Gold Farming

VWE Gold Farming Project: The Gold Farming project is a sub-project within the larger VWE This project is a collaboration between me and Brian Keegan at Northwestern University. The other PIs from the VWE project are involved in this project as well: Professor Dmitri Williams from USC, Professor Jaideep Srivastava from University of Minnesota and Professor Noshir Contractor from Northwestern University.

What is Gold Farming?

“Gold farming refers to a body of practices that involve the sale of virtual in-game resources for real-world money. The name gold farming stems from a variety of repetitive practices (¨farming¨) to accumulate virtual wealth (¨gold¨) which farmers illicitly sell to other players who lack the time or desire to accumulate their own in-game capital.

History of Gold Farming

While the earliest instances of real money trade can be traced back to the terminal-based multi-user dungeons (MUDs) of the 1970s and 80s, formal gold farming operations originated in an early massively multiplayer online role-playing game, Ultima Online, in 1997. An informal cottage industry of inconsequential scale and scope at first, the practice grew rapidly with the parallel development of an ecommerce infrastructure in the late 1990s. The complexity of gold trading organizations continued to grow as indigenously-developed massively multiplayer games as well as Western-developed games were released into East Asian markets like Japan, South Korea, and China.

Characteristics of Gold Farmers in EQ2

At the end of 2006 there were 2,122,600 unique characters out of which 9,179 were gold farmers, or 0.43% of the population. The stereotype of Gold Farmers being Chinese is borne out by the EQ2 data, 77.6% of all the Gold Farmers are indeed Chinese.

This of course does not mean that all Chinese speaking players are Gold Farmers. This is evident in the next figure which shows that only a minority of Chinese speaking players are Gold Farmers, although the percentage of Gold Farmers amongst Chinese speakers is higher as compared to speakers of other languages.

Chinese speaking players are however a small minority of total number of players in EQ2. The majority of the players in EQ2 are English speaking players.

Gold Farmer Detection

The problem of automatically detecting Gold Farmers can be posed as a machine learning problem. In layman’s terms machine learning models can be thought of as computational models which can detect/learn common patterns in the data and use this information to make predictions. The following information can be used as a feature set to the machine learning models.

• Demographic information: Demographic information about the player in the real-world. This is already anonymized so that it is not possible to link the player back to a real-world person.
• Character game statistics of players: These characteristics are of two types. ¨Demographic¨characteristics of the character like race (human, orc, elf etc), character sex, etc.;
• Cumulative statistics like total number of experience points earned, or number of monsters killed.
• Player activity sequence: Players can perform a wide range of activities within the game. The sequences of activities include but are not limited to mentoring other players, leveling up, killing monsters, completing a recipe for a potion, fighting other players, etc. The following table is an example of such a sequence.

Using the aforemenioned features and applying additional techniques from social network analysis we were able to build models which worked quite well in catching Gold Farmers.

Gold Farming and Drug Trafficking Networks

Note: The main analysis was done by Brian Keegan of Northwestern University. There have been a lot of hits to this page, so if you are linking to this page then please acknowledge Brian Keegan as well. Credit where it is due. If you want to know more about my work then more information is available here.

Since Gold Farming is the quintessential illicit activity in MMORPG, the behaviors exhibited by Gold Farmers in MMORPGs are likely to be similar to behaviors exhibited in criminal networks like Drug Traficking Networks, networks of street gang members etc. In order to study the similarities between online and offline clandestine networks we used data from EQ2 Gold Farming networks (online criminal network) and network data on a drug trafficking ring obtained from a Canadian law enforcement taskforce called Project Caviar (Offline criminal network). We discovered that Gold Farmers camouflage their activity or counterparties by emulating typical players’ trading practices – rely upon many repeated transactions with only trusted (and likely co-offending) characters. This behavior is contrasted by gold farmers’ affiliates who have substantially higher connectivity and transaction frequency than lay players.

