Modeling and Analyzing Social Networks of Games

1. 1/80 Modeling and Analyzing Social Networks of Games Modeling and Analyzing Social Networks of Games Leonardo Mauro Pereira Moraes Advisor: Robson Leonardo Ferreira Cordeiro leonardo.mauro@usp.br, robson@icmc.usp.br Institute of Mathematics and Computer Sciences (ICMC) University of Sao Paulo (USP) May 08, 2020 Research institutes Research funding

2. 2/80 Modeling and Analyzing Social Networks of Games 1 Introduction 2 Dataset 3 Detecting Digital Influencers 4 Attractive Game Characteristics 5 Conclusion

3. 3/80 Modeling and Analyzing Social Networks of Games Introduction 1 Introduction Context Motivation Hypothesis 2 Dataset 3 Detecting Digital Influencers 4 Attractive Game Characteristics 5 Conclusion

4. 4/80 Modeling and Analyzing Social Networks of Games Introduction Context Context Digital games Game is developed for electronic devices, e.g., computer, mobile, video game console. applied to entertainment, education, health, etc. One of the most popular forms of entertainment, reaching millions of players. Universe of games is in constant ascendancy. both in production and in consumption. revenues of more than U$160 billion in 20191 . 1Source: https://newzoo.com/insights/articles/newzoos-trends-to-watch-in-2020-games-esports-and-mobile/.

5. 4/80 Modeling and Analyzing Social Networks of Games Introduction Context Context Digital games Game is developed for electronic devices, e.g., computer, mobile, video game console. applied to entertainment, education, health, etc. One of the most popular forms of entertainment, reaching millions of players. Universe of games is in constant ascendancy. both in production and in consumption. revenues of more than U$160 billion in 20191 . 1Source: https://newzoo.com/insights/articles/newzoos-trends-to-watch-in-2020-games-esports-and-mobile/.

6. 5/80 Modeling and Analyzing Social Networks of Games Introduction Context Context Universe of games eSports: professional competitions. Streamers: players who produce online videos. Digital influencers: professional players and streamers. Why games? Industry interest. Research, e.g., Knowledge Discovery in Databases. Players relate to the games. e.g., buy, play, like, comment; such as Social Network. Understand player preferences. e.g., what are the characteristics of attractive games?

7. 5/80 Modeling and Analyzing Social Networks of Games Introduction Context Context Universe of games eSports: professional competitions. Streamers: players who produce online videos. Digital influencers: professional players and streamers. Why games? Industry interest. Research, e.g., Knowledge Discovery in Databases. Players relate to the games. e.g., buy, play, like, comment; such as Social Network. Understand player preferences. e.g., what are the characteristics of attractive games?

8. 6/80 Modeling and Analyzing Social Networks of Games Introduction Context Social Networks Social Network of Games (SNG) Interaction: user → game e.g., play A set of Social Networks. ... c1 c2 c3 c4 c5 c6 Figure: Network of Games Also, SNGs can change over time...

9. 6/80 Modeling and Analyzing Social Networks of Games Introduction Context Social Networks Social Network of Games (SNG) Interaction: user → game e.g., play A set of Social Networks. ... c1 c2 c3 c4 c5 c6 Figure: Network of Games Also, SNGs can change over time...

10. 6/80 Modeling and Analyzing Social Networks of Games Introduction Context Social Networks Social Network of Games (SNG) Interaction: user → game e.g., play A set of Social Networks. p1 c3 p2 c1 c2 Gplay Figure: Graphs of interactions Also, SNGs can change over time...

11. 6/80 Modeling and Analyzing Social Networks of Games Introduction Context Social Networks Social Network of Games (SNG) Interaction: user → game e.g., play, like, etc. A set of Social Networks. p1 c3 p2 c1 c2 p1 c1 p2 c2 c3 ... Glike Gplay Figure: Graphs of interactions Also, SNGs can change over time...

14. 6/80 Modeling and Analyzing Social Networks of Games Introduction Context Social Networks Social Network of Games (SNG) Interaction: user → game e.g., play, like, etc. A set of Social Networks. p1 c1 p2 c2 c3 Glike Figure: Changes over time Also, SNGs can change over time...

15. 6/80 Modeling and Analyzing Social Networks of Games Introduction Context Social Networks Social Network of Games (SNG) Interaction: user → game e.g., play, like, etc. A set of Social Networks. p1 c1 p2 c2 c3 Time ti Figure: Changes over time Also, SNGs can change over time...

16. 6/80 Modeling and Analyzing Social Networks of Games Introduction Context Social Networks Social Network of Games (SNG) Interaction: user → game e.g., play, like, etc. A set of Social Networks. p1 c1 p2 c2 c3 p1 c1 p2 c2 c3 c4 Time ti ti+1 Figure: Changes over time Also, SNGs can change over time...

17. 7/80 Modeling and Analyzing Social Networks of Games Introduction Motivation Motivation Understand player preferences in a SNG. i.e., what are the characteristics of attractive games? Support game developers. Hard task involving technical and aesthetic aspects. Focus on games of platform type.

18. 8/80 Modeling and Analyzing Social Networks of Games Introduction Motivation Motivation Games of platform, e.g., Super Mario Bros series Figure: Super Mario World. Source: Nintendo2 . 2Source: http://mario.nintendo.com/.

19. 9/80 Modeling and Analyzing Social Networks of Games Introduction Hypothesis Hypothesis (1) Dataset 1 Find/elaborate a Social Network of Games dataset. 2 Identify influencer makers ∼ digital influencer. to avoid bias in the future process. 3 Identify common characteristics of successful games. Due to the diversity, each area contains its related works.

20. 9/80 Modeling and Analyzing Social Networks of Games Introduction Hypothesis Hypothesis (2) Player Filtering (1) Dataset 1 Find/elaborate a Social Network of Games dataset. 2 Identify influencer makers ∼ digital influencer. to avoid bias in the future process. 3 Identify common characteristics of successful games. Due to the diversity, each area contains its related works.

21. 9/80 Modeling and Analyzing Social Networks of Games Introduction Hypothesis Hypothesis (2) Player Filtering (1) Dataset (3) Analysis 1 Find/elaborate a Social Network of Games dataset. 2 Identify influencer makers ∼ digital influencer. to avoid bias in the future process. 3 Identify common characteristics of successful games. Due to the diversity, each area contains its related works.

22. 9/80 Modeling and Analyzing Social Networks of Games Introduction Hypothesis Hypothesis (2) Player Filtering (1) Dataset (3) Analysis 1 Find/elaborate a Social Network of Games dataset. 2 Identify influencer makers ∼ digital influencer. to avoid bias in the future process. 3 Identify common characteristics of successful games. Due to the diversity, each area contains its related works.

