Social dynamics are based on human needs for trust, support, resource sharing, irrespective of whether they operate in real life or in a virtual setting. Massively multiplayer online role-playing games (MMORPGS) serve as enablers of leisurely social activity and are important tools for social interactions. Past research has shown that socially dense gaming environments like MMORPGs can be used to study important social phenomena, which may operate in real life, too. We describe the process of social exploration to entail the following components 1) finding the balance between personal and social time; 2) making a choice between a large number of weak ties or few strong social ties; 3) finding a social group. In general, these are the major determinants of an individual’s social life. This research looks into the phenomenon of social exploration in an activity-based online social environment. We study this process through the lens of the following research questions, 1) What are the different social behavior types? 2) Is there a change in a player’s social behavior over time? 3) Are certain social behaviors more stable than others? 4) Can longitudinal research of player behavior help shed light on the social dynamics and processes in the network? We use an unsupervised machine learning approach to come up with 4 different social behavior types - Lone Wolf, Pack Wolf of Small Pack, Pack Wolf of a Large Pack and Social Butterfly. The types represent the degree of socialization of players in the game. Our research reveals that social behaviors change with time. While lone wolf and pack wolf of small pack are more stable social behaviors, pack wolf of large pack and social butterflies are more transient. We also observe that players progressively move from large groups with weak social ties to settle in small groups with stronger ties.
Unsupervised Machine Learning Approach to Analyzing Social Behaviors in an Online Multi-Player Game
1. Unsupervised Machine Learning Approach to
Analyzing Social Behaviors in an Online Multi-
Player Game
An Empirical Study of Social Behaviors and Social Exploration in Multiplayer
Online Game.
Arpita Chandra
Dual Degree Thesis Defense
Center for Exact Humanities
IIIT – Hyderabad
Committee Members
Dr. Ponnurangam Kumaraguru (Advisor)
Dr. Nimmi Rangaswamy
Dr. Radhika Krishnan
1
2. Outline
BACKGROUND AND SETUP ANALYSIS
› EverQuest II (MMORPG)
› Construction of group instances
› Defining feature set & importance of intuitive clusters
› Clustering (k-means)
RESEARCH MOTIVATION SOCIAL BEHAVIORS
DATASET AND METHODOLOGY ANALYSIS
1
2
3
4
› Studying social behaviors
› Games as social networks
› Importance from a gaming perspective
› Defining 4 types of social behaviors
› Characteristics for each
› Social settling in – from large-weak tied groups to
smaller-strong tied ones.
› Relationship between social behaviors and churn
behavior (attrition).
2
4. Social Exploration
What is the split between social time
and alone time?
Extent of socialization
How many people do they want to
socialize with?
Breadth of socialization
What kind of ties do they want with
their neighbors - weak or strong?
Depth of socialization
How much social
activity?
Lone time vs Social
time
How many people
to socialize with?
Small Group vs
Large Group
What is the quality of
social ties??
Several weak ties vs
few strong ties
4
5. Social Exploration
in Literature
o Choosing work-group members, during the exploration phase, points of contact
are many for information seeking, learning.
o Social exploration to find an expert in social expert – “An extensive set of
experiments shows that the analysis of social activities, social relationships, and
socially shared contents helps improving the effectiveness of an expert finding
system.”
o E.W Morisson studied patterns of relationships, learning, integration into a social
group of new comers in a work place and when an individual needs to change
their social group
5
6. Research Questions
o What are different kinds of social behaviors?
o Do players change social behavior overtime?
o Are patterns of behavior change indicative of any social
phenomenon at play?
6
7. Gaming Industry estimated to be $116 billion in 2017.
$135 billion in 2018. Almost 11% increase.
Gaming networks are activity based environments.
Provide active socialization.
55.2 million online social gamers only in the United
States of America as of 2016.
Games as social networks
7
Source - https://www.wepc.com/news/video-game-statistics/#online-gaming
8. Conventional gaming systems are also going social.
Xbox live, PlayStation plus
Encouraging socialization
in games
8
9. What roles
do games play in real life?
Power and future of gaming: her research reveals
how gamers have become expert problem solvers
and collaborators
We can use games design to socially positive
ends, be it in our own lives, our communities or
our businesses.
Video games allow us to create social bonds and
connect with others at a level some of us can’t
reach in real life.
9
11. Games as warehouses
of social data
Huge player base ~ more than a billion users
worldwide
Provide rich social data
Online games are a good indicator of human
behavior in the real world.
