Protecting Digital Game Integrity : Exploring Similarity Detection Methods to Ensure Authenticity for Small Game Platforms

Sammanfattning

In a world of computers, data integrity is of utmost importance. To know where the information published online truly originates is becoming an increasingly difficult task. One place where this may be a problem is in the gaming world, where piracy is a common issue. In this master's thesis, the work is performed on a platform for small games, where there is a potential risk that games could be copied, modified, and republished under another author's name, raising concerns about preserving the authenticity of game authorship.To determine if a copy has been made, this report explores pairwise similarity detection methods to ensure authenticity. Similarity metrics were constructed from raw game data, game logic, and graphical appearance using fuzzy hashing, Abstract Syntax Trees with the Zhang–Shasha algorithm, and color histograms compared using cosine distance and earth mover's distance.The methods were evaluated using four datasets: popular game groups, near-copy games, manually selected distinct games, and 100 unlabeled games. Pairwise similarity metrics generally distinguished games within the same group from those outside it. Combined pairwise similarity measures analyzed using Kernel Principal Component Analysis and Multi-Dimensional Scaling preserved meaningful group structure and successfully detected near-copy games. Kernel Principal Component Analysis on the unlabeled dataset showed short distances between games of similar level design and color scheme.The main contribution of this thesis is the demonstration that combined feature-based similarity metrics can capture meaningful relationships between games and show promise as a tool for detecting copied games, particularly near-copy variants.