YouTube 2016 2 16
27 YouTube YouTube UGC User Generated Content CDN Content Delivery Networks LRU Least Recently Used UGC YouTube CGM Consumer Generated Media CGM CGM UGC UGC YouTube k-means YouTube YouTube 1
3 10 YouTube k-means 2
1 6 2 YouTube 8 2.1.............................. 8 2.2....................... 10 2.3 k-means............. 17 3 YouTube 22 3.1............................. 22 3.2........................ 22 3.3...................... 24 3.4...................................... 24 4 30 31 32 3
1.................................. 8 2 1 2 3 6 1 CCDF....... 11 3 1 2 3 7 14 1 CCDF....... 12 4 1 2 3 7 14 CCDF........ 12 5 1 CCDF.... 13 6 1 CCDF..... 13 7 7 CCDF.... 14 8 7 CCDF... 14 9 1 CCDF....... 15 10 1 CCDF........ 15 11 7 CCDF....... 16 12 7 CCDF..... 16 13 24 1 5 k-means................. 19 14 48 1 5 k-means................. 20 15 72 1 5 k-means................. 21 16.......... 23 17 Y = 3 d............. 27 18 d = 7 Y................. 28 19 d = 14 Y................ 29 4
1.................................. 24 2 3 7.. 26 3 3 14. 26 5
1 YouTube[1] UGC User Generated Content OTT Over The Top (YouTube ) ISP Internet Service Provider CDN Contents Delivery Networks LRU Least Recently Used [2] UGC CGM Consumer Generated Media) CGM UGC VoD Video-on-Demand UGC [3] Digg[4] [5] YouTube 6
YouTube 3 [6] YouTube 1 [7] YouTube [8] Digg YouTube 30 [9] [8] x x [8] [10] YouTube [11] [12] 1 k-means YouTube [12] 1 1 k-means 7
1: 2 YouTube YouTube YouTube YouTube 2.1 YouTube YouTube 1 YouTube API YouTube Data API version3.0 [13] 1 1 1-2015 10 14 2015 12 16 1 87, 830-2015 12 25 2016 1 21 1 40, 900 8
1 1 1 1 ID YouTube API ID ID ( ID ) 1 1 YouTube API 1 1 YouTube API 1 87, 830 1 1 1 1 30 8 10 1 5 7 1 YouTube [14] 1 ( 9
1, 000km) 30 ID ID 1 JP WestUS EastUS GB FR DE BR IN 2.2 1 2 3 6 1 2 1 2 3 7 14 1 3 2 1 SNS Social Networking Service 3 1 2 3 7 14 4 0 3 4 7 8 11 12 15 16 19 20 23 1 5 10
1 0.1 1hour 2hour 3hour 6hour CCDF 0.01 0.001 0.0001 1e-005 1 10 100 1000 10000 100000 1x10 6 1x10 7 View count 2: 1 2 3 6 1 CCDF 1 6 7 7 1 20 23 16 19 7 8 7 1 9 1 10 7 11 7 12 11
1 0.1 1day 2day 3day 7day 14day CCDF 0.01 0.001 0.0001 1e-005 1 10 100 1000 10000 100000 1x10 6 1x10 7 View count 3: 1 2 3 7 14 1 CCDF 1 0.1 1day 2day 3day 7day 14day CCDF 0.01 0.001 0.0001 1e-005 1 10 100 1000 10000 100000 1x10 6 1x10 7 1x10 8 Total view count 4: 1 2 3 7 14 CCDF 12
1 0.1 0-3hour 4-7hour 8-11hour 12-15hour 16-19hour 20-23hour CCDF 0.01 0.001 0.0001 1e-005 1 10 100 1000 10000 100000 1x10 6 1x10 7 View count 5: 1 CCDF 1 0.1 0-3hour 4-7hour 8-11hour 12-15hour 16-19hour 20-23hour CCDF 0.01 0.001 0.0001 1e-005 1 10 100 1000 10000 100000 1x10 6 1x10 7 View count 6: 1 CCDF 13
1 0.1 0-3hour 4-7hour 8-11hour 12-15hour 16-19hour 20-23hour CCDF 0.01 0.001 0.0001 1e-005 1 10 100 1000 10000 100000 1x10 6 1x10 7 View count 7: 7 CCDF 1 0.1 0-3hour 4-7hour 8-11hour 12-15hour 16-19hour 20-23hour CCDF 0.01 0.001 0.0001 1e-005 10 100 1000 10000 100000 1x10 6 1x10 7 1x10 8 Total view count 8: 7 CCDF 14
CCDF 1 0.1 0.01 WestUS(5541) EastUS(9829) GB(4145) JP(3505) FR(5153) DE(4217) BR(2783) IN(5727) 0.001 0.0001 1 10 100 1000 10000 100000 1x10 6 View count 9: 1 CCDF CCDF 1 0.1 0.01 WestUS(5541) EastUS(9829) GB(4145) JP(3505) FR(5153) DE(4217) BR(2783) IN(5727) 0.001 0.0001 1 10 100 1000 10000 100000 1x10 6 View count 10: 1 CCDF 15
