Back to Top

Steam-Player-Preference Analyzer and the AIR Status Performance Classifier

In this work, we investigate how people exhibit and construct forms of self-expression in virtual environments including computational systems such as online social networks, or videogames. For example, in everyday life people dress in certain ways to reflect their individual senses of fashion, thereby expressing their social and personal knowledge regarding clothing. However, looking at a large number of people, distinctive categories may become apparent such as “formal,” “business casual,” or “leisurewear.” Such identity-related phenomena take place in computational systems as well. In representing oneself in computational systems, certain aspects of one's identity, including preferences and knowledge, are imparted. By comparing and contrasting these representations between the different computational systems, we may begin to understand how these systems support, or hinder, the user in terms of representing themselves adequately. More importantly, we may begin to identify, and model, phenomena that exists within the real world computationally too, enabling us as developers and designers to understand the consequences and implications of choices made in the development and design of such systems.

Application Domain

As a domain of application, we have made use the social network within the game distribution platform Steam, and the commercially successful, free-to-play, multi-player first person shooter Team Fortress 2 (TF2), both developed by Valve Inc..

System Development

ICE Lab Ph.D. Student, Chong-U Lim, and his supervisor, ICE Lab Director Professor D. Fox Harrell, have developed two systems: 1) The Steam-Player-Preference Analyzer (Steam-PPA) system, a data-mining and processing system used to collect and aggregate publicly available data from Steam, TF2, and additional community websites; and 2) The Advanced Identity Representation Status Performance Classifier (AIR-SPC), a artificial intelligence system for data mining and analysis using machine learning techniques. Using these systems, we highlight a correlation between a players social network usage and the exhibited preferences for in-game avatar customization.

Status Performance

We have introduced the notion of status performance in games as the real-world monetary value that can be assigned to a players' choices in avatar customization. It takes into consideration both the exhibited choices (actively equipped items), and stored choices (inventory collected items). Using Steam-PPA, we have the capability to aggregate and calculate the real-world monetary values of TF2 virtual items. As an initial case study, we focused on hats, an extremely popular type of virtual accessory that has contributed significantly to TF2's popularity and economical success. Steam-PPA is able to calculate a user's equipped, and inventory, hat values, assigning a quantitative measure of the status performance of each individual player. Next, using our machine learning AIR-SPC system, we exhibited the capabilities of discovering implicit clusters of players based on their status performance.

Social Status/Meta-Ties

We have also introduced a means of analyzing a player's Steam social networking profile. . These abstract features are based upon "tie-strength", which is a measure of the relationship between people. Using these tie strength variables on Steam profiles, our AIR-SPC system identifies important features that describe each player's social network interactions and structures, forming we term social status or meta-ties. This method allows us to describe players according to more easily-comprehensible rules governing their behaviors on their social networks. For example, we discern between players who engage heavily in interactions only when common friends, applications, or groups are involved, compared to users who use the system mainly for trading items.

Model Construction and Prediction

Using AIR-SPC, we construct a model of the dataset of players using the machine learning technique with support vector machines. The meta-ties form the feature vectors of each player, while the status performance forms the classification label.


Chong-U Lim and D. Fox Harrell, “Modeling Player Preferences in Avatar Customization Using Social Network Data: A Case-Study Using Virtual Items in Team Fortress 2,” Proceedings of the IEEE Conference on Computational Intelligence and Games, 2013, Niagara Falls, Canada, Aug 11 – August 13. [pdf] [ppt] [bib]

Chong-U Lim. "Modeling Player Self-Representation in Multiplayer Online Games using Social Network Data," S.M. Thesis, 2013, Massachusetts Institute of Technology, USA. [pdf]


M. Granovetter, “The Strength of Weak Ties,” American journal of sociology, 1973.

H. pu, “Social Network Profiles as Taste Performances,” Journal of Computer-Mediated Communication, vol. 13, no. 1, pp. 252–275, Oct. 2007.

C. Moore, “Hats of affect: a study of affect, achievements and hats in Team Fortress 2,” Game Studies, vol. 11, no. 1, pp. 1–28, 2011.

E. Gilbert and K. Karahapos, “Predicting tie strength with social media,” Proceedings of the 27th international conference on Human factors in computing systems - CHI 09, p. 211, 2009.

R. Becker, Y. Chernihov, Y. Shavitt, and N. Zilberman, “An analysis of the Steam community network evolution,” 2012 IEEE 27th Convention of Electrical and Electronics Engineers in Israel, pp. 1–5, Nov. 2012.

D. F. Harrell, G. Vargas, and R. Perry, “Steps Toward the AIR Toolkit: An Approach to Modepng Social Identity Phenomena in Computational Media,” in Proceedings of the 2nd International Conference on Computational Creativity, 2011.

D. F. Harrell, “D. Fox Harrell. Designing empowering and critical identities in social computing and gaming. CoDesign: International Journal of CoCreation in Design and the Arts, 6(4):187–206, December 2010.

Project Images: 
Supporting Documentation: