SPIKE 2026 Abstracts

Area 1 - eSports Performance, Artificial Intelligence and Knowledge in Esports
Full Papers Short Papers

Area 1 - eSports Performance, Artificial Intelligence and Knowledge in Esports

Full Papers
Paper Nr: 8
Title:

Formal Concept Analysis as a Biclustering Approach for Champion Drafting in eSports
Authors:

Anthony Feudjio, Malika Charrad and Mondher Maddouri
Abstract: Data-driven methods are increasingly used to support strategic decision-making in competitive esports, yet champion drafting in League of Legends still relies largely on empirical knowledge and subjective judgment. This work proposes a Formal Concept Analysis (FCA) framework to support champion selection at the start of a match and to identify suitable replacements when a desired pick is unavailable due to bans or draft constraints. Champion characteristics are encoded in a binary formal context derived from aggregated in-game statistics describing offensive, defensive, and utility aspects of gameplay. Formal concepts extracted from this context define groups of champions that share similar strategic profiles, and the resulting concept lattice enables structured exploration of champion similarities as well as generalization and specialization of strategic intents. For initial picks, the framework recommends champions whose attribute sets match a targeted strategic profile, while in replacement scenarios it selects alternatives from the same or closely related concepts, thereby preserving team-level strategic coherence. Experiments on real League of Legends data show that the FCA-based recommendations are interpretable and consistent, underscoring the potential of FCA as an effective decision-support tool for draft-phase strategy in esports.
Paper Nr: 9
Title:

Mutual and Global Synergy in League of Legends: An Attributed Graph-Based Study of Champions Interactions in LoL
Authors:

Hugo Hemery, Malika Charrad, Manuel Meireles Carvalho, Faten Chakchouk and Sébastien Ricciardi
Abstract: Drafting the right team in League of Legends is a huge strategic puzzle that goes way beyond just individual mechanical skill. Most analysis tools today only look at basic statistics like win rates, failing to see how champions actually work together in a team. In this paper, we propose a graph-based framework to fix this by quantifying these complex interactions. Our model uses two distinct weighted networks to map out both intra-team synergies and inter-team counter-picks. Using an unsupervised method called Deep Graph Infomax (DGI), we organized champions into a latent embedding space based on their characteristics and their relations rather than just their roles. We successfully identified key strategic clusters, such as ”Front-to-back” or ”Dive” compositions, which follow a classic Rock-Paper-Scissors balance. This provides coaches with an automated way to manage the draft’s complexity and build much more coherent team rosters, where he can detect the goal of the enemy composition in real-time, and know better how to counter it.
Short Papers
Paper Nr: 5
Title:

Multi-Modal Highlight Detection in Broadcast Audio: A Deep Learning Approach for Event Recognition in Sports and eSports
Authors:

Nuno Costa, António Oliveira, Armindo Lobo, Ricardo Teixeira, Duarte Fernandes, Ricardo Rodrigues and Emanuel Gouveia
Abstract: The detection of highlights in broadcast streams is essential for enhancing User Experience (UX) through automated summaries and efficient content retrieval. This is particularly relevant for live streaming environments common in sports and eSports, where audiences demand near real-time analysis. This paper presents a benchmark of models for highlight detection in broadcast audio, validated on the SoccerNet dataset but applicable to general competitive gaming streams. We propose a novel multi-modal architecture combining high-level semantic audio features (YAMNet) with Natural Language Processing (NLP) of transcribed commentary (analogous to eSports shoutcasting). Results show that fusing audio event detection with semantic text analysis significantly outperforms uni-modal baselines. The proposed framework offers a computationally efficient solution for AI-based broadcasting technologies, enabling scalable automation for content creators and improved viewer experiences.
Paper Nr: 6
Title:

Optimizing Sim Racing Performance Using Machine Learning and Evolutionary Algorithms
Authors:

Fazilat Hojaji, Adam Toth and Mark Campbell
Abstract: The ideal racing line is a key determinant of lap-time performance, defining the trajectory that minimizes time while maximizing vehicle stability and control. Despite the rise of esports, methods for systematically analysing and optimizing driver performance in sim racing remain limited. This study demonstrates how the integration of machine learning (ML) and evolutionary algorithms (EA) can identify critical telemetry metrics and guide performance optimization. Using a professional racing simulator and MoTeC i2 Pro, telemetry data were collected from 135 participants who completed 1,180 laps on the Laguna Seca circuit in Assetto Corsa Competizione (v1.9). Laps were clustered by performance, and a hybrid feature-selection approach combining correlation analyses and ML models identified the top metrics predictive of lap times, including speed, trail-braking duration, steering angle, oversteer, and lane deviation. These metrics were incorporated into an EA fitness function to optimize sector-level KPIs and generate idealized laps. The EA converged rapidly, achieving substantial reductions in predicted lap times within the first 50 generations and producing smoother, more stable, and faster laps than the human best. Lane deviation, oversteer, trail braking duration, and longitudinal acceleration showed the largest improvements. The EA-optimized laps offer actionable insights for high-performance driving, demonstrating measurable gains in speed, control, and stability, and providing practical guidance for driver coaching and performance enhancement in sim racing.