Extended Analysis of Multiplayer Platform Users Activity Based on the Virtual and Real Time...
Transcript of Extended Analysis of Multiplayer Platform Users Activity Based on the Virtual and Real Time...
Extended version of paper: Jankowski J., Analysis of Multiplayer Platform Users Activity Based on the
Virtual and Real Time Dimension Lecture Notes in Computer Science Volume 6984, 2011, pp 312-315
Analysis of Multiplayer Platform Users Activity Based
on the Virtual and Real Time Dimension
Jaroslaw Jankowski
Faculty of Computer Science and Information Technology
West Pomeranian University of Technology
ul. Zolnierska 49, 71-410 Szczecin, Poland
Abstract. The paper proposes an approach to modeling online multiplayer
systems users’ behavior and segmentation in terms of usage patterns. The
solutions presented, are based on analysis of time series in the real and virtual
time dimension. The proposed method can be applied in the field of community
platforms, virtual worlds and massively multiplayer online systems to capture
quantitative characteristics of online platforms usage.
Keywords: multiplayer platforms, time series analysis, web users’ behavior
1 Introduction
Together with the development of web systems, the need to conduct analyses that
focus on users’ behavior, increases. Research in the field of online communities,
virtual worlds, and massively multiplayer platforms and games, among other, relates
to users engagement [29] and social dynamics [7]. A big volume of the collected data
requires the usage of data mining methods, which are developed in the area of
personalization [22], semantic processing [9] and users’ segmentation [27]. Solutions
available in the cited papers are mainly focused on the analysis of behaviors in the
scope of websites in the connection with content, external links and navigation
patterns [23][12]. Less attention is paid to the segmentation of multiplayer platform
users and similarities based on time sequences related to their activity and platform
usage characteristics. In that context the goal of this article is to propose an approach
focused on the segmentation of users in the scope of trends and system usage
frequency. For the purposes of the considered behavior patterns, the dimension of real
and virtual time was introduced. The analysis of dependencies between the real and
virtual time activity and users segmentation based on usage patterns, delivers valuable
information about trends and can be used in planning platform development and
marketing actions. The structure of the article is as follows: paragraph two includes
the analysis of methods used for analysing website users activity, paragraph three
presents the assumptions of the proposed solution, paragraph four describes the
empirical research and analysis. User segmentation based on proposed approach is
presented in paragraph five. A summary and some final remarks are presented in
paragraph six.
Extended version of paper: Jankowski J., Analysis of Multiplayer Platform Users Activity Based on the
Virtual and Real Time Dimension Lecture Notes in Computer Science Volume 6984, 2011, pp 312-315
2 Motivation and related work
Multiplayer online platforms can be thought off as multi-dimensional environments,
which connect the features of economic, technical and social systems. They work
usually in the form of virtual worlds or massive multiplayer games, where users
develop a parallel identity, perform different roles and participate in creating complex
social relations. They function somehow simultaneously with mainstream social
applications, such as Facebook or Twitter. They can also be integrated with them in
the form of social games and provide different forms of entertainment or
communication than web platforms. Depending on the used technology they acquire a
2D, 2.5D or 3D representation and integrate extensions improving the communication
with other users by audio or video sensations. Depending on the scope of usage and
functional specification, they are divided into several groups [32]. S. Kish [19]
proposes classification of systems in this area in five main categories including
relation to real world. These systems are divided into MMORPGs (Massive
Multiplayer Online Role Playing Games), more general MMO (Massive Multiplayer
Online), metaverses (social oriented), MMOLEs (Learning and Training
Environments), intraverses (internal corporate systems) and paraverses (mirror
worlds). In all of this groups, there is a growing demand for implementation of new
methods for data processing. The analyses aim to recognize different occurrences
and trends to acquire knowledge that have a significant role in the development of
online services. It creates interest in the area of methods of social network analysis
[24], diffusion of marketing messages [1], recommendations [13][11][17] or
dynamics of network structure development [21]. There are also developed solutions
deriving from methods of knowledge discovery in databases, which with reference to
online systems acquire a specialized form of web mining algorithms. They can be
applied, among others, at the basis of agent systems [10][5], electronic marketing
platforms [6], and in design of intelligent interfaces. In most kind of applications time
dimension is taken into consideration and it is a base for activity pattern analysis and
loyalty estimation of online service users. In this area it is important to determine a
quantitative representation of time–related behavior characteristics, and compare the
various patterns of behavior.
New approaches and methods are developed that are focused on the analysis of data
that enable the acquisition of additional knowledge about online consumers’ behavior.
Temporal data mining is one of the developing fields of methods for knowledge
extraction from data bases focused on time characteristics. C.M. Antunes and L.O.
Arlindo presented a review of approaches and methods used in this field [2], and the
scope of appliance in different fields were identified. The survey of methods
presented in the study [28] relates to temporal data bases and indicates development
direction of methods focused on acquiring knowledge about hidden patterns and
analysis of changeability of sets with temporal characteristics. The study [16]
develops the aspect of web data specification and the need to search for dedicated
solutions that include local solutions which, when searching for generalized
dependencies for all data sets, are not detected. The presented methods are focused on
acquiring temporal, indirect, frequent patterns and their temporal, extended patterns in
the area of identifying Web users with distinct interests. Changeability of consumers’
behavior in time and evolution in time connected with different factors, as well as
developing new technologies in time, is emphasized in the works of V. Venkatesh and
M.G. Morris [31], E. Karahanna and others [18]. As B. Hernandez-Ortega, J. Jimenez
and others proved, there are significant differences in the behavior of potential, new
or permanent consumers [14]. The identification of their behavior enables adequate
segmentation and offers targeting or adaptation of functionality to a given group of
recipients. S.M. Beitzel and others presented an analysis focused on the temporal
aspects and arrangement in time of data from logs from servers connected with search
engines, and they proved that there is a possibility of temporal analyses in this
environment [3]. The concept of analysis of time series focused on analysis of Web
search system logs was presented by Y. Zhang, B.J. Jansen and others [36]. The
analysis of time series enabled the detection of dynamic occurrences and consumers’
behavior patterns. The temporal nature of data from transactional systems provided
the opportunity to isolate short-term patterns and to generalize multi-scope analyses.
Analysis in this area can be used to identify the consumers’ characteristics and also
increase their loyalty, which is a significant element of marketing actions [8].
The development of technologies based on communities, increases users’ interactions
and can be an element of strategies focused on loyalty building, e.g. in the form of
weblogs in the scope of corporative systems, which was indicated by S.C. Herring
and others [15]. Interpersonal engagement is an additional factor of loyalty building
also in the scope of technological platforms, which was proved by C. Wagner and N.
Bolloju [34]. What is also important in creating loyalty towards advanced online
services are the limitations of introduction threshold, where getting to know a new
system is time-consuming. F. Wangenheima and T. Bayona state that behavioral
aspects of loyalty translate into the purchase of a service and recognition of the
superiority of an individual service over other subjects on the market [35]. H.P. Lu
and S. Wang identify a dependency between users’ satisfaction and loyalty in
connection to Internet addiction [20]. The research also indicates a phenomenon
characteristic for online entertainment services, where loyalty can occur even with the
lack of satisfaction and according to H. Tsai and H. Huang [30] it can discourage
searching for alternatives. The presented studies focus on the aspect of representation
and temporal data analyses. Nevertheless, they do not include the specification of
multiplayer platforms and functioning of users in virtual reality, for which different
structuring of time lapse can be determined than in real systems, and with a use of
multidimensional analysis, the evaluation of loyalty level or characteristic of using
individual service or platform can be done. The approach presented in this paper
identifies two dimensions of temporal data analysis based on real and virtual time that
enable the segmentation of users, focusing on time series and trends identification.
The approach presented in the next part gives the opportunity to conduct analyses that
enable users to determine audiences’ parameters, identify their characteristics,
behavior trends, estimate loyalty and engagement levels of users, as well as detect
tendencies.
3 Conceptual framework for usage patterns analysis
The article herein proposes an approach towards data analysis and characteristics of
multiplayer platforms usage related to the players activity and engagement. Users
usually perform distinct roles in the virtual worlds or online games, building their
position by achieving levels of initiation and expertise. Their representation is
functioning in cyber space in a sense of place and environment. It can be assumed that
here exists also another dimension of time, more related to the virtual life and activity
within the online system than to real life. Time series analysis in such case will
usually show some periods of activity and inactive real days when the user is not
logged into the system, and seasonality can be observed with low activity during late
hours or offline periods. Analyzing together the time dimension related to the period
spent in the system and real life time, is mixing both realities. In our approach we
propose to separate them into two time dimensions. The main focus is to analyze the
Extended version of paper: Jankowski J., Analysis of Multiplayer Platform Users Activity Based on the
Virtual and Real Time Dimension Lecture Notes in Computer Science Volume 6984, 2011, pp 312-315
relation between them and measure the distance and dynamics of their changes
together with comparison to the other users, and the possibility to create user
segments. Theoretical background related to a different time interpretation for
different periods is based on the approach proposed by S. Radas and S.M. Shugan
who de-seasonalised processes by changing the speed at which time flows [26].
Research showed that during low seasons time can be slowed down and during high
seasons can be speeded up. In our research we don’t change the speed of time flow.
Instead, we considered time flow in the online system, and time spent offline without
contacts with multiplayer environments as separate dimensions.
Lapse of time in the online system (herein referred to as ‘virtual time’) can be
considered separately from real time. For mapping and parameters’ determination
purposes, the two-dimensional bi-temporal model of time representation was adopted,
in which occurrences are registered in terms of real and virtual time. There can be
introduced different time separations that will lead to an increase of analysis accuracy
by decomposition of the time period into sub periods and determination of time
granulation. In order to describe such a dependency, the virtual time factor was
defined, which gives the relation of virtual and real time, according to the formula
F=v*n/d, where d represents the number of real time days, v the number of virtual
time days and n is the number of time intervals taken into consideration for
monitoring the users’ activity. Such an approach enables monitoring the lapse of both
types of time for each user and determining a level of activeness. For the purpose of
data analysis consistency with assumed time granulation at the t moment for every i-
th user, the St snapshot of parameters that are monitored in an individual time period
is generated. Parameters can include social characteristics, user’s activeness, number
of loggings and use of particular system functions according to the formula:
1
,1,1
1
,1t
00
,1
1,= S
tk
t
i
k
i
t
i
t
t
tk
i
k
i PT
PDPPt
(1)
where Ps,t parameters were assumed for every i-th user, with i = 1,…,n for every type
of s parameter where s= 1,…,m. For every t time moment the aggregated parameter
values are determined in relation to previous periods, and so are the dynamics of
change Dt in relation to the previous period and the total number of periods T. Time
dependency between real and virtual time is a measure of the user’s engagement and
the dynamics of service usage. A set of snapshot parameters depends on the
parameters and abilities of a system and aims to present the dynamics of system usage
and changeability of the user’s determined features. For the purpose of sequence
analysis, the methods focused on similarity between sequence A and B can be
introduced in the form of similarity d(A,B) and the correlation factor can be used to
measure it. Similarity can be defined as a distance among the Minkowski series,
determined consistently with the following formula:
PPn
i
ii yxBAd
/1
1
),(
(2)
In the case P=2, the distance among series is a Euclidean distance. Determination of
distance in such a way causes limitations in case of a series with different length and a
series with time-lag with changeable amplitude. When comparing two series in the
scope of real and virtual time, there can be a time-lag that should not influence the
total similarity evaluation. To evaluate similarity, the time warping distance method
was applied. For two series X(x1,x2,…,xn) and Y(y1,y2,…,ym), with lengths n and m
respectively, the M matrix is defined as an interpoint relation between series X and Y
where element Mij indicates the distance d(xi,yj) between xi and yj. Dependency among
a series is determined by the time warping path. The algorithm determines the
warping path with the lowest cost between two series according to the formula [25]:
kkW
wwwwdYXD ,...,,,min, 21 (3)
where dk=d(xi,yj) indicates distance represented by wk=(i,j) on w path. A series with a
higher level of similarity can be better compared because of alignment and
dependencies resulting from dynamic time distance. In the proposed procedure, the
users’ sequences can be compared to distinguish similar behavior and compared to the
pattern of ideal time lapse similar to real time. On the basis of time series and
behavior patterns it is possible to create a learning system focused on detection of
users with big potential, for which the engagement in system usage and becoming a
heavy user has a high probability. For the purpose of further analysis, the usage of
classification and regression trees was adopted [4]. The selection of subsections was
applied on the basis of predictor variables, which are used to forecast the belonging of
cases and objects to classes determined by the dependent variables. In the applied
method the first phase of division consists of determining the division knots and a
predictor variable. The classification is made on the basis of the value representing
the distance from the ideal vector for individual units of time. Similarity in the frames
of separate groups relates to the patterns of behavior. The forecasts of behaviors in
later periods can be generated based on sequences with determined lengths. In the
next part of this article this approach is verified on the basis of empirical data and a
particular area of applications was identified.
4 Empirical data analysis based on real and virtual time
In the next stage an analysis based on empirical data was conducted. Real and virtual
time was identified for the multiplayer platform Timik.pl (a virtual world platform),
targeting web users in Poland. The system in question is presented in the register of
virtual worlds published by Virtual Worlds Management, and has more than 850k
registered users [33]. The main function relates to communication, social area,
multiplayer games and chats. Data received for research purpose remained
anonymous and were delivered in encoded form. We took into consideration 4410
users who created accounts during three months prior to the research and their
activeness was registered at least for 10 real days in that period. We analysed the
period of one hundred days and the number of real days was equal to virtual days. For
particular users, changeable dependencies of two dimensions of time can be observed.
Fig. 1 shows the time dependency for real time Rd and virtual time Vd. On axes x and
y, real and virtual days were marked. For user u1 there is a consistency of measures,
and a high frequency of system usage is observed. User u6 initially showed an
increase in the distance between real and virtual time, with Vd =12 and Rd=20 at point
A. In the following days the distance is slightly changed to Vd=19, and there is a
stability up to point B where the distance rapidly increases and eventually equals
Vd=49 and Rd=100. For user u5, stability occurs from Vd=7 up to point C after which
the distance rapidly increases between dimensions. Dependencies among users can be
a basis for segmentation depending on behavior similarities. The analysis was made
including similarities and a distance to an ideal vector, which represents a state of
maximum engagement. In such a situation Vd is consistent with Rd.
Extended version of paper: Jankowski J., Analysis of Multiplayer Platform Users Activity Based on the
Virtual and Real Time Dimension Lecture Notes in Computer Science Volume 6984, 2011, pp 312-315
Fig. 1. Relative users’ behavior characteristics u1, u5
and u6.
Fig. 2. Average values of real day in relation to
virtual day.
Fig. 2 presents average values of real days for corresponding virtual days for the
initial set of users. For example, on the 11th virtual day the average value of the real
day was 25. In the initial period of service usage, the average value increases, and in
the following period it decreases. In the following step the dynamic time warping
algorithm was used to determine the similarity of the time series with the ideal vector.
Fig. 3 presents the changeable dependency for particular users with a change of
sequence length from virtual dimension. Together with the change of sequence length,
the minimum and maximum distance to the ideal vector increases, which is illustrated
by a regression line.
Fig. 3. Distance from ideal vector calculated with dynamic time warping.
Fig. 4. Detailed distance characteristics for users u1-u1000.
A similarity to the ideal vector results not only from the sequence length, but also
from the character of changes and lag of virtual time vector. Such characteristics are
included by dynamic time wrap method that enables, among others, a presentation
with comparison to the sequence time-lag. Fig. 4 illustrates in detail the distance for
sequences from 1-1000 (window W1 from Fig. 3) and distances from the ideal vector.
Point A represents user u92 with dtw=68 and sequence length 74. Point B shows user
u183 with dtw=574 and sequence length 64. Point C shows user u332 with dtw=2875
and sequence length 52. Point D denotes user u719 and dtw=3294 and sequence
length= 36. Sequence similarity and segmentation can be conducted by searching for
optimal alignment, which enables similarity analysis in case of time-lag. During the
analysis of behavior pattern characteristics, the similarity and time-lag of determined
behaviors can be identified. The punctual sequence comparison does not provide such
possibility. Fig. 5 shows alignment for sequences A and B previously analysed, where
one of the sequences is treated as a reference point and constitutes a reference pattern
for comparison of other sequences. Matching pairs of points determine areas of
similarity in time and can be visualised as cost density for sequences A and B, as
shown on Fig. 6.
B C
Fig. 5. Aligning time series A and B. Fig. 6. Cost density for sequences A and B.
Dependencies between sequences can be analysed with asymmetric presentation,
where the lag is determined in relation to a reference sequence. Such a presentation
for sequence A and B, and A and C can be modelled as well. Using this method, the
warping curve can show the direction of changes and lag in relation to the reference
points. Analyses of dependencies of virtual and real time can be used for early
detection of active users, and for estimation of the possibility of conversion on
permanent users who often participate in the service. Depending on the audiences’
specification, the set of pattern sequences can be determined, as well as a set of
informative supports, for which estimation is realized with adequate accuracy.
5 User segmentation based on usage patterns
In the next step we analysed patterns of behavior-analysed users, whose activity was
registered at least during fifty days in relation to the ideal vector. The main goal was
to verify an algorithm based on classification trees in order to forecast users activity
based on distance between virtual time and real time. We performed a division of
users into four classes depending on their activity during the period of our research.
The first class included users, which did not use the platform for maximally ten days
during the analysed period. There were 68 users in this group. The second group
included 71 users with a lack of activity for up to twenty days. The third group
consisted of 115 users with lack of activity for up to 30 days. The fourth group
gathered 144 users who did not use the system during the period in question for up to
40 days. For every user ui, there were current values determined, connected with a
real time R(ui), virtual time V(ui) and the distance D(ui) from the ideal vector. Table 1
presents the fragment of input data for a sample of chosen users
U={u1,u19,u39,u55,u203} with different patterns of system usage during one hundred
virtual days vd1 - vd100. The real time dimension Rd(ui) corresponding to each virtual
day was determined for each user. For example, for user u19 on the second day of
virtual time vd2 the corresponding value of real time was Rd(u19,vd2)=7. Table 1
shows the break in system usage and lack of user’s activity on the {2,3,4,5,6} day of
real time. Distance for this user amounts to 5 because in reference to the ideal vector
the corresponding amount of a real time should amount to 2, and is 7. In the next
period the distance for user u19 does not increase which shows the stable usage of a
system in the following days . Another pattern occurs for user u55 for whom in the
following days the distance gradually increases, which indicates the decrease of
activity in the following days.
Table 1. Selected users from different classes.
Uid Usage Input vd1 vd2 vd3 vd4 vd5 vd6 vd7 vd8 vd9 vd10 C
u1 100% Rd(u1) 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00
1.00 D(u1) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
u19 91% Rd(u19) 1.00 7.00 8.00 9.00 10.00 11.00 12.00 13.00 14.00 15.00 1.00
Extended version of paper: Jankowski J., Analysis of Multiplayer Platform Users Activity Based on the
Virtual and Real Time Dimension Lecture Notes in Computer Science Volume 6984, 2011, pp 312-315
D(u19) 0.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00 5.00
u39 85% Rd(u39) 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00 11.00 12.00
1.00 D(u39) 2.00 2.00 2.00 2.00 2.00 2.00 2.00 2.00 2.00 2.00
u55 82% Rd(u55) 5.00 7.00 16.00 19.00 20.00 21.00 22.00 23.00 24.00 25.00
2.00 D(u55) 4.00 5.00 13.00 15.00 15.00 15.00 15.00 15.00 15.00 15.00
u203 62% Rd(u1) 13.00 14.00 15.00 19.00 27.00 28.00 29.00 38.00 47.00 48.00
4.00 D(u1) 12.00 12.00 12.00 15.00 22.00 22.00 22.00 30.00 38.00 38.00
Ideal vector 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00
The aim of our research and analysis presented in this paper, was to determine
whether it is possible to forecast behaviors of users, classify them and allocate them in
a certain group at the initial stage of multiplayer platform usage. For this reason,
classification trees were built based on input data. The analysis was conducted for a
different sequence length relatively to the initial values for ten and twenty days of
service usage. Table 2 presents the classification matrix for the acquired results. In the
case of a short time of analysis concerning the first ten days of activity in relation to
the forecast horizon of the whole analysed period, the analysis is burdened with a big
deviation and most of the users were included in the fourth class. When the analysed
period is extended to 20 days, the accuracy of prediction increased for classes 1, 3,
and 4. The results for the second class of users were burdened with a bigger deviation.
Table 2. Classification matrix for 10 and 20 days sequences.
Period Class Forecast class 1
Forecast class 2 Forecast class 3 Forecast class 4
10 days
1 4.41% 13.24% 11.76% 70.59%
2 0.00% 25.00% 26.39% 48.61%
3 0.00% 8.62% 33.62% 57.76%
4 0.69% 2.78% 14.58% 81.94%
Elements 4 41 87 268
Total % 1.00% 10.25% 21.75% 67.00%
20 days
1 64.71% 5.88% 23.53% 5.88%
2 29.17% 26.39% 23.61% 20.83% 3 27.59% 5.17% 57.76% 9.48%
4 15.97% 7.64% 25.69% 50.69%
Elements 120 40 137 103 Total % 30.00% 10.00% 34.25% 25.75%
Fig. 7 and Fig. 8 illustrate the results of classification in reference to real data for the
initial ten and twenty values of service entries. There is a risk of system usage
decrease among the users from the fourth group. Intervention actions can be
conducted towards them, aiming at the recognition of their needs and determination of
the reason of such behavior. The users from the third group are characterised by a
decreased activity and this group can be subject to more detailed analyses. The users
from the first group are most active and the detailed analyses of their profiles, social
activeness or the realized transactions can constitute the basis for determination of
behaviors and features of a system, which contributes to big activeness.
Fig. 7. Classification accuracy for 10-day
usage patterns.
Fig. 8. Classification accuracy for 20-day
usage patterns.
The proposed solution enables to determine the scope of analysis, which should be
conducted for the individual group of audiences, and to reduce the number of profiles,
which should be a subject to a more accurate analysis. The conducted analysis
indicates the possibility of using the presented approach based on virtual and real time
during the research concerning behavior patterns related to the usage of multiplayer
platforms. The combination of dynamic time warping and classification methods in a
virtual and real time enables the introduction of quantitative measures of similarity in
relation to the ideal vector. The research conducted based on empirical data showed
that in the initial phase of platform usage it is possible to identify behavior patterns,
which indicate the usage of application in the future. The presented approach
supplement the current proposed methods focused on users analysis. It allows the
comparison of sequences in relation to the ideal vector or reference to the behavior of
other users with the system usage patterns. Using this method, the identification of
users that are at risk of stopping service usage is possible. It can help to introduce
appropriate intervention actions to increase activeness and frequency of usage.
6 Summary
In parallel to the increase of Internet systems’ complexness and the increasing
competition in this sector, there is a bigger demand for introduction of new data
analysis methods, and for examining the behavior of users participating in Internet
services. The proper recognition of needs and behavior tendencies provides the basis
for making rational decisions and better adaptation of online services to users’ needs.
High dynamics and changeability of users’ behavior show the need for
implementation of solutions that take time characteristics into account. The solutions
presented so far in the literature did not include two-dimensional approach towards
data character and time series. The approach presented in this paper enables
quantitative estimation of users’ behavior with the use of reference sequences and
users’ segmentation based on time conditioning. The applied methods of dynamic
time warping enable quantitative presentation of similarity including usage patterns.
The presented research proves the possibility of using this approach in different fields
and initiating new research areas focused on recognition of audiences characteristics
with bi-temporal representation.
Extended version of paper: Jankowski J., Analysis of Multiplayer Platform Users Activity Based on the
Virtual and Real Time Dimension Lecture Notes in Computer Science Volume 6984, 2011, pp 312-315
References
1. Acar, A.S., Polonsky, M.: Online Social Networks and Insights into Marketing
Communications, Journal of Internet Commerce. 6(4), 55--72 (2008)
2. Antunes, C.M., Arlindo, L.O.: Temporal Data Mining: an overview. Workshop on
Temporal Data Mining. In: 7th ACM SIGKDD International Conference on Knowledge
Discovery and Data Mining, pp. 1--13, San Francisco (2001)
3. Beitzel, S.M., Jensen, E.C., Chowdhury, A., Grossman, D., Frieder, O.: Hourly Analysis of
a Very Large Topically Categorized Web Query Log. In: ACM SIGIR 2004 Workshop on
Information Retrieval in Context, pp. 321--328, Sheffield (2004)
4. Breiman, L., Friedman J., Olshen R., Stone C.: Classification and Regression Trees.
Chapman and Hall. New York (1984)
5. Chau, M., Zeng, D., Chen, M., Huang, M., Hendriawan, D.: Design and Evaluation of a
Multi-Agent Collaborative Web Mining System. Decision Support Systems. 35(1), 167--
183 (2003)
6. Chun-Ling, Z., Zun-Feng, L., Jing-Rui, Y.: The Application Research on Web Log Mining
in E-Marketing. In: 2nd International Conference e-Business and Information System
Security, pp. 1--4, Wuhan (2010)
7. Ducheneaut, N., Yee, N., Nickel, E., Moore, R.: Alone Together - Exploring the Social
Dynamics of Massively Multiplayer Online Games. In: Proceedings of ACM CHI 2006
Conference on Human Factors, pp. 407-416, Quebec (2006)
8. Dick, A.S., Basu, K.: Customer Loyalty: Toward an Integrated Conceptual Framework.
Journal of the Academy of Marketing Science. 22(2), 99--113 (1994)
9. Eirinaki, M., Lampos, H., Vazirgiannis, M., Varlamis, I.: Sewep: Using Site Semantics and
a Taxonomy to Enhance the Web Personalization Process. In: ACM conference on
Knowledge Discovery in Data, pp. 99--108, Washington (2003)
10. Gomes, M.F., Canuto, A.M.: Carcara: A Multi-agent System for Web Mining Using
Adjustable User Profile and Dynamic Grouping. In: IEEE/WIC/ACM International
Conference on Intelligent Agent Technology, pp. 187--190, Hong-Kong (2006)
11. Gürsel, A., Sen, S.: Producing Timely Recommendations From Social Networks. In:
Decker, S., Sichman, J., Sierra, C., Castelfranchi, C. (eds.) Proceedings of 8th International
Conference on Autonomous Agents and Multiagent Systems, pp.10--5, Budapest (2009)
12. Hawwash, B., Nasraoui, O.: Mining and Tracking Evolving Web User Trends From Large
Web Server Logs. Statistical Analysis and Data Mining. 3(2), 106--125 (2010)
13. He, J., Chu, W.W.: A Social Network Based Recommender System. Annals of Information
Systems. Special Issue on Data Mining for Social Network Data, 12, 47--74 (2009)
14. Hernandez-Ortega, B., Jiménez-Martinez, J., Martín-DeHoyos, M.J.: Differences Between
Potential, New and Experienced e-Customers: Analysis of e-Purchasing Behavior. Internet
Research, 18(3), 248--265 (2008)