23. 10/80 Modeling and Analyzing Social Networks of Games Dataset 1 Introduction 2 Dataset Related Work SMMnet Characteristics Discussion 3 Detecting Digital Influencers 4 Attractive Game Characteristics 5 Conclusion

24. 11/80 Modeling and Analyzing Social Networks of Games Dataset Related Work Related Work Given the popularity of the games, there are several game datasets available: From market games... World of Warcraft [Lee et al. (2011)] StarCraft [Lin et al. (2017)] League of Legends [Aung et al. (2018)] Independent games... Platformer Experience [Karpouzis et al. (2015)] Heroes of Elibca [Lim and Harrell (2015)]

25. 12/80 Modeling and Analyzing Social Networks of Games Dataset Related Work Problem Problem However, how to work with SNG? There is no dataset, so far. Proposal A popular SNG with different games and many user. Extract a data history of player-game interactions. e.g., likes, plays, etc.

26. 12/80 Modeling and Analyzing Social Networks of Games Dataset Related Work Proposal Problem However, how to work with SNG? There is no dataset, so far. Proposal A popular SNG with different games and many user. Extract a data history of player-game interactions. e.g., likes, plays, etc.

27. 13/80 Modeling and Analyzing Social Networks of Games Dataset SMMnet Super Mario Maker (SMM) Released September 2015 Nintendo, Tokyo, Japan; Players can: play, like a game level; create their own levels. etc. four different styles; and four game difficulties. Figure: Super Mario Maker If you played every level in #SuperMarioMaker for 1 minute each, it would take you nearly 14 years to play them all! (Nin- tendo of America3 ) - [+7.2M game levels] 3Source: https://twitter.com/NintendoAmerica/status/732624228428750848/.

28. 13/80 Modeling and Analyzing Social Networks of Games Dataset SMMnet Super Mario Maker (SMM) Released September 2015 Nintendo, Tokyo, Japan; Players can: play, like a game level; create their own levels. etc. four different styles; and four game difficulties. Figure: Super Mario Maker If you played every level in #SuperMarioMaker for 1 minute each, it would take you nearly 14 years to play them all! (Nin- tendo of America3 ) - [+7.2M game levels] 3Source: https://twitter.com/NintendoAmerica/status/732624228428750848/.

29. 14/80 Modeling and Analyzing Social Networks of Games Dataset SMMnet Super Mario Maker (SMM) (a) Super Mario Bros (b) Super Mario Bros 3 (c) Super Mario World (d) Super Mario Bros U Figure: Levels styles - Games of platform

30. 15/80 Modeling and Analyzing Social Networks of Games Dataset SMMnet Data Collection SMM metadata is available in a website4 . Crawler5 : 1 smm-course-search: searches for game IDs, and stores into Database (DB). 2 super-mario-maker-client and smm-maker-profile: query the game IDs and collect the data from SMM Bookmark. SMM Bookmark DB smm-course-search smm-maker-profile super-mario-maker-client Crawler 4Source: https://supermariomakerbookmark.nintendo.net/. 5Source: https://www.npmjs.com/~leomaurodesenv.

31. 16/80 Modeling and Analyzing Social Networks of Games Dataset SMMnet SMMnet Dataset Crawled at each 2-hours by a 5-months period over 115 thousand game levels; over 880 thousand players; 7 millions of interactions. create, play, clear, like; first clear and time record. In addition, data from four countries. Computationally unfeasible to capture: (top-1) Japan and (top-2) United States. Thus, we captured (top-3 to 5): Germany (DE), France (FR), Canada (CA). Also, Brazil (BR) (top-18).

32. 16/80 Modeling and Analyzing Social Networks of Games Dataset SMMnet SMMnet Dataset Crawled at each 2-hours by a 5-months period over 115 thousand game levels; over 880 thousand players; 7 millions of interactions. create, play, clear, like; first clear and time record. In addition, data from four countries. Computationally unfeasible to capture: (top-1) Japan and (top-2) United States. Thus, we captured (top-3 to 5): Germany (DE), France (FR), Canada (CA). Also, Brazil (BR) (top-18).

33. 17/80 Modeling and Analyzing Social Networks of Games Dataset Characteristics Data Characteristics Highlights the most recent game style (marioBrosU) marioBros3 8.1% (9.3k) marioBrosU 51.7% (59.5k) marioWorld 19.0% (21.9k) marioBros 21.2% (24.3k) Table: Level styles Style Data marioBrosU 59.5k (51.7%) marioBros 24.3k (21.2%) marioWorld 21.9k (19.0%) marioBros3 9.3k (08.1%)

34. 18/80 Modeling and Analyzing Social Networks of Games Dataset Characteristics Data Characteristics Highlights the normal game difficulty expert 24.1% (27.8k) superExpert 5.4% (6.2k) normal 44.6% (51.4k) easy 25.8% (29.7k) Table: Level difficulties Difficulty Data easy 29.7k (25.8%) normal 51.4k (44.6%) expert 27.8k (24.1%) superExpert 6.2k (05.4%)

35. 19/80 Modeling and Analyzing Social Networks of Games Dataset Characteristics Data Characteristics Pretty balanced data from CA, FR and DE FR 35.2% (40.5k) CA 30.0% (34.5k) DE 32.4% (37.3k) BR 2.4% (2.7k) Table: Levels per country Country Data France (FR) 40.5k (35.2%) Germany (DE) 37.3k (32.4%) Canada (CA) 34.5k (30.0%) Brazil (BR) 2.7k (02.4%)

36. 20/80 Modeling and Analyzing Social Networks of Games Dataset Discussion Discussion First real dataset of a SNG based on the worldwide well-known Super Mario Maker. Collection of over 115 thousand levels; over 880 thousand players; nearly 7 million interactions. Moraes and Cordeiro (2019b) Schema to a Relational Database Management System (RDBMS). Applicability in the research fields of Social Networks and Artificial Intelligence. For example, to detect Digital Influencers. Also, we believe that researchers, game designers and you will find further creative proposals for this dataset.

40. 21/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers 1 Introduction 2 Dataset 3 Detecting Digital Influencers Motivation Related Work Proposal Experiments Discussion 4 Attractive Game Characteristics 5 Conclusion

41. 22/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Motivation Motivation Digital influencers It exists since the popularization of social media; Publish online content (e.g., videos, blogs, forums); Many followers reacting to their contents. Hence Companies invest to endorse their products. Direct relevance in viral marketing; High influence in new trends.

42. 22/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Motivation Motivation Digital influencers It exists since the popularization of social media; Publish online content (e.g., videos, blogs, forums); Many followers reacting to their contents. Hence Companies invest to endorse their products. Direct relevance in viral marketing; High influence in new trends.

43. 23/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Motivation Motivation Detection of influencers How to detect a influencer? What are the influencers’ characteristics? Influencers’ characteristics Publish many contents; Receive many “likes”; but not only that. Evolution of “likes”, a trend of peaks.

44. 23/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Motivation Motivation Detection of influencers How to detect a influencer? What are the influencers’ characteristics? Influencers’ characteristics Publish many contents; Receive many “likes”; but not only that. Evolution of “likes”, a trend of peaks.

45. 24/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Motivation Motivation The evolution of “likes” over time Time Likes Figure: Normal Time Likes p e a k Figure: Influencer

46. 25/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Motivation Problem Problem How to detect digital influencers? in a Social Network of Games; with millions of players; with many types of interaction; which can change over time.

47. 26/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Related Work Related Work How spot the influencers? There is no research on SNG. Thus, we outline the state-of-the-art (influencer detection in social network) Top-rank Morone et al. (2016) Wang et al. (2017) Feature extraction Liu et al. (2014) Qi et al. (2015) Chino et al. (2017)

48. 26/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Related Work Related Work How spot the influencers? There is no research on SNG. Thus, we outline the state-of-the-art (influencer detection in social network) Top-rank Morone et al. (2016) Wang et al. (2017) Feature extraction Liu et al. (2014) Qi et al. (2015) Chino et al. (2017)

49. 27/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Related Work Related Work - Top-rank Algorithms based-on centrality indices; developed to homogeneous network; it’s single type of node our problem is a heterogeneous network. developed to static network. our problem is a set of dynamic networks. Example of degree centrality Gdev, rank by number of developed games. Glike, rank by number of likes given. Unfortunately, more elaborated techniques fail too.

50. 27/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Related Work Related Work - Top-rank Algorithms based-on centrality indices; developed to homogeneous network; it’s single type of node our problem is a heterogeneous network. developed to static network. our problem is a set of dynamic networks. Example of degree centrality Gdev, rank by number of developed games. Glike, rank by number of likes given. Unfortunately, more elaborated techniques fail too.

51. 28/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Related Work Related Work - Feature Extraction Algorithms developed to one dynamic network. our problem has two or more networks. or heavily depend on specific node characteristics. e.g., in Chino et al. (2017), text of the comments. e.g., in Qi et al. (2015), microblog text, number of followers. Summary These works cannot tackle our problem. However, they point that methods for dynamic graph perform feature extraction from the users to classify them.

52. 28/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Related Work Related Work - Feature Extraction Algorithms developed to one dynamic network. our problem has two or more networks. or heavily depend on specific node characteristics. e.g., in Chino et al. (2017), text of the comments. e.g., in Qi et al. (2015), microblog text, number of followers. Summary These works cannot tackle our problem. However, they point that methods for dynamic graph perform feature extraction from the users to classify them.

53. 29/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Our proposal - Main Idea Proposal 1 Model the temporal aspects using data streams; because of the evolution of “likes”.

54. 30/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Stream Modeling Time G like Figure: Glike changing over time

55. 30/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Stream Modeling Likes (a) c1 stream Likes (b) c2 stream Likes (c) c3 stream Likes (d) c4 stream Figure: Stream Modeling on the toy Glike Time G like Figure: Glike changing over time

56. 30/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Stream Modeling Likes (a) c1 stream Likes (b) c2 stream Likes (c) c3 stream Likes (d) c4 stream Figure: Stream Modeling on the toy Glike Time G like Figure: Glike changing over time

57. 30/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Stream Modeling Likes (a) c1 stream Likes (b) c2 stream Likes (c) c3 stream Likes (d) c4 stream Figure: Stream Modeling on the toy Glike Time p3 G like Figure: Glike changing over time

58. 30/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Stream Modeling Likes (a) c1 stream Likes (b) c2 stream Likes (c) c3 stream Likes (d) c4 stream Figure: Stream Modeling on the toy Glike p1 c1 p2 c2 c3 Time ti ti+1 ti+2 c4 p3 c4 p3 c4 G like ti+3 p1 c1 p2 c2 c3 p1 c1 p2 c2 c3 p1 c1 p2 c2 c3 Figure: Glike changing over time

59. 31/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Our proposal - Main Idea Proposal 1 Model the temporal aspects using data streams; because of the evolution of “likes”. 2 Extract features to be used by classification algorithms. a Extract features from stream; b Model the player.

60. 32/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction p1 Figure: Player Modeling

61. 32/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction p1 p2 Figure: Player Modeling

62. 32/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction p1 p2 p3 ... Figure: Player Modeling

63. 32/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction p1 Figure: Player Modeling

64. 32/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction p1 c1 [a1,b2...] Figure: Player Modeling

65. 32/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction p1 c2 [a2,b2...] c1 [a1,b2...] Figure: Player Modeling

66. 32/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction p1 c2 [a2,b2...] p1 = c1 U c2 A = {a1, a2} B = {b1, b2} c1 [a1,b2...] ... Figure: Player Modeling

67. 33/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction Game’s features Linear Regression (LR) Delta Rank (DR) Coefficient of Angle (CA)

68. 33/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction Game’s features Linear Regression (LR) Delta Rank (DR) Coefficient of Angle (CA) Likes Figure: Stream example

69. 34/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Linear Regression (LR) Least Squares: fc(ti) = αc + βc · ti Model: (αc, βc) αc = 0.57 βc = 1.25 Measure: R2 c R2 c = 0.88 Angle: ]c ]c = 51.34o Likes Likes

70. 34/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Linear Regression (LR) Least Squares: fc(ti) = αc + βc · ti Model: (αc, βc) αc = 0.57 βc = 1.25 Measure: R2 c R2 c = 0.88 Angle: ]c ]c = 51.34o Likes Likes

71. 34/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Linear Regression (LR) Least Squares: fc(ti) = αc + βc · ti Model: (αc, βc) αc = 0.57 βc = 1.25 Measure: R2 c R2 c = 0.88 Angle: ]c ]c = 51.34o t1 t2 t3 t4 t5 t6 t7 Time 0 2 4 6 8 10 Like s c: 51.34o R2 c : 0.88 Linear Regression Likes LR curvature

72. 35/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Delta Rank (DR) ∆c,i = likesc,i+1 − likesc,i ∆c = {∆c,1, ∆c,2, . . .} Set: ∆c ∆c = {1, 3, 2, 0, 1, 0} Entropy: S∆c S∆c = 1.33 t1 t2 t3 t4 t5 t6 t7 Time 0 2 4 6 8 10 Likes c,1 c,2 c,3 c,4 c,5 c,6 S c: 1.33 c,1: 1 c,2: 3 c,3: 2 c,4: 0 c,5: 1 c,6: 0 Delta Rank Likes

73. 35/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Delta Rank (DR) ∆c,i = likesc,i+1 − likesc,i ∆c = {∆c,1, ∆c,2, . . .} Set: ∆c ∆c = {1, 3, 2, 0, 1, 0} Entropy: S∆c S∆c = 1.33 t1 t2 t3 t4 t5 t6 t7 Time 0 2 4 6 8 10 Likes c,1 c,2 c,3 c,4 c,5 c,6 S c: 1.33 c,1: 1 c,2: 3 c,3: 2 c,4: 0 c,5: 1 c,6: 0 Delta Rank Likes

74. 35/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Delta Rank (DR) ∆c,i = likesc,i+1 − likesc,i ∆c = {∆c,1, ∆c,2, . . .} Set: ∆c ∆c = {1, 3, 2, 0, 1, 0} Entropy: S∆c S∆c = 1.33 t1 t2 t3 t4 t5 t6 t7 Time 0 2 4 6 8 10 Like s c, 1 c, 2 c, 3 c, 4 c, 5 c, 6 S c : 1.33 c, 1: 1 c, 2: 3 c, 3: 2 c, 4: 0 c, 5: 1 c, 6: 0 Delta Rank Likes greatest value

