Application of the Artificial Society Approach to Multiplayer Online Games: A Case Study on Effects of a Robot Rental Mechanism



Similar documents
EVALUATING REPUTATION OF ONLINE-GAME PLAYERS BASED ON INCOMING CHAT MESSAGES


Data Mining. Anyone can tell you that it takes hard work, talent, and hours upon hours of

THE WHE TO PLAY. Teacher s Guide Getting Started. Shereen Khan & Fayad Ali Trinidad and Tobago

Modeling and Design of Intelligent Agent System

Laboratory work in AI: First steps in Poker Playing Agents and Opponent Modeling

Leveling-Up in Heroes of Might and Magic III

Univariate Regression

Evolutionary denoising based on an estimation of Hölder exponents with oscillations.

VISUALIZATION OF DENSITY FUNCTIONS WITH GEOGEBRA

Multi-agent System Learning Support Software with Fighting Games

Center: Finding the Median. Median. Spread: Home on the Range. Center: Finding the Median (cont.)

Equilibrium: Illustrations

Gamma Distribution Fitting

A Performance Study of Load Balancing Strategies for Approximate String Matching on an MPI Heterogeneous System Environment

Organizational Learning and Knowledge Spillover in Innovation Networks: Agent-Based Approach (Extending SKIN Framework)

Online Appendices to the Corporate Propensity to Save

Single Level Drill Down Interactive Visualization Technique for Descriptive Data Mining Results

Beyond RTTP: Gaming Your Classroom

Master s Thesis. A Study on Active Queue Management Mechanisms for. Internet Routers: Design, Performance Analysis, and.


Intelligent Agents Serving Based On The Society Information

Acing Math (One Deck At A Time!): A Collection of Math Games. Table of Contents

On evaluating performance of exploration strategies for an autonomous mobile robot

Analysis of Bayesian Dynamic Linear Models

Modelling Quest Data for Game Designers Simon Joslin School of Software Engineering Queensland University of Technology PO Box

Least Squares Estimation

Applying Design Patterns in Distributing a Genetic Algorithm Application

Agent-based Modeling of Disrupted Market Ecologies: A Strategic Tool to Think With

A Cloud Based Solution with IT Convergence for Eliminating Manufacturing Wastes

A Color Placement Support System for Visualization Designs Based on Subjective Color Balance

Using Excel (Microsoft Office 2007 Version) for Graphical Analysis of Data

The Design and Implementation of an Android Game: Foxes and Chickens

You Are What You Bet: Eliciting Risk Attitudes from Horse Races

Learning tasks from observation and practice

Institute of Actuaries of India Subject CT3 Probability and Mathematical Statistics

COMMON CORE STATE STANDARDS FOR

Simple Linear Regression Inference

Adaptation of Rapid Prototyping Model for Serious Games Development

Basic Probability and Statistics Review. Six Sigma Black Belt Primer

1) Write the following as an algebraic expression using x as the variable: Triple a number subtracted from the number

Behavioral Entropy of a Cellular Phone User

ESTIMATING THE DISTRIBUTION OF DEMAND USING BOUNDED SALES DATA

DATA MINING TECHNIQUES SUPPORT TO KNOWLEGDE OF BUSINESS INTELLIGENT SYSTEM

Evaluation of a New Method for Measuring the Internet Degree Distribution: Simulation Results

A Study on the Comparison of Electricity Forecasting Models: Korea and China

Follow-up Study of an Application of Design of Experiments to a Technical Trading System

Random Fibonacci-type Sequences in Online Gambling

Chicago Booth BUSINESS STATISTICS Final Exam Fall 2011

The Rich Are Different! Pareto law from asymmetric interactions in asset exchange models

Evaluation of Students' Modeling and Programming Skills

Spatial sampling effect of laboratory practices in a porphyry copper deposit

DECENTRALIZED SCALE-FREE NETWORK CONSTRUCTION AND LOAD BALANCING IN MASSIVE MULTIUSER VIRTUAL ENVIRONMENTS

Determining optimal window size for texture feature extraction methods

AKEY challenge in communication networks is to design

A STUDY REGARDING INTER DOMAIN LINKED DOCUMENTS SIMILARITY AND THEIR CONSEQUENT BOUNCE RATE

Automatic Gameplay Testing for Message Passing Architectures

Markov Chains for the RISK Board Game Revisited. Introduction. The Markov Chain. Jason A. Osborne North Carolina State University Raleigh, NC 27695

Department of Mathematics, Indian Institute of Technology, Kharagpur Assignment 2-3, Probability and Statistics, March Due:-March 25, 2015.

