Browsing by Author "Shim, Kyong Jin"
Now showing 1 - 6 of 6
Results Per Page
Sort Options
Item A computational approach to the study of player and group performance in massively multiplayer online games.(2011-12) Shim, Kyong JinThe market for video games skyrocketed over the past decade. Massively Multiplayer Online Games (MMOGs) have become increasingly popular and have communities comprising over 47 million subscribers by the year 2008. With their increasing popularity, researchers are realizing that video games can be a means to fully observe an entire isolated universe. Each action is logged, and the level of granularity and completeness with which information is collected is unmatched by any real-life experimental setup. They serve as unprecedented tools to theorize and empirically model the social and behavioral dynamics of individuals, groups, and networks within large communities. Virtual world applications usually have a thin-client architecture, practically all user actions are captured in the click-stream logged at the server. This dataset contains a comprehensive record of every user's in-network activities, accomplishments, interactions, economic status, etc. A brief record of the user's side information (i.e. profile data) is also stored. It is common for popular social networking and collaborative systems to have hundreds of thousands of users generating copious amounts of data based on the many different activities they are participating in at any given time. The data also has a temporal component which is often an integral part of the analysis and introduces further relationships that must be accounted for. Thus, while providing an exciting new tool for the social sciences, the virtual worlds also present a set of difficult and novel computational challenges. In the gaming community today, there is a growing interest in understanding player behaviors both inside and outside the gaming space. Game companies are interested in finding out how their games are played, if they are being played as intended, how the different game mechanics are being played out and how the different game playing patterns lead to a high level of satisfaction and entertainment for customers. Retrospective analyses after the game launch on existing game features can reveal information on which features enhance player's gaming experience and to which demographic segments they especially appeal to. Features negatively correlated with gaming experience can be considered for removal while those positively correlated with gaming experience can be further enhanced. For new game features, prospective analyses before the game launch can reveal information on which features might appeal to certain player population segments with a certain level of confidence and user-oriented testing can focus on these features for further validation. This thesis work presents the first comprehensive quantitative analysis of an important aspect of MMOG game play, namely player and group performance. While there are many different game genres (i.e. action, shooter, action-adventure, adventure, role-playing, and simulation) and many dimensions comprising players' game-play experience, in certain game genres such as MMOGs, close connection has been reported between player enjoyment and completing challenges and mastering skills. A systematic study of individual game player characteristics, group composition and characteristics, social interactions amongst the group members, and game environments can reveal a great deal about what are the recipes for success in achieving various objectives in the game. Broadly, this thesis work seeks to develop 1) Player performance metrics and prediction models, 2) Player activity prediction model, 3) Player enjoyment prediction model, and 4) Group performance metrics and prediction models. Lastly, we contribute a single, generic framework for player and group behavior analysis that is applicable to other MMOGs with minimal configuration changes.Item Data Mining Based Predictive Models for Overall Health Indices(2010-04-14) Rajkumar, Ridhima; Shim, Kyong Jin; Srivastava, JaideepIn this study, we infer health care indices of individuals using their pharmacy medical and prescription claims. Specifically, we focus on the widely used Charlson Index. We use data mining techniques to formulate the problem of classifying Charlson Index (CI) and build predictive models to predict individual health index score. First, we present comparative analyses of several classification algorithms. Second, our study shows that certain ensemble algorithms lead to higher prediction accuracy in comparison to base algorithms. Third, we introduce cost-sensitive learning to the classification algorithms and show that the inclusion of cost-sensitive learning leads to improved prediction accuracy. The built predictive models can be used to allocate health care resources to individuals. It is expected to help reduce the cost of health care resource allocation and provisioning and thereby allow countries and communities lacking the ability to afford high health care cost to provide health indices (coverage), provide individuals with health index which takes into consideration their overall health and thereby improve quality of individual health assessment (quality), and improve reliability of decision making by focusing on a set of objective criteria for all individuals (reliability).Item Evaluation of Protein Backbone Alphabets: Using Predicted Local Structure for Fold Recognition(2010-07-14) Shim, Kyong JinOptimally combining available information is one of the key challenges in knowledge-driven prediction techniques. In this study, we evaluate six Phi and Psi-based backbone alphabets. We show that the addition of predicted backbone conformations to SVM classifiers can improve fold recognition. Our experimental results show that the inclusion of predicted backbone conformations in our feature representation leads to higher overall accuracy compared to when using amino acid residues alone.Item Inferring Player Rating from Performance Data in Massively Multiplayer Online Role-Playing Games (MMORPGs)(2009-05-20) Shim, Kyong Jin; Pathak, Nishith; Srivastava, JaideepIn recent years, researchers have taken notice that virtual environments such as EverQuest II serve as a major mechanism for socialization. In particular, educational research has found virtual environments to be a sound venue for studying learning, collaboration, social participation, literacy in online space, and learning trajectory at the individual level as well as at the group level. The present research is concerned with learning in virtual environments and examines online player performance in EverQuest II, a popular massively multiplayer online role-playing game (MMORPG) developed by Sony Online Entertainment. The study uses the game's player performance data to devise performance metrics for online players. The study reports three major findings. First, we show that the game's point scaling system over-estimates performances of lower level players and under-estimates performances of higher level players. We present a novel point scaling system based on the game's player performance data that addresses the under-estimation and over-estimation problems. Second, we present a highly accurate predictive model for player performance as a function of past behavior. Third, we show that playing in groups has impact on player performance and that individual characteristics alone are not sufficient for explaining individual's performance, which calls for a different set of performance metrics methods. The discrepancy between the point scaling system in the game and observed player performance can be used as a guide to modify the existing system to better reflect the expected learning behaviors in different levels.Item Player Performance Prediction in Massively Multiplayer Online Role-Playing Games (MMORPGs)(2010-02-04) Shim, Kyong Jin; Sharan, Richa; Srivastava, JaideepRecent years have seen an ever increasing number of people interacting in the online space. Massively multiplayer online role-playing games (MMORPGs) are personal computer or console-based digital games where thousands of players can simultaneously sign on to the same online, persistent virtual world to interact and collaborate with each other through their in-game characters. In recent years, researchers have found virtual environments to be a sound venue for studying learning, collaboration, social participation, literacy in online space, and learning trajectory at the individual level as well as at the group level. While many games today provide web and GUI-based reports and dashboards for monitoring player performance, we propose a more comprehensive performance management tool (i.e. player scorecards) for measuring and reporting operational activities of game players. This study uses performance data of game players in EverQuest II, a popular MMORPG developed by Sony Online Entertainment, to build performance prediction models for game players. The prediction models provide a projection of player's future performance based on his past performance, which is expected to be a useful addition to existing player performance monitoring tools. First, we show that variations of PECOTA and MARCEL, two most popular baseball home run prediction methods, can be used for game player performance prediction. Second, we evaluate the effects of varying lengths of past performance and show that past performance can be a good predictor of future performance up to a certain degree. Third, we show that game players do not regress towards the mean and that prediction models built on buckets using discretization based on binning and histograms lead to higher prediction coverage.Item Prediction of Protein Subcellular Localization: A Machine Learning Approach(2010-06-24) Shim, Kyong JinSubcellular localization is a key functional characteristic of proteins. Optimally combining available information is one of the key challenges in today's knowledge-based subcellular localization prediction approaches. This study explores machine learning approaches for the prediction of protein subcellular localization that use resources concerning Gene Ontology and secondary structures. Using the spectrum kernel for feature representation of amino acid sequences and secondary structures, we explore an SVM-based learning method that classifies six subcellular localization sites: endoplasmic reticulum, extracellular, Golgi, membrane, mitochondria, and nucleus.