75. 36/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Coefficient of Angle (CA) θc, i = tan−1 ∆c,i ti+1−ti θc = {θc,1, θc,2, . . .} Set: θc θc = {45.00 , 71.60 , 63.40 , 0, 45.00 , 0} t1 t2 t3 t4 t5 t6 t7 Time 0 2 4 6 8 10 Likes c,1 c,2 c,3 c,4 c,5 c,6 c,1: 45.0o c,2: 71.6o c,3: 63.4o c,4: 0.0o c,5: 45.0o c,6: 0.0o Coefficients of Angle Likes

76. 36/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Coefficient of Angle (CA) θc, i = tan−1 ∆c,i ti+1−ti θc = {θc,1, θc,2, . . .} Set: θc θc = {45.00 , 71.60 , 63.40 , 0, 45.00 , 0} t1 t2 t3 t4 t5 t6 t7 Time 0 2 4 6 8 10 Likes c,1 c,2 c,3 c,4 c,5 c,6 c,1: 45.0o c,2: 71.6o c,3: 63.4o c,4: 0.0o c,5: 45.0o c,6: 0.0o Coefficients of Angle Likes

77. 36/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Coefficient of Angle (CA) θc, i = tan−1 ∆c,i ti+1−ti θc = {θc,1, θc,2, . . .} Set: θc θc = {45.00 , 71.60 , 63.40 , 0, 45.00 , 0} t1 t2 t3 t4 t5 t6 t7 Time 0 2 4 6 8 10 Like s c, 1 c, 2 c, 3 c, 4 c, 5 c, 6 c, 1: 45.0o c, 2: 71.6o c, 3: 63.4o c, 4: 0.0o c, 5: 45.0o c, 6: 0.0o Coefficients of Angle Likes h i g h e s t g r o w t h

78. 37/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction Table: Feature Extractors Based on Game’s features Player modeling LR Linear Regression βc, R2 c, ]c β = {βc1, βc2, . . .}, R2 = {R2 c1, R2 c2, . . .}, ] = {]c1, ]c2, . . .} DR Difference ∆c, S∆c ∆ = ∆c1 ∪ ∆c2 ∪ . . ., S = {S∆c1 , S∆c2 , . . .} CA Angle θc θ = θc1 ∪ θc2 ∪ . . .

79. 37/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Proposal Feature Extraction Table: Feature Extractors Based on Game’s features Player modeling LR Linear Regression βc, R2 c, ]c β = {βc1, βc2, . . .}, R2 = {R2 c1, R2 c2, . . .}, ] = {]c1, ]c2, . . .} DR Difference ∆c, S∆c ∆ = ∆c1 ∪ ∆c2 ∪ . . ., S = {S∆c1 , S∆c2 , . . .} CA Angle θc θ = θc1 ∪ θc2 ∪ . . . FALL Combination . . . . . .

80. 38/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Dataset SMMnet6 Snapshot of 3-months period. +880k players and +75k game maps that received +380k likes. Dataset Canadian (CAN) players - train/test Manually labeled the top-100 players by the number of “likes”, 41 influencers. consensus: influencers who published in popular websites of the game community. French (FRA) players - generality test 6Source: https://www.kaggle.com/leomauro/smmnet/.

81. 38/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Dataset SMMnet6 Snapshot of 3-months period. +880k players and +75k game maps that received +380k likes. Dataset Canadian (CAN) players - train/test Manually labeled the top-100 players by the number of “likes”, 41 influencers. consensus: influencers who published in popular websites of the game community. French (FRA) players - generality test 6Source: https://www.kaggle.com/leomauro/smmnet/.

82. 39/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 1 - Features Extractors Task: Analyze each feature extractor; LR, DR, CA, individually FALL, combination Evaluated 28 classification algorithms; standard input parameter Top-100 CAN players with a 5-fold cross-validation.

83. 40/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 1 - Features Extractors Table: Top-3 best classifiers for each extractor Classifier Accuracy Precision Recall F1-score LR Decision Tree 0.670 (±0.08) 0.621 (±0.11) 0.645 (±0.10) 0.595 (±0.10) Bernoulli NB 0.665 (±0.11) 0.432 (±0.15) 0.600 (±0.10) 0.489 (±0.13) Quadratic Disc. 0.665 (±0.11) 0.432 (±0.15) 0.600 (±0.10) 0.489 (±0.13) DR Decision Tree 0.690 (±0.07) 0.680 (±0.09) 0.658 (±0.07) 0.632 (±0.07) Extra Tree 0.670 (±0.07) 0.616 (±0.10) 0.653 (±0.08) 0.607 (±0.09) Gradient Boost. 0.670 (±0.06) 0.593 (±0.10) 0.613 (±0.06) 0.579 (±0.08) CA Gradient Boost. 0.740 (±0.07) 0.745 (±0.07) 0.766 (±0.08) 0.722 (±0.07) Bagging 0.696 (±0.09) 0.648 (±0.10) 0.675 (±0.11) 0.640 (±0.10) Extra Tree 0.680 (±0.07) 0.650 (±0.09) 0.687 (±0.09) 0.644 (±0.08) F ALL Logistic Reg. 0.808 (±0.11) 0.808 (±0.11) 0.733 (±0.16) 0.745 (±0.14) Ridge CV 0.775 (±0.11) 0.733 (±0.13) 0.675 (±0.16) 0.685 (±0.14) Linear SVC 0.750 (±0.11) 0.750 (±0.11) 0.683 (±0.15) 0.688 (±0.14)

84. 41/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 2 - Parameter Tuning Task: Improve the top-3 classifiers (FALL). Logistic Reg. Ridge CV Linear SVC Grid search through the hyperparameter space. More than 2,500 tests Logistic Reg., 6 parameters (C, dual, fit_intercept, max_iter, penalty, solver) Ridge CV, 4 parameters (alphas, cv, fit_intercept, store_cv_values) Linear SVC, 6 parameters (C, dual, fit_intercept, loss, max_iter, penalty)

85. 41/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 2 - Parameter Tuning Task: Improve the top-3 classifiers (FALL). Logistic Reg. Ridge CV Linear SVC Grid search through the hyperparameter space. More than 2,500 tests Logistic Reg., 6 parameters (C, dual, fit_intercept, max_iter, penalty, solver) Ridge CV, 4 parameters (alphas, cv, fit_intercept, store_cv_values) Linear SVC, 6 parameters (C, dual, fit_intercept, loss, max_iter, penalty)

86. 42/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 2 - Parameter Tuning Table: Tuning the top-3 classifiers Classifier Accuracy Precision Recall F1-score Default F ALL Logistic Reg. 0.808 (±0.11) 0.808 (±0.11) 0.733 (±0.16) 0.745 (±0.14) Ridge CV 0.775 (±0.11) 0.733 (±0.13) 0.675 (±0.16) 0.685 (±0.14) Linear SVC 0.750 (±0.11) 0.750 (±0.11) 0.683 (±0.15) 0.688 (±0.14) Tuning F ALL Logistic Reg. 0.871 (±0.07) 0.903 (±0.05) 0.859 (±0.08) 0.857 (±0.08) Ridge CV 0.806 (±0.22) 0.790 (±0.27) 0.823 (±0.21) 0.785 (±0.26) Linear SVC 0.839 (±0.09) 0.860 (±0.10) 0.831 (±0.10) 0.827 (±0.10)

87. 43/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 3 - Generality Test Task: Evaluated the generality of the feature extractor. Best configuration Extractor: FALL Classifier: Logistic Reg. (tuning) Train: Top-100 CAN players Find influencers in top-100 FRA players.

88. 44/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 3 - Generality Test Influencer detection Labeled 27 players as influencers, automatically in which 21 were manually confirmed as true influencers. Precision: 77.8% Result: A trained algorithm in one country was able to infer influencers in another nationality.

89. 44/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Experiments Exp. 3 - Generality Test Influencer detection Labeled 27 players as influencers, automatically in which 21 were manually confirmed as true influencers. Precision: 77.8% Result: A trained algorithm in one country was able to infer influencers in another nationality.

90. 45/80 Modeling and Analyzing Social Networks of Games Detecting Digital Influencers Discussion Discussion Novel framework to detect digital influencers; Data stream modeling for SNG; Three feature extraction techniques; to model the evolution of “likes” and extract player’s features. Mapped the problem to a classification task; Evaluated the extractors; Parameter tuning; Generality test. which indicate that our proposal is generic to model the behaviour of influencers from different nationalities. Besides, analyze other types of Social Networks. e.g., find influencers in Facebook.

95. 46/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics 1 Introduction 2 Dataset 3 Detecting Digital Influencers 4 Attractive Game Characteristics Context Related Work Proposal Analysis Discussion 5 Conclusion

96. 47/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Context Game Design Draw a level; Technical and aesthetic aspects - challenges; Game designers seek for tools to facilitate the creation process. Figure: Super Mario Maker

99. 48/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Context Game Design Game objects Pick the items and characters. Millions of possible combinations. What combinations are the best ones to attract the players? It is commonly unclear. Figure: Super Mario Maker

100. 48/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Context Game Design Game objects Pick the items and characters. Millions of possible combinations. What combinations are the best ones to attract the players? It is commonly unclear. Figure: Super Mario Maker

101. 49/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Context Problem Problem Draw a game level. for game of platform; with millions object combinations; What are the best ones? Why? Help designers to develop better games. Offer new possibilities as a source of inspiration.

102. 49/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Context Problem Problem Draw a game level. for game of platform; with millions object combinations; What are the best ones? Why? Help designers to develop better games. Offer new possibilities as a source of inspiration.

103. 50/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Related Work Related Work Automated tools Guzdial and Riedl (2018) - games of platform. Karavolos et al. (2018) - shooting games. in the training, a huge dataset and the desired output. does not consider any attractive games. Co-creation tools Guzdial et al. (2019) - games of platform. Replicates the user style - unable to create new patterns. may not offer a new source of inspiration. does not consider any attractive games.

104. 50/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Related Work Related Work Automated tools Guzdial and Riedl (2018) - games of platform. Karavolos et al. (2018) - shooting games. in the training, a huge dataset and the desired output. does not consider any attractive games. Co-creation tools Guzdial et al. (2019) - games of platform. Replicates the user style - unable to create new patterns. may not offer a new source of inspiration. does not consider any attractive games.

105. 51/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Related Work Related Work Automated tools Beaupre et al. (2018) - labyrinth games. generic design pattern of game objects. optimization of a fitness function. impractical to evaluate the quality of new content. Summary Generic patterns are interesting and very inspiring! Also, none work presents the best object combinations; or analyzes on attractive games.

106. 51/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Related Work Related Work Automated tools Beaupre et al. (2018) - labyrinth games. generic design pattern of game objects. optimization of a fitness function. impractical to evaluate the quality of new content. Summary Generic patterns are interesting and very inspiring! Also, none work presents the best object combinations; or analyzes on attractive games.

107. 52/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Proposal Proposal Identify common characteristics of successful games. 1 Identify attractive games; Filter influencers. Select top games. Extract game objects. 2 Identify patterns. Analysis.

108. 52/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Proposal Proposal Identify common characteristics of successful games. 1 Identify attractive games; Filter influencers. Select top games. Extract game objects. 2 Identify patterns. Analysis.

109. 53/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Proposal - Attractive Games 1 Filter out influencers - avoid bias in the analysis. 2 Collect data from the top remaining levels. 3 Extract features that describe their items and characters. 4 Generic patterns - aggregate similar features.

110. 53/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Proposal - Attractive Games 1 Filter out influencers - avoid bias in the analysis. 2 Collect data from the top remaining levels. 3 Extract features that describe their items and characters. 4 Generic patterns - aggregate similar features.

111. 53/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Proposal - Attractive Games ... ... ... ... 1 Filter out influencers - avoid bias in the analysis. 2 Collect data from the top remaining levels. 3 Extract features that describe their items and characters. 4 Generic patterns - aggregate similar features.

112. 53/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Proposal - Attractive Games ... ... ... ... ... ... { + 1 Filter out influencers - avoid bias in the analysis. 2 Collect data from the top remaining levels. 3 Extract features that describe their items and characters. 4 Generic patterns - aggregate similar features.

113. 54/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Player Filtering Note Influencers may receive recognition in their levels due to the popularity, and not necessarily due quality of levels. How? Filter out digital influencers. Solution Detecting digital influencers - approach

116. 55/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Player Filtering How? Top-100 Canadian and top-100 French players. (number of stars received) CAN - 59 non-influencers - 760 game levels. FRA - 73 non-influencers - 1,984 game levels. Selecting successful games. (number of stars received) Unfortunately, some of these levels had been deleted. Top-720 successful games. top-360 from CAN; and top-360 from FRA.

117. 55/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Player Filtering How? Top-100 Canadian and top-100 French players. (number of stars received) CAN - 59 non-influencers - 760 game levels. FRA - 73 non-influencers - 1,984 game levels. Selecting successful games. (number of stars received) Unfortunately, some of these levels had been deleted. Top-720 successful games. top-360 from CAN; and top-360 from FRA.

118. 56/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Data Collection Note SMMnet dataset stores basic information about game usage. without game objects. Solution We manually downloaded the 720 levels from SMM using one Nintendo Wii U console and an original copy.

119. 56/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Data Collection Note SMMnet dataset stores basic information about game usage. without game objects. Solution We manually downloaded the 720 levels from SMM using one Nintendo Wii U console and an original copy. Figure: Nintendo Wii U

120. 57/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Data Collection Diversity - quasi-balanced 720 levels from 2 nationalities. CAN and FRA. Style Many marioBros 244 (33.89%) marioBrosU 225 (31.25%) marioWorld 162 (22.50%) marioBros3 89 (12.36%) Difficult Many easy 147 (20.42%) normal 286 (39.72%) expert 146 (20.28%) superExpert 141 (19.58%)

121. 58/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Feature Extraction Process Extract a histogram of items and characters. represent a game level with features that store the occurrences for each possible game object. open-source code7 extract features from binaries. For example, FNoteBlock counts how many items of type “Note Block”. FP iranhaP lant counts for characters of type “Piranha Plant”. . . . for all 173 unique items or characters. 7Source: https://www.npmjs.com/package/smm-course-viewer.