CCDF 1 0.1 0.01 WestUS(5541) EastUS(9829) GB(4145) JP(3505) FR(5153) DE(4217) BR(2783) IN(5727) 0.001 0.0001 1 10 100 1000 10000 100000 1x10 6 View count 11: 7 CCDF CCDF 1 0.1 0.01 WestUS(5541) EastUS(9829) GB(4145) JP(3505) FR(5153) DE(4217) BR(2783) IN(5727) 0.001 0.0001 10 100 1000 10000 100000 1x10 6 1x10 7 Total view count 12: 7 CCDF 16
2.3 k-means n k-means YouTube [12] 1 k-means 1 k-means v n 0 1 n v k-means 5 24 13 13(a) 1 5 16 1 13(b) 5 24 24 SNS 24 30 13(c) 13(a) 5 7 13(d) 5 24 5 48 14 14(a) 3 17
1 14(b) 3 4 12 3 3 30 14(c) 3 7 14(d) 3 5 72 15 15(a) 3 1 15(b) 3 30 15(c) 3 7 15(d) 3 72 24 48 3 7 10, 000 3 18
Normalized view count 1 0.8 0.6 0.4 0.2 cluster1 (29225) cluster2 (15564) cluster3 (17590) cluster4 (15036) cluster5 (10415) View count 3500 3000 2500 2000 1500 1000 500 cluster1 (29225) cluster2 (15564) cluster3 (17590) cluster4 (15036) cluster5 (10415) 0 4 8 12 16 20 24 Hour 0 0 24 48 72 96 120 144 168 Hour after upload (a) (b) 1 25000 20000 cluster1 (29225) cluster2 (15564) cluster3 (17590) cluster4 (15036) cluster5 (10415) 1 0.1 cluster1 (29225) cluster2 (15564) cluster3 (17590) cluster4 (15036) cluster5 (10415) View count 15000 10000 CCDF 0.01 0.001 5000 0.0001 0 0 5 10 15 20 25 30 Day after upload 1e-005 1 10 100 1000 10000 100000 1e+006 1e+007 View count (c) 1 (d) 7 13: 24 1 5 k-means 19
Normalized view count 1 0.8 0.6 0.4 0.2 cluster1 (25656) cluster2 (16873) cluster3 (12446) cluster4 (20304) cluster5 (12551) View count 4000 3500 3000 2500 2000 1500 1000 500 cluster1 (25656) cluster2 (16873) cluster3 (12446) cluster4 (20304) cluster5 (12551) 0 6 12 18 24 30 36 42 48 Hour 0 0 24 48 72 96 120 144 168 Hour after upload (a) (b) 1 30000 25000 cluster1(25656) cluster2(16873) cluster3(12446) cluster4(20304) cluster5(12551) 1 0.1 cluster1 (25656) cluster2 (16873) cluster3 (12446) cluster4 (20304) cluster5 (12551) View count 20000 15000 10000 CCDF 0.01 0.001 5000 0.0001 0 0 5 10 15 20 25 30 Day after upload 1e-005 1 10 100 1000 10000 100000 1e+006 1e+007 View count (c) 1 (d) 7 14: 48 1 5 k-means 20
Normalized view count 1 0.8 0.6 0.4 0.2 cluster1 (27234) cluster2 (12079) cluster3 (10727) cluster4 (18786) cluster5 (19004) View count 3000 2500 2000 1500 1000 500 cluster1 (27234) cluster2 (12079) cluster3 (10727) cluster4 (18786) cluster5 (19004) 0 6 12 18 24 30 36 42 48 54 60 66 72 Hour 0 0 24 48 72 96 120 144 168 Hour after upload (a) (b) 1 30000 25000 cluster1 (27234) cluster2 (12079) cluster3 (10727) cluster4 (18786) cluster5 (19004) 1 0.1 cluster1 (27234) cluster2 (12079) cluster3 (10727) cluster4 (18786) cluster5 (19004) View count 20000 15000 10000 CCDF 0.01 0.001 5000 0.0001 0 0 5 10 15 20 25 30 Day after upload 1e-005 1 10 100 1000 10000 100000 1e+006 1e+007 View count (c) 1 (d) 7 15: 72 1 5 k-means 21
3 YouTube Y 3.1 A B P (B A) P (B A) = P (B)P (A B) P (A) n F 1,, F n C p(c F 1,, F n ) p(c F 1,, F n ) = p(c)p(f 1,, F n C) p(f 1,, F n ) C p(c, F 1,, F n ) = p(c)p(f 1 C)p(F 2 C)p(F 3 C) C classify classify(f 1,, f n ) = arg max c n p(c = c) p(f i = f i C = c) (1) i=1 3.2 16 H Y d H Y + d H Y Y Y 22
16: 1. 2. YouTube Data API 1 3. Y YouTube Data API 1 1 4. Y d YouTube Data API 1 1 5. 1 Y + d Y 1 6. 1 Y d d 23