A Sarsa based Autonomous Stock Trading Agent

Online Learning of Genetic Network Programming (GNP)

Solution Let us regress percentage of games versus total payroll.

Products reliability assessment using Monte-Carlo simulation

WORKED EXAMPLES 1 TOTAL PROBABILITY AND BAYES THEOREM

A Genetic Algorithm-Evolved 3D Point Cloud Descriptor

How To Find Out If International Debt Securities And Gdp Are Related

Optimum Design of Worm Gears with Multiple Computer Aided Techniques

Encourage students interest in computer science, game design, engineering and art with ProjectFUN!

THREE DIMENSIONAL REPRESENTATION OF AMINO ACID CHARAC- TERISTICS

Common Core Unit Summary Grades 6 to 8

IMPROVING TRADITIONAL EARNED VALUE MANAGEMENT BY INCORPORATING STATISTICAL PROCESS CHARTS

Oscillations of the Sending Window in Compound TCP

GUIDANCE FOR ASSESSING THE LIKELIHOOD THAT A SYSTEM WILL DEMONSTRATE ITS RELIABILITY REQUIREMENT DURING INITIAL OPERATIONAL TEST.

CHAPTER 13 SIMPLE LINEAR REGRESSION. Opening Example. Simple Regression. Linear Regression

Process Modelling from Insurance Event Log

Comparing Alternate Designs For A Multi-Domain Cluster Sample

A web marketing system with automatic pricing. Introduction. Abstract: Keywords:

The Correlation in the Global Context of Financial Markets and the Evolution of Emerging Market of Romania Through the Bucharest Stock Exchange

Research on information propagation analyzing odds in horse racing

SIMPLIFIED PERFORMANCE MODEL FOR HYBRID WIND DIESEL SYSTEMS. J. F. MANWELL, J. G. McGOWAN and U. ABDULWAHID

MULTI-CRITERIA PROJECT PORTFOLIO OPTIMIZATION UNDER RISK AND SPECIFIC LIMITATIONS

arxiv:physics/ v2 [physics.comp-ph] 9 Nov 2006

Artificial Intelligence Beating Human Opponents in Poker

A Robust Method for Solving Transcendental Equations

Answers to Concepts in Review

Course Text. Required Computing Software. Course Description. Course Objectives. StraighterLine. Business Statistics

An Application of Inverse Reinforcement Learning to Medical Records of Diabetes Treatment

Data Mining: Algorithms and Applications Matrix Math Review

Industry Environment and Concepts for Forecasting 1

Efficient Statistical Methods for Evaluating Trading Agent Performance

Genetic algorithms for changing environments

On Correlating Performance Metrics

Stock Trading by Modelling Price Trend with Dynamic Bayesian Networks

Data Analysis. Using Excel. Jeffrey L. Rummel. BBA Seminar. Data in Excel. Excel Calculations of Descriptive Statistics. Single Variable Graphs

Determinants of the Total Quality Management Implementation in SMEs in Iran (Case of Metal Industry)

Regularized Logistic Regression for Mind Reading with Parallel Validation

Agent-Based Pricing Determination for Cloud Services in Multi-Tenant Environment

Using Adaptive Random Trees (ART) for optimal scorecard segmentation

Supplement to Call Centers with Delay Information: Models and Insights

Load balancing in a heterogeneous computer system by self-organizing Kohonen network

Transcription:

Application of the Artificial Society Approach to Multiplayer Online Games: A Case Study on Effects of a Robot Rental Mechanism Ruck Thawonmas and Takeshi Yagome Intelligent Computer Entertainment Laboratory Department of Human and Computer Intelligence, Ritsumeikan University Kusatsu, Shiga 525-8577, Japan ruck@ci.ritsumei.ac.jp Abstract This paper discusses a case study where the artificial society approach is used to examine effects of a robot rental mechanism in a multiplayer online game, under development at the authors laboratory. Simulation results show the effectiveness of the proposed rental mechanism, namely, better game balance and inflation control. These results also indicate that the artificial society approach has high potential for use in multiplayer online game applications in order to systematically predict effects of each service or component.. Introduction Development of multiplayer online games involves high costs. A major reason for this is that effects of each service or component are hardly known before launching of beta versions. To remove unwanted effects or add more wanted effects, ad hoc development is forced to continue after releasing of the games. This paper aims to present a solution to the above issue. Our solution is to use the artificial society approach [, 2] for analyzing each service or component in advance. In the artificial society approach, computer simulation is used to explore and understand social and economic issues. The approach has been widely used since 99s among social scientists. In this paper, based on the artificial society approach, we study effects of introducing a robot rental mechanism into a multiplayer online game called The ICE. 2. Game concepts of The ICE Here we describe the game concepts of The ICE. This game is under development at the authors laboratory using a customized middleware of Community Engine Inc [3]. Besides educational purpose, it will be used as a platform for examining various data mining techniques in [4-6], recently proposed by the authors group and tested on a multiplayer online game simulator called Zereal [7]. In The ICE, each player is initially given a programmable robot. A major role of the player is that of programming the robot and participating with it in a snow battle at a battle field against the team of another player and his robot. A reward is distributed to the winner player of the battle. The fun of this game is the in-game programming. Though they might have difficulty in programming their robots in the beginning, through trial-and-error processes and advices from virtual friends or Game Masters, players should gradually overcome such hurdles step by step. They should also become more and more attached to their own robots. However, there might be some players who have a relatively low programming ability. Rather than preventing this type of players from playing the game, we propose below a robot rental mechanism for balancing the player status, in terms of the amount of virtual versus the. This should result in higher player satisfaction. 3. Robot rental mechanism and its economical perspective The proposed robot rental mechanism is outlined as follows:

A:.8 B:.33 PROGRAMMING.7 C:..33.3 B:.33.33 BATTLING CHATTING.5.33 A:.5 B:.33.5 Table. Evolution of player ability state evolvable ability PROGRAMMING p i, s i BATTLING p i, s i, c i CHATTING p i WATCHING.5 REGISTERING Figure. State transition diagram of the players. A player who wishes to rent out a clone of the current version of his robot must register that version at the robot shop of the corresponding game zone. 2. To use registered clones besides his own robot in a snow battle, a player must go to the robot shop, specify the number of clones he wants, and pay a rental fee to the shop owner, a non-player character. Upon receiving the fee, the shop owner randomly reproduces clones from the list of the registered versions of the robots and rent out the clones to the customer. At the end of each snow battle, all clones are voided. 3. If the team of a clone wins a snow battle, a part of the rental fee is given to the programmer of the robot, from which the clone is reproduced. From an economical perspective, we divide the players into the following two categories. Producer who is good at programming and thus programs his robot for not only winning a snow battle but also renting out its clones. Consumer who is poor in programming and hence needs to additionally use clones in a snow battle for increasing the chance to win. By having both producers and consumers in the game, an economy [8] is formed where virtual money, consisting of the battle rewards and the robot rental fees, is circulated. In this economy, the cost of X is represented by time, the number of simulation cycles in the simulator discussed below, required to complete X. For example, the programming cost of a robot is the number of simulation cycles required to program the robot. We expect that the proposed rental mechanism will weaken the power-law relation between the and the amount of of the players. Namely, with the proposed rental mechanism, players having higher performance robots do not always get richer in the game. Another more evident effect of the proposed rental mechanism is that of controlling inflation in the game. Without this mechanism, the system can not absorb the money from the players. Hence, the amount of of each player keeps on growing, causing inflation. 4. Game simulator In this section, we discuss how we simulate The ICE and the proposed robot rental mechanism. In our simulator, a player is modeled by an agent, and the ability of player i at a given simulation step, ability i, is defined by ability i = floor(p i + s i + c i ), () where floor() is the floor function, and p i, s i, and c i are the degree of programming ability, the degree of skillfulness, and the degree of concentration of player i at the current simulation step, respectively. The above p i, s i, and c i evolve according to the state transition diagram of the players shown in Fig., where the ovals represent states and the arrows represent transitional probabilities between states. Table shows the states at the end of which each corresponding ability is slightly increased by a small random number α. The initial state of a player can be one among PRO- GRAMMING, BATTLING, and WATCHING with the same probability. The description of each state in Fig. is given in detail as follows: PROGRAMMING indicates the state where the robot of the player is programmed. The performance of the current version of the player i s robot (and its clones), perform i, that can be achieved is defined by perform i = op i + dp i + sp i + pr i, (2) where op i, dp i, sp i, and pr i are the offense power, the defense power, the speed, and the precision