11
12. EVERQUEST II
o Originally developed by Sony
Online Entertainment, Nov
2004.
o As of Jan, 2005, over 300,000
active users.
o Had a subscription based
model during it’s initial years.
o Subscription packs available
for 1,3,6,12 months.
12
13. Player Activity Logs
Took player logs for 4 consecutive weeks for the analysis.
o 4 weeks is the natural cycle for a subscription-based model in
Everquest II.
o Analyze social behavior transformation on a weekly basis for 4
weeks.
13
14. RQ1: What are different types of social
behaviors?
14
15. Methodology
Identify group
instances from player
logs
STEP 1
Feature selection
STEP 2
Clustering
STEP 3
Parse through
the player logs
sequentially and
identify entries
with same
log_date,
group_size,
server_name,
location_id
Make group
instances
Calculating
different in-
game metrics
for socialization,
enagagement
Extracting
feature list
Clustering using
k-means with k
from the
previous step
and feature list
from step 2
Determining the
number of
naturally
occurring
clusters (k) using
the elbow
method
15
16. Group Instances
A group instance consisted of entries that logged the same values of server name, log time, location id, and
group size
16
17. Engagement
And Socialization
Metrics
o No. of Group Sessions: This feature
denotes the total no. of sessions
participated in by player in groups.
o No of Lone sessions: Sessions played
alone.
o Total no. of sessions : Sum of group
sessions and lone sessions.
o No. of neighbors: Unique number of
players played with .
Total observed history (T) - 4 weeks.
Unit of analysis (∆t) – 1 week
What comprises a game session?: Series of in-
game activity separated by no more than 30
minutes
17
18. Noise Reduction
o We only take engaged social players
who have
total no. of sessions > threshold value, τ
for each week, defined as:
τ = μ(log (total no. of sessions)) – σ (log(total
no. of sessions)
𝜏
= 𝜇(𝑙𝑜𝑔(Total no. of sessions))
− 𝜎(𝑙𝑜𝑔 𝑡𝑜𝑡𝑎𝑙 𝑛𝑜. 𝑜𝑓 𝑔𝑟𝑜𝑢𝑝 𝑠𝑒𝑠𝑠𝑖𝑜𝑛𝑠 )
𝜇 is mean and σ is standard deviation
Week 14
18
19. Importance
of intuitive clusters
o Computationally, clustering reduces the dimensionality of the dataset.
o A number of studies in literature stress upon the importance of producing clusters that are
easy to interpret and are intuitive within the framework of the dataset.
o Rather than using traditional methods for reducing feature dimensionality like PCA, an
interpretable and intuitive feature set was recognized to reflect the social connectedness of
players
19
20. Defining Feature Set
o No of neighbors: Unique players a players plays with.
o Fraction of group sessions: Ratio of group sessions to total number of sessions
o Average Tie Strength:
𝑇𝑜𝑡𝑎𝑙 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐺𝑟𝑜𝑢𝑝 𝑆𝑒𝑠𝑠𝑖𝑜𝑛𝑠
𝑇𝑜𝑡𝑎𝑙 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑁𝑒𝑖𝑔ℎ𝑏𝑜𝑟𝑠
20
21. Clustering
o Used k means to cluster around the
features.
o Used the elbow method or
minimum within cluster sum of
squared errors to determine
number of clusters (k).
21
23. Types of Social Behaviors
Based on centroids from k-means
23
24. Social Behavior
Types
o Lone Wolf (LW)
o Pack Wolf of Small Pack (PWS)
o Pack Wolf of Large Pack (PWL)
o Social Butterfly (SB)
24
25. Social Behavior Types
Lone Wolf (LW)
Pack Wolf of Small Pack
(PWS)
Pack Wolf of Large Pack
(PWL)
Social Butterfly (SB)
• Spends more time
playing alone.
• Interacts with a very
small number of
players. (mean =
6.82)
• Has a high tie
strength with their
neighbors. (2.63)
• Interacts with small
number of players
(mean = 26.35)
• Tie strength is lower
as compared to lone
wolves. (mean =
0.87)
• Interacts with a large
number of players.
(mean = 58.86)
• Plays in groups and
spends less time
playing alone
• Tie strength is lower
than first two. (mean
= 0.41)
• Interacts with a very
high number of
players. (mean =
115.04)
• Tie strength is low.
(mean = 0.29)
• Has a lot of weak
connections.
25
26. Social Behavior
Types
o Lone Wolf (LW)
o Pack Wolf of Small Pack (PWS)
o Pack Wolf of Large Pack (PWL)
o Social Butterfly (SB)
No. of neighbors
increases
26
27. Social Behavior
Types
o Lone Wolf (LW)
o Pack Wolf of Small Pack (PWS)
o Pack Wolf of Large Pack (PWL)
o Social Butterfly (SB)
Tie strength
decreases
27
30. Behavior Change
1862 players analyzed Some behaviors are more
resilient to change.
Change in behaviors from
week 14 to week 17
LW PWS PWL SB
LW 74.90% 18.70% 5.00% 1.30%
PWS 43.70% 37.70% 15.70% 2.90%
PWL 19.20% 37.30% 32.90% 10.50%
SB 14.00% 29.00% 39.00% 18.00%
Week 14 labels
Week 17 labels
30
31. Social Behavior
Types
o Lone Wolf (LW)
o Pack Wolf of Small Pack (PWS)
o Pack Wolf of Large Pack (PWL)
o Social Butterfly (SB)
Propensity to
change behavior
increases.
31
32. RQ3: Can longitudinal research of in-game social
behaviors shed light on the social dynamics that
exist in the network ?
Bpaths: What do transition of social behaviors look like?
32
33. o A behavior path (Bpath) for a player X may look something like this for a
period of 4 weeks. e.g
BpathX = (PWS, PWS, LW, LW)
o Each entry corresponds to behavior in the respective week.
Behavior Pathways
Bpath
33
36. Towards a
smaller group and stronger ties
Lone Wolf Pack Wolf of Small Pack Pack Wolf of Large Pack Social Butterfly
o 75% remained the
same.
o Stable behavior
o 56.5% exhibited
behaviors remained
same or became less
social.
o 49% starting out as
PWL became less
social but not LW
o 22% either remained
PWL or shifted SB.
o Like high
socialization.
o 82% of the players
who started out as
SB became less
social.
36
37. Benefits of
high socialization
o Degree of socialization is large
o Points of contact are many
o Learn skills and strategies from different players and groups
o Social Exploration
o Information Diffusion
Also in line with existing literature on social exploration; most people start off by interacting with many people for information
gain, learning and familiarity.
37
38. Social Behavior and Churn
How are social behaviors linked with churning behavior or attrition?
38
39. What is Churn?
o Customer churn is when an existing customer, user, player,
subscriber leaves the network or stops ends the relationship with a
company.
39
40. Churn Behavior
Week 14 to 15
o We see that lone wolves have the
highest churn or attrition rate
amongst all.
o The same trends are seen for all
the weeks of the observed history.
o This supports the common notion
of
“player engages player”
40
43. Churn Behavior –
Low Socialization
o Less social responsibility towards other players to return to the
game.
o The social influence of other players or groups is also limited.
o Strong group cohesiveness; high tie strength. Even the slight
undercurrents that bring about a change in the group structure
or dynamics can have a big impact.
43
44. Churn Behavior -
High socialization
o This could mean that these players are in the early stages of the
social exploration process; actively looking for a group to settle
down with.
o High number of neighbors; many options to play game sessions
with even if a few churn.
44
45. Related Publications
o Chandra, Arpita, Zoheb Borbora, Ponnurangam Kumaraguru, and Jaideep
Srivastava. "Finding Your Social Space: Empirical Study of Social Exploration
in Multiplayer Online Games." (2019).
o Borbora, Zoheb, Arpita Chandra, Ponnurangam Kumaraguru, and Jaideep
Srivastava. "On Churn and Social Contagion." networks 26 (2019): 29.
45
47. Summary
o Examined social behaviors in an MMORPG - Everquest II.
o Came up with behavior types based on the characteristics of k-means centroids.
o Examined behaviors longitudinally & uncovering social phenomena.
o The methodology can be easily replicated for other games and activity based networks.
o Trends in behavior change, suggest a process of ’social settling in or social exploration’
operating in the network.
o The analysis also reveals that there is a link between social behaviors and churn behavior.
As such, this supports the common notion that "player engages player” and that
socialization is an important factor for retaining players
47
48. Limitations
o A larger sample size would have been instrumental in making
better assessments on behavior evolution and statistically
significant most traversed Bpaths.
o Furthermore, the analysis was done on one dataset. However,
this research could be easily reproduced to study social
interactions on similar platforms that facilitate social interactions.
48
49. Future Work
o Finding out the optimal time duration that the process of social
settling in takes.
o Investigate whether different social personality types take
different amounts of time to settle into a social group.
49
50. Acknowledgements
o I would like to express my heartfelt gratitude to Prof P.K for being
an amazing ‘Guru’ and mentor in this endeavor. Sir, here’s to you!
o Also, a sincere word of thanks to Dr. Jaideep Srivastava and Dr.
Zoheb Borbora for their continued support and being wonderful
people to work with. I interned with Dr. Srivastava in my
sophomore year.
o Last but not the least, I would also like to remember Prof.
Navjyoti and wish he was here to see me defend my thesis.
50
51. References
1. Chandra, Arpita, et al. "Finding Your Social Space: Empirical Study of Social Exploration in Multiplayer Online
Games." (2019).
2. Hinds, Pamela J., et al. "Choosing work group members: Balancing similarity, competence, and
familiarity." Organizational behavior and human decision processes 81.2 (2000): 226-251.
3. Bozzon, Alessandro, et al. "Choosing the right crowd: expert finding in social networks." Proceedings of the 16th
International Conference on Extending Database Technology. ACM, 2013.
4. Morrison, Elizabeth Wolfe. "Newcomers' relationships: The role of social network ties during
socialization." Academy of management Journal 45.6 (2002): 1149-1160.
5. Williams, Dmitri, et al. "The virtual worlds exploratorium: Using large-scale data and computational techniques for
communication research." Communication Methods and Measures 5.2 (2011): 163-180. - ”In a partnership with a
corporation that hosts an MMO, a 20-person team of scholars is engaged in the study of behavior within a game
and also game activities that parallel those in “real life” (e.g., economic activity, social networking, group
processes). ”
6. Romero, Margarida, Mireia Usart, and Michela Ott. "Can serious games contribute to developing and sustaining
21st century skills?." Games and Culture 10.2 (2015): 148-177.
7. Shim, Kyong Jin, and Jaideep Srivastava. "Behavioral profiles of character types in EverQuest II." Proceedings of the
2010 IEEE Conference on Computational Intelligence and Games. IEEE, 2010.
51
52. References
8. Bauckhage, Christian, Anders Drachen, and Rafet Sifa. "Clustering game behavior data." IEEE Transactions on Computational
Intelligence and AI in Games 7.3 (2014): 266-278. - The goal of user behavior analysis in game business intelligence is an
interpretable representation of the data at hand and the patterns residing in the data, as the representation basically has to assist a
human in analyzing huge amounts of game data
9. MLADrachen, Anders, et al. "Guns, swords and data: Clustering of player behavior in computer games in the wild." 2012 IEEE
conference on Computational Intelligence and Games (CIG). IEEE, 2012. – “Interpretability which is important in a practical
development context, where the results of a clustering analysis should be as easy as possible to interpret”
10. Borbora, Zoheb, Arpita Chandra, Ponnurangam Kumaraguru, and Jaideep Srivastava. "On Churn and Social
Contagion." networks 26 (2019): 29.
11. https://www.wepc.com/news/video-game-statistics/#online-gaming
12. Farhangi, Sanaz. "Reality is broken to be rebuilt: how a gamer’s mindset can show science educators new ways of contribution to
science and world?." (2012): 1037-1044.
13. Kawale, Jaya, Aditya Pal, and Jaideep Srivastava. "Churn prediction in MMORPGs: A social influence based approach." 2009
International Conference on Computational Science and Engineering. Vol. 4. IEEE, 2009.
52
Literature talks about the process of social exploration in different contexts, often particular to the type of social network.
Hinds, Pamela J., et al. "Choosing work group members: Balancing similarity, competence, and familiarity." Organizational behavior and human decision processes 81.2 (2000): 226-251.
Bozzon, Alessandro, et al. "Choosing the right crowd: expert finding in social networks." Proceedings of the 16th International Conference on Extending Database Technology. ACM, 2013.
Morrison, Elizabeth Wolfe. "Newcomers' relationships: The role of social network ties during socialization." Academy of management Journal 45.6 (2002): 1149-1160.
Studying social exploration through the lens of social behaviors.
Different from passive socialization.
Active social networks.
* Source - https://www.wepc.com/news/video-game-statistics/#online-gaming
These various “mechanics,” or game incentive schemes (Sellers, 2006), prompt players to act together and give rise to a large number of social and communication phenomena worth studying. - The Virtual Worlds Exploratorium: Using Large-Scale Data and Computational Techniques for Communication Research (paper)
Jane McGonigal.