We used various centrality measures where a centrality measure computes the relative importance of a node in a network e.g., how many paths in the network pass through that node. Gold farmers and their affiliates had significantly lower closeness centrality scores than the unaffiliated population at large. This suggests farmers and their affiliates generally position themselves on the periphery of the transaction network such that most actors would have to pass through several intermediary characters to trade with them. This is further evidence that although affiliates serve a unique role brokering connections between the farmers and typical characters, they exhibit structural characteristics of both farmers and non-affiliates.

The figure above shows the distribution of in-degree (the number of players who have sold something to this player) and out-degree (the number of player to whom I have sold something) of Gold Farmers, their affiliates and non-affiliates. In case of gold farmers there is a shift above a threshold suggests there are more gold farmers engaging in more intense trading than would be expected by extrapolation. This effect may be an adaptation that promotes operational efficiency, security, or resilience, thus high-frequency trading.

We observe the subset of the affiliate network consisting of only identified farmers exhibits a clear pattern of dissortative mixing that resembles the dissortative pattern observed in the offline Caviar drug trafficking network. The presence of dissortative mixing in both the drug trafficking and gold farming networks is key evidence that behaviors in online, virtual worlds also map onto behaviors found in the offline, real world.

Another way to measure similarity between the two networks to to simulate attacks on these networks. Attacks on the gold farming and affiliate networks were simulated using the random failure, degree attack, and edge attack targeting strategies. Figure 4 shows the results of these attacks on both the fraction of the affiliate network located inside the largest connected component (LCC) as well as the fraction of the affiliate network that are isolates. Notably, the edge attack strategy of extracting nodes with strong link weights performs more poorly than both degree attack and random node failure.

Figures 5 and 6 compare the identified farmer network to the Caviar drug-trafficking network. Both the real world and virtual criminal networks exhibit very similar performance and resilience under degree attack and random failures. Removing fewer than 1% of the nodes by attack keeps the fraction of the network in the LCC (largest connected component) relatively high and the number of isolates in the network relatively low. However, these networks are an order of magnitude more sensitive to node removal than the affiliate networks analyzed in Figure 4; removing approximately 5% of nodes by degree attack cuts the fraction of nodes in the largest connected component below 50% while increasing the fraction of isolates to approximately 50%.

Taken together, this analysis shows the farmer and affiliate networks have substantial resilience to both random failures and determined attacks over several orders of magnitude before fracturing into many disconnected components, a pattern which is also found in a real-world drug trafficking network. The affiliate network composed of farmers, unidentified farmers, and legitimate players exhibits even less sensitivity to attack than the clandestine networks alone. These findings suggest that farmers are able to effectively conceal their interaction patterns against the background of legitimate trade activity which also provides substantial resilience to interdiction.

Summary:

• The common stereotype that most Gold Farmers are Chinese is correct. In fact in EverQuest 2 more than three quarters of all Gold Farmers are Chinese.
• It is possible to identify and construct a set of attributes of Gold Farmers which can be used to build machine learning for automatically detecting Gold Farmers.
• Social Network Analysis plays a crucial role in identifying Gold Farmers.
• Criminal network in the online world (Gold Farming Networks) behave in a manner similar to criminal networks (Drug Trafficking Networks) in the offline world.

Published Work:

Here are the list of my related publications:

• Muhammad Aurangzeb Ahmad, Brain Keegan, Jaideep Srivastava, Dmitri Williams, Noshir Contractor, “Mining for Gold Farmers: Automatic Detection of Deviant Players in MMOGS”  Proceedings of the 2009 IEEE Social Computing (SocialCom-09). Symposium on Social Intelligence and Networking (SIN-09). Vancouver, Canada, August 29-31, 2009.
• Brian Keegan, Muhammad Aurangzeb Ahmad, Dmitri Williams, Jaideep Srivastava, Noshir Contractor, “Dark Gold: Statistical Properties of Clandestine Networks in Massively-Muliplayer Online Games” IEEE Social Computing Conference (SocialCom-10) Minneapolis, MN, USA, August 20-22, 2010.