of the player i s robot, respectively. These four parameters are calculated by the nearest integer not larger than the result of p i added by a random value from to s i. The cost, i.e., the number of simulation cycles required to complete this state, depends linearly on the and inversely on the degree of concentration of the player. It is calculated by floor(perform i /c i ). At the end of this state, the version of the player i s robot is incremented by. BATTLING indicates the state where the player together with his robot and the rented clones are engaged in a snow battle. Player i rents clones for battling if the performance of the current version of his robot is not larger than a threshold θ i, i.e., perform i θ i. Here θ i is defined by the average performance of the former versions of his robot and the rented clones that player i previously used in the lost battles. If perform i θ i holds, player i rents up to τ clones at the robot shop. The total rental fee is the sum of the rental fee of each rented clone, where the rental fee of a clone is defined as the rental fee of the registered version of the robot from which the clone is reproduced. The out-going branches from this state depend on the following three conditions: Condition A: where the player s team wins the current battle, Condition B: where the player s team loses the current battle, and Condition C: where the player s team skips the battle because his robot has not been programmed yet. A snow battle consists of 5 turns. The team that first wins 3 turns will be the winner of the battle. In a given turn, if the total performance of the current version of the player i s robot and the rented clones added by s i and a random value from toc i is larger than that of the opponent team, the player i s team will win that turn. If his team wins, player i will be rewarded with λ money units, and the rental fee of his robot be increased by β H. If player i additionally uses clones in the win battle, the programmer of each robot, from which its clone is reproduced, will be given the rental fee of the clone multiplied by γ; In addition, the rental fee of the rented version of each robot is increased by β L. The number of simulation cycles required in this state is one cycle added by the number of cycles used to wait for the availability of a battle field and/or the availability of an opponent team. CHATTING indicates the state where the player chats with his friends. This state requires five simulation cycles. REGISTERING indicates the state where the player registers the current version of his robot at the robot shop. This state requires one simulation cycle. WATCHING indicates the state where the player merely watches a snow battle between another two teams. This state requires one simulation cycle. 5. Simulation results In our simulation, the main variables are as follows: 4 game zones 8 battle fields per one game zone 5 players per one game zone simulation cycles money units as the initial amount of money in hand for each player The initial degrees of p i, s i, and c i are uniformly assigned a random integer value from to 6, and α is randomly selected from.,., and. θ i is initially set to. For simulating the game without the proposed rental mechanism, this value is fixed through out the simulation. The initial rental fee for each registered version of a robot is set to money units. Rent related parameters τ, β H, β L, and γ are set to 3, 2.,.5, and., respectively. λ is set to money units. Figure 2 shows the scatter plot of the normalized amount of versus the normalized robot performance of the players for (a) where the rental mechanism is not implemented, and (b) where the rental mechanism is implemented. As can be readily seen, trend of the power-law exponent is more pronounced in Fig. 2.a than in Fig. 2.b. Namely, without the proposed rental mechanism, players who have higher performance robots are richer. When the proposed rental mechanism is implemented, this trend is less evident.

without rental.8.8 (a) without rental.8.8 (a).8 with rental.8 (b) Figure 2. Scatter plot of the normalized amount of versus the normalized of the players for (a) without rental and (b) with rental To further investigate the above claim, we conducted nonlinear regression [9] of the data in Fig. 2 using the simplex algorithm for fitting function y i = ax b i, where x i and y i are the and the amount of of player i, respectively, and a as well as b are initially set to. The resulting regression coefficients are shown in Table 2. The resulting regression curves overlayed on the scatter plots of Fig. 2 are depicted in Fig. 3. These results confirm our claim. Other statistics shown in Table 3 are the maximum, the mean, the median, and the standard deviation of both the and the amount of money in hand of the players. Note that a larger difference between the mean and the median indicates that the distribution of the data is more skewed, and this occurs in the case where the proposed rental mechanism is not implemented, especially for the amount of money in.8 with rental.8 (b) Figure 3. Scatter plot of the normalized amount of versus the normalized of the players with the regression curve being overlayed for (a) without rental and (b) with rental hand of the players. The standard deviation of both the and the amount of of the players are also much higher in the case where the proposed rental mechanism is not implemented. The results obtained so far conform to our expectation that the proposed robot rental mechanism can result in better game balance, in terms of the amount of versus the of the players. In addition, it should also be noted that the amount of of the players where the proposed rental mechanism is implemented is much lower. This means that the proposed rental mechanism well controls inflation in the system. Now we further analyze the data in the case where the proposed rental mechanism is implemented. We first consider as producers the players who register