1: ID Y Y =H 1 2 Y =L abcdefghijk 1.0 0.5 0.4 5 H lmnopqrstuv 0.5 0.2 0.0 3 L wxyz1234567 0.2 0.8 0.6 4 H 3.3 1 d 1 1% 2 d d 1% 3.1 n F 1,, F n 2.3 Y v Y 0 1 Y Y Y + 1 3.4 87, 830 p(c = c) p(f i = f i C = c) p(c = c) p(f i = f i C = c) 3.1 1 classify NBC: Naive Bayes Classifier VCS: View Count based Selection 24
Y d = 7 Y = 3 2 3 8 1% 2 8 7 1% 3 7 d = 14 Y = 3 3 3 15 1% 2 15 15 1% 3 14 Y = 3 d 1 17(a) 2 17(b) 1 2 d = 7 Y 1 18(a) 2 18(b) Y/3 25
2: 3 7 8 1% 2 8 1% 0.785 0.956 0.697 0.860 3: 3 14 15 1% 2 15 1% 0.707 0.933 0.674 0.837 3 Y d = 14 Y 1 19(a) 2 19(b) d = 7 Y 3 26
1 0.8 Accuracy 0.6 0.4 0.2 NBC 0 7 14 VCS 21 d 28 (a) d 1 0.8 Accuracy 0.6 0.4 0.2 NBC 0 7 14 VCS 21 d 28 (b) d 17: Y = 3 d 27
1 NBC VCS 0.8 Accuracy 0.6 0.4 0.2 0 3 6 9 12 15 18 21 24 Y (hour) (a) d 1 NBC VCS 0.8 Accuracy 0.6 0.4 0.2 0 3 6 9 12 15 18 21 24 Y (hour) (b) d 18: d = 7 Y 28
1 NBC VCS 0.8 Accuracy 0.6 0.4 0.2 0 3 6 9 12 15 18 21 24 Y (hour) (a) d 1 NBC VCS 0.8 Accuracy 0.6 0.4 0.2 0 3 6 9 12 15 18 21 24 Y (hour) (b) d 19: d = 14 Y 29
4 YouTube 1 k-means 3 Good 30
NTT 31
[1] YouTube. https://www.youtube.com/. [2] N. Kamiyama, R. Kawahara, T. Mori, and H. Hasegawa, Multicast Pre-distribution VoD System, IEICE transactions on communications, vol. E96-B, pp. 1459 1471, June 2013. [3] K. Lerman and T. Hogg, Using a Model of Social Dynamics to Predict Popularity of News, in Proceedings of the nineteenth international conference on World Wide Web, pp. 621 630, Apr. 2010. [4] Digg. http://digg.com/. [5] F. Figueiredo, F. Benevenuto, and J. M. Almeida, The tube over time: characterizing popularity growth of YouTube videos, in Proceedings of the fourth ACM international conference on Web search and data mining, pp. 745 754, Feb. 2011. [6] Y. Borghol, S. Mitra, S. Ardon, N. Carlsson, D. Eager, and A. Mahanti, Characterizing and modelling popularity of user-generated videos, Performance Evaluation, vol. 68, pp. 1037 1055, Nov. 2011. [7] G. Gürsun, M. Crovella, and I. Matta, Describing and forecasting video access patterns, in Proceedings of IEEE INFOCOM, pp. 16 20, Apr. 2011. [8] G. Szabo and B. A. Huberman, Predicting the popularity of online content, Communications of the ACM, vol. 53, pp. 80 88, Aug. 2010. [9] H. Pinto, J. M. Almeida, and M. A. Gonçalves, Using early view patterns to predict the popularity of youtube videos, in Proceedings of the sixth ACM international conference on Web search and data mining, pp. 365 374, Feb. 2013. [10] G. Chatzopoulou, C. Sheng, and M. Faloutsos, A first step towards understanding popularity in YouTube, in Proceedings of INFOCOM IEEE Conference on Computer Communications Workshops, pp. 1 6, Mar. 2010. [11] J. M. Tirado, D. Higuero, F. Isaila, and J. Carretero, Multi-model prediction for enhancing content locality in elastic server infrastructures, in Proceedings of the 32
eighteenth International Conference on High Performance Computing, pp. 1 9, Dec. 2011. [12] Y. Kitade, Analyzing popularity dynamics of YouTube content and its application to content cache design, Master s thesis, Graduate School of Information Science and Technology, Osaka University, Feb. 2015. [13] YouTube Data API. https://developers.google.com/youtube/v3/. [14] YouTube Help. https://support.google.com/youtube/answer/2991785/. 33