124. 59/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Feature Aggregation Note These features are specifically for SMM. Solution Aggregate similar features into general, high-level features; for suit our framework to any game of platform; reduce the data sparsity and dimensionality. For example, FI,{vehicle}= FLakituCloud + FKoopaClown + . . .

128. 60/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Feature Aggregation Based in Beaupre et al. (2018) that propose five aggregated features: (1) Collectable, (2) Harmful, (3) Solid, (4) Avatar, and (5) Other. focus on features Collectable and Harmful. excessively abstract for game designers. Note, games commonly contain several subtypes of objects; thus, we propose to expand this process.

131. 61/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Proposal Feature Aggregation Subtype Category Type Aggregated Feature Character Item Power-up Support Jump Move Teleport Puzzle Bonus Move Stable Walk Fly Size Normal Ability Shooter Vitality Fragile Robust ⊻ Legend ∧ ⊻ ∧ exclusive non-exclusive Big Non-Shooter Vehicle Figure: Our proposal for feature aggregation

132. 62/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Characterizing Popular Games Review We collected 720 popular levels. Extracted 173 features and combined them into 32 high-level features. 8 subcategories of items. 24 subcategories of characters. Proposal How spot combinations of objects that usually occur on these levels? Subspace-clustering-based analysis; followed by Association rules analysis on the clusters.

133. 62/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Characterizing Popular Games Review We collected 720 popular levels. Extracted 173 features and combined them into 32 high-level features. 8 subcategories of items. 24 subcategories of characters. Proposal How spot combinations of objects that usually occur on these levels? Subspace-clustering-based analysis; followed by Association rules analysis on the clusters.

134. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle}

135. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach G1 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle} Clusters: G1 = { FI,{vehicle} ,[4,6] }

136. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach G2 G1 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle} Clusters: G1 = { FI,{vehicle} ,[4,6] } G2 = { FC,{robust,stable} ,[10,15] }

137. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach G3 G2 G1 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle} Clusters: G1 = { FI,{vehicle} ,[4,6] } G2 = { FC,{robust,stable} ,[10,15] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] }

138. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach G3 G2 G1 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle} Clusters: G1 = { FI,{vehicle} ,[4,6] } G2 = { FC,{robust,stable} ,[10,15] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } (2) Pruning G1 = { FI,{vehicle} ,[0,1] } G2 = { FC,{robust,stable} ,[0,2] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } G4 = { FI,{vehicle} ,[0,0] , FC,{fragile,walk} ,[0,8] } G5 = { FC,{robust,stable} ,[10,15] , FC,{fragile,walk} ,[0,8] } ...

139. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach G3 G2 G1 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle} Clusters: G1 = { FI,{vehicle} ,[4,6] } G2 = { FC,{robust,stable} ,[10,15] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } (2) Pruning G1 = { FI,{vehicle} ,[0,1] } G2 = { FC,{robust,stable} ,[0,2] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } G4 = { FI,{vehicle} ,[0,0] , FC,{fragile,walk} ,[0,8] } G5 = { FC,{robust,stable} ,[10,15] , FC,{fragile,walk} ,[0,8] } ... Filtering: (a) clusters with high presence of zeros.

140. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach G3 G2 G1 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle} Clusters: G1 = { FI,{vehicle} ,[4,6] } G2 = { FC,{robust,stable} ,[10,15] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } (2) Pruning G1 = { FI,{vehicle} ,[0,1] } G2 = { FC,{robust,stable} ,[0,2] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } G4 = { FI,{vehicle} ,[0,0] , FC,{fragile,walk} ,[0,8] } G5 = { FC,{robust,stable} ,[10,15] , FC,{fragile,walk} ,[0,8] } ... Filtering: (a) clusters with high presence of zeros. (b) default subspace interval.

141. 63/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Approach G3 G2 G1 0 2 4 6 8 0 5 10 15 20 (1) Subspace clustering F C ,{robus t,s table } FI,{vehicle} Clusters: G1 = { FI,{vehicle} ,[4,6] } G2 = { FC,{robust,stable} ,[10,15] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } (2) Pruning G1 = { FI,{vehicle} ,[0,1] } G2 = { FC,{robust,stable} ,[0,2] } G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } G4 = { FI,{vehicle} ,[0,0] , FC,{fragile,walk} ,[0,8] } G5 = { FC,{robust,stable} ,[10,15] , FC,{fragile,walk} ,[0,8] } ... Filtering: (a) clusters with high presence of zeros. (b) default subspace interval. (3) Analysis G3 = { FI,{vehicle} ,[4,6] , FC,{robust,stable} ,[10,15] } G5 = { FC,{robust,stable} ,[10,15] , FC,{fragile,walk} ,[0,8] } ...

142. 64/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Spotting Clusters Techniques CLIQUE (Agrawal et al., 1998); P3C (Moise et al., 2006); and STATPC (Moise and Sander, 2008). Output P3C and STATPC failed to obtain appropriate results. no clusters; or clusters using all dimensions. probability tests needs many instances. our dataset has ∼ 57.42% sparsity. CLIQUE inferred 3, 573 clusters. with dimensionality from 1 up to 9.

143. 64/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Spotting Clusters Techniques CLIQUE (Agrawal et al., 1998); P3C (Moise et al., 2006); and STATPC (Moise and Sander, 2008). Output P3C and STATPC failed to obtain appropriate results. no clusters; or clusters using all dimensions. probability tests needs many instances. our dataset has ∼ 57.42% sparsity. CLIQUE inferred 3, 573 clusters. with dimensionality from 1 up to 9.

144. 65/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Pruning Clusters Note Design of a new game level starts with an empty scenario. irrelevant similarity, absence of many objects. Filtering 1 Filter out clusters that have λ percent sparsity; and λ = 50%, similar to the sparsity of the entire dataset. 2 Boundaries are [0, 0] in at least one feature of their subspaces. From 3, 573 clusters, 318 remained.

147. 66/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Pruning Clusters Note Some clusters may contain an irrelevant size - number of game levels. Filtering 1 Observe the size of the clusters. 2 Cut-off the highest and lowest groups.

148. 66/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Pruning Clusters Note Some clusters may contain an irrelevant size - number of game levels. Filtering 1 Observe the size of the clusters. 2 Cut-off the highest and lowest groups.

149. 67/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Pruning Clusters 0 50 100 150 200 250 300 Cluster rank 0 100 200 300 400 500 600 700 Cluster size Figure: Head cut-off of clusters There are a few large clusters, they include nearly all of levels. Pruned the 25 largest clusters. their sizes vary from 718 to 583; while the 26th has only 70 levels. 293 clusters remained.

153. 68/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Pruning Clusters 26 76 126 176 226 276 Cluster rank 0 10 20 30 40 50 60 70 Cluster size Figure: Tail cut-off of clusters Many tiny clusters also exist. Pareto principle - 80/20 rule; ∼ 80% of the effects come from ∼ 20% of the causes. 59 largest clusters, whose sizes range from 24 to 70 levels.

157. 69/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Analysis Review We selected 59 clusters of combinations from 720 successful levels. describe grouping of game objects that occur frequently. Proposal Each cluster was studied individually by observing features that form its subspace; and cluster’s boundaries. Note CLIQUE tends to infer many redundant clusters. Thus, we manually distributed them into classes.

158. 69/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Analysis Review We selected 59 clusters of combinations from 720 successful levels. describe grouping of game objects that occur frequently. Proposal Each cluster was studied individually by observing features that form its subspace; and cluster’s boundaries. Note CLIQUE tends to infer many redundant clusters. Thus, we manually distributed them into classes.

159. 70/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Subspace Clustering - Results Most interesting patterns were found into 18 clusters. Many combinations for FI,{vehicle} item. e.g., FC,{robust,stable,shooter}, or FC,{robust,stable,big}, or {FC,{fragile,fly,big}, FC,{robust,stable,big}}, etc. Specific characters in levels with bonus items. i.e., FC,{fragile,fly,big}, FC,{fragile,fly,shooter,big}, and FC,{robust,fly,shooter} Characters combinations. i.e., {FC,{robust,fly,shooter}, FC,{robust,stable,big}} {FC,{robust,fly,shooter}, FC,{robust,stable,shooter}} Some obvious combinations.

163. 71/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules Review We have already analyzed 59 clusters of successful game levels. highlighting patterns that are obvious or unspecific. very interesting combinations were detected into 18 clusters. Proposal What is the co-occurrence of these game objects? i.e., X → Y , if X then Y (probability). Association rules analysis.

164. 71/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules Review We have already analyzed 59 clusters of successful game levels. highlighting patterns that are obvious or unspecific. very interesting combinations were detected into 18 clusters. Proposal What is the co-occurrence of these game objects? i.e., X → Y , if X then Y (probability). Association rules analysis.

165. 72/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules - Approach { FI,{vehicle} , FC,{robust,stable} } { FI,{vehicle} , FC,{fragile,fly,big} } { FI,{vehicle} , FC,{fragile,fly,big} , FC,{robust,stable} } ... (1) Frequent itemsets 1 Frequent itemsets filtered from interesting clusters. 2 Detect and evaluate association rules. discovering interesting relations; based on probability of relationships.

166. 72/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules - Approach { FI,{vehicle} , FC,{robust,stable} } { FI,{vehicle} , FC,{fragile,fly,big} } { FI,{vehicle} , FC,{fragile,fly,big} , FC,{robust,stable} } ... (1) Frequent itemsets { FI,{vehicle} } = {FC,{robust,stable} } { FI,{vehicle} , FC,{fragile,fly,big} } = {FC,{robust,stable} } { FI,{vehicle} , FC,{robust,stable} } = {FC,{fragile,fly,big} } ... (2) Rules 1 Frequent itemsets filtered from interesting clusters. 2 Detect and evaluate association rules. discovering interesting relations; based on probability of relationships.

167. 73/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules - Frequent Itemsets Procedure Transform the histogram dataset into a binary matrix. the presence and the absence of each type of object. Algorithm Apriori (Agrawal et al., 1994) to identify itemsets. with support equal to or greater than 0.05; that appears in at least 36 distinct levels. in total, 93, 825 itemsets were identified. Filtering itemsets by interesting clusters. 7 frequent itemsets remained. note, the size of the clusters range from 24 to 70.

170. 74/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules - Rules Procedure Rule X → Y , CONF ≥ 0.5. Y happens in at least half of the occurrences of X. in total, 20 rules were found. Characteristics of the rules: CONF ∈ [0.5, 1.0]; SUPP ∈ [0.06, 0.32] additional metrics LIFT, CONV , LEV evaluating the degree of dependence between X and Y .

171. 74/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules - Rules Procedure Rule X → Y , CONF ≥ 0.5. Y happens in at least half of the occurrences of X. in total, 20 rules were found. Characteristics of the rules: CONF ∈ [0.5, 1.0]; SUPP ∈ [0.06, 0.32] additional metrics LIFT, CONV , LEV evaluating the degree of dependence between X and Y .

172. 75/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Analysis Association Rules - Results Interesting patterns found: Consequent FC,{robust,stable} combines with: FC,{fragile,walk,shooter,big}; also combined with FI,{vehicle}. 0.98 and 1.0 confidence, 0.11 and 0.06 support, respectively. FC,{fragile,fly,big}; also combined with FI,{vehicle}. 0.92 and 0.97 confidence, 0.14 and 0.08 support, respectively. it is popular character, strongly combined with these objects. Antecedent {FC,{fragile,fly,big}} induces the consequent {FC,{robust,stable,big}} with support 0.12 and confidence 0.78. they do not appear alone in the clusters, but there is still a strong correlation. Some FI,{vehicle} rules; and Few obvious rules.

176. 76/80 Modeling and Analyzing Social Networks of Games Attractive Game Characteristics Discussion Discussion How to identify game object combinations of attractive games? Novel framework for detect non-obvious combinations of game characters and items that commonly occur in successful levels. Extracting features from games Player filtering - only non-influencers remained. Data collection - manually downloaded the top-720 game levels. Feature extraction - histogram of items and characters. Feature aggregation - high-level features. Characterizing popular games Subspace clustering analysis; and further Association rules analysis. Besides, this framework can be used for other types of games; and that this research can also be expanded using locale.

180. 77/80 Modeling and Analyzing Social Networks of Games Conclusion 1 Introduction 2 Dataset 3 Detecting Digital Influencers 4 Attractive Game Characteristics 5 Conclusion

181. 78/80 Modeling and Analyzing Social Networks of Games Conclusion Conclusion Main Contributions First Social Network of Games dataset. Moraes and Cordeiro (2019b) Brazilian Symposium on Databases (SBBD 2019) Dataset Showcase Workshop (DSW) Novel framework to automatic detect digital influencers. Moraes and Cordeiro (2019a) International Conference on Enterprise Information Systems (ICEIS 2019) Extensive analysis on common characteristics of successful games. a histogram dataset from 720 games. writing for a journal submission.

182. 79/80 Modeling and Analyzing Social Networks of Games Conclusion Conclusion Additional Contribution Measurements of Normalized Difference Vegetation Index (NDVI) for sugarcane crops classification. (Kunze. et al., 2018) Awards API to retrieve player profiles received the Innovation Award from JavaScript Classes in August 20178 . Moraes and Cordeiro (2019a) was selected as one of the best papers of the ICEIS 2019. 8Source: https://www.jsclasses.org/smm-maker-profile.

183. 80/80 Modeling and Analyzing Social Networks of Games Conclusion Thank you for your attention! Questions?