Table 2. Regression coefficients for the case where the rental mechanism is not implemented, and the case where the rental mechanism is implemented coefficients without rental with rental a.28.58 b.6.82 their robots for renting out more often than average players, and the rest of the players as consumers. Based on this, the players are divided into 47 producers and 953 customers, where the average number of times the players register their robots is 29.4. The scatter plot for each type of players with the regression curve, obtained by the nonlinear regression discussed above, being overlayed is shown in Fig. 4.a and Fig. 4.b, respectively. The resulting regression coefficients for the producers and the consumers are given in Table 4. Table 5 shows for each type of players the maximum, the mean, the median, and the standard deviation of both the and the amount of of the players. These results reveal that the power-law relation between the and the amount of of the players is slightly higher for the producers. In addition, the producers have higher performance robots and are richer. The distribution of the amount of money in hand of the consumers is less skewed. 6. Conclusions and future works In this paper, it has been shown that the artificial society approach can systematically examine effects of the proposed robot rental mechanism. Namely, the power-law relation between the and the amount of of the players is weakened by the proposed rental mechanism. Players can be divided into producers and consumers. The producers have relatively higher status, thus the motivation for being the players of this type should be preserved. At the same time, the players of the consumer type should enjoy playing the game because their status can be maintained by renting clones for snow battles and they do not differ much in terms of the and the amount of. In addition, inflation in the game is successfully controlled by the proposed rental mechanism. Consequently, it is expected that high player satisfaction can be achieved if the robot rental mechanism is actually implemented Table 3. Major statistics of both the and the amount of of the players for (a) without rental and (b) with rental (a) without rental maximum 34 427 mean 5.74 979.7 median 9 55 stdev 57.34 9383.5 (b) with rental maximum 5 8822 mean 83.53 84 median 85 767 stdev 9.6 76.67 Table 4. Regression coefficients for the producers and the consumers coefficients producers consumers a.53 b.95.65 in The ICE. Our future works include studying effects with more complex player behaviors, such as heterogeneous player agents with non-supervised learning ability, and comparing the results with those actually obtained from The ICE when in operation. Acknowledgement This work has been supported in part by the Ritsumeikan University s Kyoto Art and Entertainment Innovation Research, a project of the 2 st Century Center of Excellence Program funded by the Japan Society for Promotion of Science. Both authors are grateful to the other members of the Intelligent Computer Entertainment Laboratory for their fruitful discussions.

.8 producers.8.8 (a) consumers.8 (b) Figure 4. Scatter plot of the normalized amount of versus the normalized of the players with the regression curve being overlayed for (a) producers and (b) consumers References [] Epstein, J.M. and Axtell, R., Growing Artificial Societies: Social Science from the Bottom Up, The MIT Press, Cambridge, Massachusetts, 996. [2] Gilbert, N. and Troitzsch, K.G., Simulation for the Social Scientist, Open Univ Press, Buckingham and Philadelphia, 999. [3] www.ce-lab.net/en/. [4] Thawonmas, R., Ho, J.Y., and Matsumoto, Y., Identification of Player Types in Massively Multiplayer Online Games, Proc. the 34th Annual conference of International Simulation And Gaming Association (ISAGA23), Chiba, Japan, pp. 893-9, Aug., 23. Table 5. Major statistics of both the and the amount of of the players for (a) the producers and (b) the consumers (a) producers maximum 5 8822 mean 92.58 275.6 median 92 223 stdev 5.59 83.72 (b) consumers maximum 22 546 mean 73.6 472.23 median 75 484 stdev 8.73 943 [5] Ho, J.Y., Matsumoto, Y., and Thawonmas, R., MMOG Player Identification: A Step toward CRM of MMOGs, Proc. the 6th Pacific Rim International Workshop on Multi-Agents (PRIMA23), Seoul, Korea, pp. 8-92, Nov., 23. [6] Ho, J.Y., and Thawonmas, R., Episode Detection with Vector Space Model in Agent Behavior Sequences of MMOGs, Proc. Future Business Technology Conference 24 (FUBUTEC 24), IN- SEAD, Fontainebleau, France, March 2-3, 24. (one of the candidates for the best paper award) [7] Tveit, A., Rein, O., Iversen, J.V. and Matskin, M., Scalable Agent-Based Simulation of Players in Massively Multiplayer Online Games, Proc. the 8th Scandinavian Conference on Artificial Intelligence (SCAI23), Bergen, Norway, Nov., 23. [8] Castronova, E., Virtual Worlds: A First-Hand Account of Market and Society on the Cyberian Frontier, The Gruter Institute Working Papers on Law, Economics, and Evolutionary Biology, Vol. 2: Article, 2. [9] Seber, G.A.F., and Wild, C.J., Nonlinear Regression, Wiley-Interscience, 23.