15. Herring, S.C., Scheidt, L.A., Wright, E., Bonus, S.: Weblogs as a Bridging Genre.
Information Technology & People, 18(2), 142--171 (2005)
16. Hu, X., Yin, Y., Zhang, B.: Mining Temporal Web Interesting Patterns. In: International
Conference on Computational Intelligence and Security, pp. 227--231, Heilongjiang (2007)
17. Jung, J.J.: Visualizing Recommendation Flow on Social Network. Journal of Universal
Computer Science. 11(11), 1780--1791 (2005)
18. Karahanna, E., Straub D.W., Chervany, N.L.: Information Technology Adoption Across
Time: a Cross-Sectional Comparison of Pre-Adoption and Post-Adoption Beliefs. MIS Q.
23, 183--213, (1999)
19. Kish S.: Second Life: Virtual Worlds and the Enterprise, http://www.lunchoverip.com
/2007 /10/ second-life-vir.html (2007)
20. Lu, H.P., Wang S.: The Role of Internet Addiction in Online Game Loyalty: an Exploratory
Study. Internet Research. 18(5), 499--519 (2008)
21. Marsili, M., Slanina, F., Vega-Redondo, F.: Dynamics of Social Networks (The Rise and
Fall of a Networked Society). In: Proceedings of the National Academy of Sciences. 101,
1439--1442 (2004)
22. Mobasher B., Cooley R., Srivastava J.: Automatic Personalization Based on Web Usage
Mining. Communications of the ACM. 43(8), 142--151 (2000)
23. Nasraoui, O., Krishnapuram, R.: One Step Evolutionary Mining of Context Sensitive
Associations and Web Navigation Patterns. In: SIAM conference on Data Mining, pp. 531-
-547, Arlington (2002)
24. Newman, M.E.J.: Finding Community Structure in Networks Using the Eigenvectors of
Matrices. Physical Review. 74(3), p.036104 (2006)
25. Rabiner, L. R., Juang, B.: Fundamentals of Speech Recognition, Prentice-Hall, 1993
26. Radas, S., Shugan S. M.: Seasonal Marketing and Timing Introductions. Journal of
Marketing Research. 35(3), 296-315 (1998)
27. Rho, J., Moon, B., Kim, Y., Yang, D.: Internet Customer Segmentation. Using Web Log
Data. Journal of Business & Economics Research. 2(11), 59--74 (2004)
28. Roddick, J.F., Myra S.: A Survey of Temporal Knowledge Discovery Paradigms and
Methods, IEEE Transactions on Knowledge and Data Engineering. 14(4), 750--767 (2002)
29. Sweetster, P., Wyeth, P.: GameFlow: A Model for Evaluating Player Enjoyment in Games.
ACM Computer and Entertainment. 3(3), 1--23 (2005)
30. Tsai, H., Huang, H.: Determinants of e-Repurchase Intentions: An Integrative Model of
Quadruple Retention Drivers. Information & Management. 44, 231--239 (2007)
31. Venkatesh, V., Morris, M. G., Davis, G. B., Davis, F. D.: User acceptance of information
technology: Toward a unified view, MIS Quarterly. 27(3), 425--478 (2003)
32. Virtual World Types: Navigating the VW Jungle, http://www.artesia.si/blog/ categories/
Virtual-worlds (2008)
33. Virtual Worlds for Kids, http://www.virtualworldsmanagement.com/2009/youth-01-26-
2009.html (2009)
34. Wagner, C., Bolloju, N.: Supporting Knowledge Management in Organizations with
Conversational Technologies. Journal of Database Management. 16(2), 1--8 (2005)
35. Wangenheim, F., Bayon, T.: Satisfaction, Loyalty and Word of Mouth Within the Customer
Base of Utility Provider. Journal of Consumer Behaviour. 3(3), 211--220 (2004)
36. Zhang, Y., Jansen, B.J., Spink, A.: Time Series Analysis of a Web Search Engine
Transaction Log. Information Processing and Management. 45(2) , 230--245 (2009)