It’s no coincidence that Youtube’s biggest channel is a gamer.
More research that games like MMORPGs can teach valuable skills which can be used in real life.
Williams, Dmitri, et al. "The virtual worlds exploratorium: Using large-scale data and computational techniques for communication research." Communication Methods and Measures 5.2 (2011): 163-180. - ”In a partnership with a corporation that hosts an MMO, a 20-person team of scholars is engaged in the study of behavior within a game and also game activities that parallel those in “real life” (e.g., economic activity, social networking, group processes). ”
Romero, Margarida, Mireia Usart, and Michela Ott. "Can serious games contribute to developing and sustaining 21st century skills?." Games and Culture 10.2 (2015): 148-177.
Sources: gameindustry.biz, gamasutra
Subscription available for 1,3,6,12 months.
4 weeks were randomly chosen from the dataset.
Session – based on the definition of session in previous studies based on the same dataset.
4 weeks were randomly chosen from the dataset. 4 weeks is a natural cycle of a subscription based model like in everquest. In order to study behavior change we wanted same players. A lot of players churn after their subscription ends.
Shim, Kyong Jin, and Jaideep Srivastava. "Behavioral profiles of character types in EverQuest II." Proceedings of the 2010 IEEE Conference on Computational Intelligence and Games. IEEE, 2010.
Kawale, Jaya, Aditya Pal, and Jaideep Srivastava. "Churn prediction in MMORPGs: A social influence based approach." 2009 International Conference on Computational Science and Engineering. Vol. 4. IEEE, 2009.
Bauckhage, Christian, Anders Drachen, and Rafet Sifa. "Clustering game behavior data." IEEE Transactions on Computational Intelligence and AI in Games 7.3 (2014): 266-278. - The goal of user behavior analysis in game business intelligence is an interpretable representation of the data at hand and the patterns residing in the data, as the representation basically has to assist a human in analyzing huge amounts of game data
MLADrachen, Anders, et al. "Guns, swords and data: Clustering of player behavior in computer games in the wild." 2012 IEEE conference on Computational Intelligence and Games (CIG). IEEE, 2012. – “Interpretability which is important in a practical development context, where the results of a clustering analysis should be as easy as possible to interpret”
Week 14
Fraction of group session: group time vs lone time.
50% players changed their behavior from week 1 to week 2.
plotted the most common traversed Bpaths by the players. The empirically seen top 4 Bpaths for each behavior type as starting state are shown. Smaller number were insignificant and hence not shown
Most people are not coming back if more social. LW is the only one where they become more social and return back.
It is rare that a less social behavior transitions into a more social behavior. This suggests a social dynamic which seems to indicate a process of social settling in
Converting to a more social state might be because some quests are particularly hard to complete alone or in small groups.
Choose which social group they want to be a part of
Also consistent with seeking behavior of newcomers in organizational behavior. -
Hinds, Pamela J., et al. "Choosing work group members: Balancing similarity, competence, and familiarity." Organizational behavior and human decision processes 81.2 (2000): 226-251.
Bozzon, Alessandro, et al. "Choosing the right crowd: expert finding in social networks." Proceedings of the 16th International Conference on Extending Database Technology. ACM, 2013.
Morrison, Elizabeth Wolfe. "Newcomers' relationships: The role of social network ties during socialization." Academy of management Journal 45.6 (2002): 1149-1160.
Borbora, Zoheb, Arpita Chandra, Ponnurangam Kumaraguru, and Jaideep Srivastava. "On Churn and Social Contagion." networks 26 (2019): 29.
This requires an investment of time and returning to the game to continue till the process of settling down and social exploration doesn’t end.
Borbora, Zoheb, Arpita Chandra, Ponnurangam Kumaraguru, and Jaideep Srivastava. "On Churn and Social Contagion." networks 26 (2019): 29.
This requires an investment of time and returning to the game to continue till the process of settling down and social exploration doesn’t end.
Borbora, Zoheb, Arpita Chandra, Ponnurangam Kumaraguru, and Jaideep Srivastava. "On Churn and Social Contagion." networks 26 (2019): 29.
Social behaviors characterized by high degree of socialization have lower propensity to churn than social behaviors characterized by less degree of socialization.
Social behaviors characterized by high degree of socialization have lower propensity to churn than social behaviors characterized by less degree of socialization.