184. 1/5 Modeling and Analyzing Social Networks of Games Reference I Abisheva, A., Garimella, V. R. K., Garcia, D., and Weber, I. (2014). Who watches (and shares) what on youtube? and when?: Using twitter to understand youtube viewership. In Proceedings of the 7th ACM International Conference on Web Search and Data Mining, WSDM ’14, pages 593–602, New York, NY, USA. ACM. Agrawal, R., Gehrke, J., Gunopulos, D., and Raghavan, P. (1998). Automatic subspace clustering of high dimensional data for data mining applications. SIGMOD Rec., 27(2):94–105. Agrawal, R., Srikant, R., et al. (1994). Fast algorithms for mining association rules. In Proc. 20th int. conf. very large data bases, VLDB, volume 1215, pages 487–499. Aung, M., Bonometti, V., Drachen, A., Cowling, P., Kokkinakis, A. V., Yoder, C., and Wade, A. (2018). Predicting skill learning in a large, longitudinal moba dataset. In 2018 IEEE Conference on Computational Intelligence and Games (CIG), pages 1–7. Barabási, A.-L. and Pósfai, M. (2016). Network science. Cambridge university press, Cambridge, USA. Beaupre, S., Wiles, T., Briggs, S., and Smith, G. (2018). A design pattern approach for multi-game level generation. In Fourteenth Artificial Intelligence and Interactive Digital Entertainment Conference, pages 1–5. Browne, C., Liapis, A., and Winands, M. (2019). Special issue on computer aided game and puzzle design. ICGA Journal, (Preprint):1–2. Chino, D. Y. T., Costa, A. F., Traina, A. J. M., and Faloutsos, C. (2017). VolTime: Unsupervised Anomaly Detection on Users’ Online Activity Volume, pages 108–116. SIAM International Conference on Data Mining.

185. 2/5 Modeling and Analyzing Social Networks of Games Reference II Gros, D., Hackenholt, A., Zawadzki, P., and Wanner, B. (2018). Interactions of twitch users and their usage behavior. In Meiselwitz, G., editor, Social Computing and Social Media. Technologies and Analytics, pages 201–213, Cham. Springer International Publishing. Guzdial, M., Liao, N., Chen, J., Chen, S.-Y., Shah, S., Shah, V., Reno, J., Smith, G., and Riedl, M. O. (2019). Friend, collaborator, student, manager: How design of an ai-driven game level editor affects creators. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems, CHI ’19, pages 624:1–624:13, New York, NY, USA. ACM. Guzdial, M. and Riedl, M. (2018). Automated game design via conceptual expansion. In AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, pages 31–37. Hilvert-Bruce, Z., Neill, J. T., Sjöblom, M., and Hamari, J. (2018). Social motivations of live-streaming viewer engagement on twitch. Computers in Human Behavior, 84:58–67. Huizinga, J. (1955). Homo ludens: A study of the Play-Element in Culture, volume 4. Boston, MA: The Beacon Press. Karavolos, D., Liapis, A., and Yannakakis, G. N. (2018). Using a surrogate model of gameplay for automated level design. In 2018 IEEE Conference on Computational Intelligence and Games (CIG), pages 1–8. Karpouzis, K., Yannakakis, G. N., Shaker, N., and Asteriadis, S. (2015). The platformer experience dataset. In 2015 International Conference on Affective Computing and Intelligent Interaction, pages 712–718, USA. ACII.

186. 3/5 Modeling and Analyzing Social Networks of Games Reference III Kunze., L. F., Amaral., T., Moraes., L. M. P., Oliveira., J. J. M., Junior., A. G. B., de Sousa., E. P. M., and Cordeiro., R. L. F. (2018). Classification analysis of ndvi time series in metric spaces for sugarcane identification. In Proceedings of the 20th International Conference on Enterprise Information Systems - Volume 1: ICEIS,, pages 162–169. INSTICC, SciTePress. Lee, Y.-T., Chen, K.-T., Cheng, Y.-M., and Lei, C.-L. (2011). World of warcraft avatar history dataset. In Proceedings of the Second Annual ACM Conference on Multimedia Systems, MMSys ’11, pages 123–128, New York, NY, USA. ACM. Lim, C.-U. and Harrell, D. F. (2015). Comparing clustering approaches for modeling players’ values through avatar construction. AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. Lin, Z., Gehring, J., Khalidov, V., and Synnaeve, G. (2017). Stardata: A starcraft ai research dataset. AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. Liu, N., Li, L., Xu, G., and Yang, Z. (2014). Identifying domain-dependent influential microblog users: A post-feature based approach. In AAAI, pages 3122–3123. Lucas, S. M. (2009). Computational intelligence and ai in games: A new ieee transactions. IEEE Transactions on Computational Intelligence and AI in Games, 1(1):1–3. Mandelbrot, B. (1960). The pareto-levy law and the distribution of income. International Economic Review, 1(2):79–106.

187. 4/5 Modeling and Analyzing Social Networks of Games Reference IV Moise, G. and Sander, J. (2008). Finding non-redundant, statistically significant regions in high dimensional data: A novel approach to projected and subspace clustering. In Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’08, pages 533–541, New York, NY, USA. ACM. Moise, G., Sander, J., and Ester, M. (2006). P3c: A robust projected clustering algorithm. In Sixth International Conference on Data Mining (ICDM’06), pages 414–425. Moraes, L. M. P. and Cordeiro, R. L. F. (2019a). Detecting influencers in very large social networks of games. In Proceedings of the 21st International Conference on Enterprise Information Systems - Volume 1: ICEIS,, pages 93–103, Crete, Greece. INSTICC, SciTePress. Moraes, L. M. P. and Cordeiro, R. L. F. (2019b). Smmnet: A social network of games dataset. In Brazilian Symposium on Databases (SBBD) 2019 - Dataset Showcase Workshop (DSW), pages 429–438, Ceará, Brazil. Brazilian Computer Society (SBC). Morone, F., Min, B., Bo, L., Mari, R., and Makse, H. A. (2016). Collective influence algorithm to find influencers via optimal percolation in massively large social media. Scientific reports, 6:30062. Newman, M. (2005). Power laws, pareto distributions and zipf’s law. Contemporary Physics, 46(5):323–351. Pei, S., Morone, F., and Makse, H. A. (2018). Theories for Influencer Identification in Complex Networks, pages 125–148. Springer International Publishing, Cham. Pivec, M. and Kearney, P. (2007). Games for learning and learning from games. Organizacija, 40(6).

188. 5/5 Modeling and Analyzing Social Networks of Games Reference V Qi, L., Huang, Y., Li, L., and Xu, G. (2015). Learning to rank domain experts in microblogging by combining text and non-text features. In 2015 International Conference on Behavioral, Economic and Socio-cultural Computing (BESC), pages 28–31. Salen, K., Tekinbaş, K. S., and Zimmerman, E. (2004). Rules of play: Game design fundamentals. MIT press. Sjöblom, M. and Hamari, J. (2017). Why do people watch others play video games? an empirical study on the motivations of twitch users. Computers in Human Behavior, 75:985–996. Wang, X., Zhang, X., Yi, D., and Zhao, C. (2017). Identifying influential spreaders in complex networks through local effective spreading paths. Journal of Statistical Mechanics: Theory and Experiment, 2017(5):053402. Yannakakis, G. N. and Togelius, J. (2015). A panorama of artificial and computational intelligence in games. IEEE Transactions on Computational Intelligence and AI in Games, 7(4):317–335. Yannakakis, G. N. and Togelius, J. (2018). Artificial Intelligence and Games. Springer International Publishing, Cham.

Modeling and Analyzing Social Networks of Games

Recommended

Recommended

More Related Content

Similar to Modeling and Analyzing Social Networks of Games

Similar to Modeling and Analyzing Social Networks of Games (20)

Recently uploaded

Recently uploaded (20)

Modeling and Analyzing Social Networks of Games