Cashflow Tracing: Detecting Online game bots leveraging financial analysis with Recurrent Neural Networks

Abstract

Since game bots and Gold Farmer Group (GFG) create critical damage toward the ecosystem of MMORPGs, game companies have struggled to detect bot characters with various methods. Primarily, previous researches identified GFGs by analyzing particular behavior patterns of bots, but these methods have become easier to be neutralized as bots started to mimic normal characters. Moreover, the spread of mobile MMORPGs decreased the diversity of character behavior; thus, classification of behavior patterns between bots and users becomes a more challenging task. To address this problem, we propose a bot detection method which is generally applicable toward modern MMORPGs in both PC and mobile environment. We focused on the analogy that bot characters and normal characters show different patterns of financial activities. As game bots are born to collect game assets for Real Money Trade (RMT), they show patterned changes in financial status to maximize its efficiency. On the other hand, normal characters take various types of financial activity as users play various in-game contents, not only accumulate the asset. Throughout the study, our series of analysis propose contributions as follow. First, we clarified financial sequences of game bots are different from normal characters; therefore, the sequential form of financial features precisely describes the financial pattern of characters. Second, we established a bot detection model with Recurrent Neural Networks (RNN) trained with the aforementioned financial sequences. With the real-world log data extracted from three PC games (Lineage, Aion, Blade and Soul), and one mobile game (Lineage M), we validated the proposed detection model effectively identifies game bots from normal users. Lastly, our detection model is widely applicable in both PC MMORPG and mobile MMORPG.