filtering.