偵測線上遊戲作弊檢舉系統中的毀謗行為

Abstract

近來「大型多人線上角色扮演遊戲」十分受歡迎，然而傳統線上遊戲所面臨之玩家作弊行為一樣存在於此類遊戲，並且持續地增加當中，其中最常見的作弊方式即是遊戲自動程式的使用。要讓遊戲本身能自動偵測遊戲自動程式十分困難，因此一些遊戲公司嘗試使用投票機制來防範作弊，玩家可於遊戲中藉由投票的方式直接檢舉行為可疑的玩家。使用這樣的投票系統，隨之而來的是一個十分重要且待解決的議題---惡意共謀毀謗。惡意玩家在共謀後蓄意的投票檢舉毀謗其他合法玩家，使得投票機制無法公平正確地運作。 在此論文中，我們提出基於投票歷史紀錄（包括投票者與被投票者之間的關係）的兩種偵測機制，用來找出毀謗合法玩家的單個或多個惡意玩家群。機制一是從投票者的角度出發，直接地偵測出惡意玩家群；機制二則是從被投票者的角度找出受害的正當玩家群，再間接地找到惡意玩家群。我們經由模擬來顯現這兩種機制的偵測能力，實驗顯示當惡意玩家群的毀謗次數達到三次與五次時，本文所提之機制其偵測正確率分別可以超過83%與97%。最後，經由一系列於不同遊戲環境之實驗顯示，我們所提之判斷機制皆能適應並且仍然保有良好的偵測效能。

In recent years, MMORPGs have become extremely popular; however, the cheating seen in traditional online games has also been increasing in MMORPG. One of the most common forms of cheating in MMORPG is the increased use of game bots, i.e., auto-playing game clients. Because of the difficulty of automatically detecting a game bot, some game companies have begun to use voting schemes, in which each player can vote that another suspicious player is a game bot. However, in a voting-based system, collusion, e.g., unfairly low ratings ('bad-mouthing'), is a critical concern to be resolved. In this work, we propose two mechanisms based on the vote history of the voter and the votee to detect a collusion cluster whether it be a single cluster or multiple clusters: 1) voter-based collusion cluster detection, and 2) votee-voter-based collusion cluster detection. Detection accuracy can exceed 83% and 97% when the collusion cluster bad-mouths three times and five times, respectively. In addition, we evaluate our schemes under different environments, and high quality performance is maintained.