未來，影音串流的爆炸性流量勢必會使得核心網路無法負荷，多接入邊端計
算(MEC)已經成為減輕影音串流流量的重要技術了，MEC伺服器部屬在基站附
近，提供了儲存和運算能力，以MEC應用程序在邊端提供預測熱門影片並快取的服
務。我們提出了一個MEC架構，並在這架構實現流量卸載。我們設計了一個遷移式學
習框架基於全球熱門領域和本地熱門領域，其中包括資料前處理、遷移式模型、快取
策略的制定，最後效能表現上非常突出。本文比較眾多機器學習方法中的優缺點，實
驗結果說明了NN和TF是最佳的解決方案，可以提升使用者的體驗品質。

In the future, mobile video traffic have been grown brustly, causing network traffic exceed
tolerable capacity. Multi-access edge computing (MEC) has become the important
technology to reduce video streaming traffic. MEC provide application and service with
large storage capacity and high computation ability. In this paper, we deploy the MEC
App in the MEC architecture to offload backhaul network traffic. We propose the transfer
learning framework to transfer knowledge from auxiliary task that is the global YouTube
dataset to help on target task that is the local YouTube dataset, as far as we know, this
idea is first proposed. Finally, we compare with various machine learning models, and
investigate their performance.

摘要
Abstract
Contents
List of Figures
List of Tables
1 INTRODUCTION----------------------------------1
2 RELATED WORK----------------------------------4
2.1 Content Popularity Prediction ----------------4
2.2 Transfer learning-----------------------------5
2.3 Relation between global and local videos------6
3 SYSTEM MODEL AND CACHING PREDICTION PROBLEM---7
3.1 System Model ---------------------------------7
3.2 Problem Formulation---------------------------9
4 PROPOSED MACHINE LEARNING MODEL FOR VIDEO POPULARITY PREDICTION -----------------------------------------------------11
4.1 Overview-------------------------------------11
4.2 Unsupervised learning------------------------12
4.2.1 Canopy+K-Means Clustering--------------------12
4.2.2 Birch clustering-----------------------------13
4.2.3 Mean-Shift Clustering------------------------15
4.2.4 Expectation–maximization clustering----------16
4.3 Supervised learning--------------------------17
4.3.1 Hoeffding tree-------------------------------17
4.3.2 Neural network-------------------------------18
4.3.3 Transfer learning----------------------------20
5 EXPERIMENT METHODS FOR VIDEO POPULARITY PREDIC-
TION -------------------------------------22
5.1 Dataset--------------------------------------23
5.2 Implementation of unsupervised learning------25
5.3 Implementation of supervised learning--------25
5.4 Transfer learning framework------------------26
5.4.1 Data preprocessing---------------------------26
5.4.2 Transfer learning model ---------------------28
5.4.3 Cache decision policy------------------------28
6 NUMERICAL RESULTS AND DISCUSSIONS------------31
7 Conclusion-----------------------------------38
Acknowledgement--------------------------------------39
Bibliography-----------------------------------------40

1. C. Index. Cisco visual networking index: Global mobile data traffic forecast, 20162021. White Paper, Jun 2017.
2. M. Patel, B. Naughton, and C. Chan. Mobile-edge computing introductory technical white paper. White Paper, 2014.
3. T. Johnson and D. Shasha. 2q: A low overhead high performance buffer management replacement algorithm. In Proceedings of the 20th International Conference on Very Large Data Bases, VLDB ’94, pages 439–450, San Francisco, CA, USA, 1994. Morgan Kaufmann Publishers Inc.
4. D. Lee, J. Choi, J. H. Kim, S. H. Noh, S. L. Min, Y. Cho, and C. S. Kim. Lrfu: a spectrum of policies that subsumes the least recently used and least frequently used policies. IEEE Transactions on Computers, 50(12):1352–1361, Dec 2001.
5. F. Figueiredo, J. M. Almeida, M. A. Gonçalves, and F. Benevenuto. Trendlearner: Early prediction of popularity trends of user generated content. Information Sciences, 349-350:172 – 187, 2016.
6. M. S. ElBamby, M. Bennis, W. Saad, and M. Latva-aho. Content-aware user clustering and caching in wireless small cell networks. In 2014 11th International Symposium on Wireless Communications Systems (ISWCS), pages 945–949, Barcelona, Spain, Aug 2014.
7. J. Nogueira, D. Gonzalez, L. Guardalben, and S. Sargento. Over-the-top catch-up tv content-aware caching. Proc. IEEE Symp. Comput. Commun. (ISCC), pages 1012–1017, Jun 2016.
8. J. Li, S. Hong, S. Xia, and S. Luo. Neural network based popularity prediction for iptv system. J. Netw., 7:2051–2056, 2012.
9. S. D. Roy, T. Mei, W. Zeng, and S. Li. Towards cross-domain learning for social video popularity prediction. IEEE Transactions on Multimedia, 15(6):1255–1267, Oct 2013.
10. T. Hou., G. Feng, S. Qin, and W. Jiang. Proactive content caching by exploiting transfer learning for mobile edge computing. International Journal of Communication Systems, 31(11):3706, 2018.
11. M. Zink, K. Suh, Y. Gu, and J. Kurose. Characteristics of youtube network traffic at a campus network – measurements, models, and implications. Computer Networks, 53(4):501 – 514, 2009. Content Distribution Infrastructures for Community Networks.
12. Youtube trending video statistics with subscriber, 2018.
https://www.kaggle.com/sgonkaggle/youtube-trend-with-subscriber.
13. N. Ben Hassine, D. Marinca, P. Minet, and D. Barth. Expert-based on-line learning and prediction in content delivery networks. pages 182–187, Sep. 2016.
14. E. B. Abdelkrim, M. A. Salahuddin, H. Elbiaze, and R. Glitho. A hybrid regression model for video popularity-based cache replacement in content delivery networks. pages 1–7, Dec 2016.
15. V. D. Silva and A. T. Winck. Video popularity prediction in data streams based on context-independent features. In Proceedings of the Symposium on Applied Computing, SAC ’17, pages 95–100, New York, NY, USA, 2017. ACM.
16. E. Baştuğ, M. Bennis, and M. Debbah. A transfer learning approach for cacheenabled wireless networks. pages 161–166, May 2015.
17. T. Hou, G. Feng, S. Qin, and W. Jiang. Proactive content caching by exploiting transfer learning for mobile edge computing. pages 1–6, Dec 2017.
18. Y. Zhu, Y. Chen, Z. Lu, S. J. Pan, G. R. Xue, Y. Yu, and Q. Yang Qiang. Heterogeneous transfer learning for image classification. pages 1304–1309, 2011.
19. J. Deng, Z. Zhang, E. Marchi, and B. Schuller. Sparse autoencoder-based feature transfer learning for speech emotion recognition. pages 511–516, Sep. 2013. 20. S. J. Pan and Q. Yang. A survey on transfer learning. IEEE Transactions on Knowledge and Data Engineering, 22(10):1345–1359, Oct 2010.
21. M. Zink, K. Suh, Y. Gu, and J. F. Kurose. Watch global, cache local: Youtube network traffic at a campus network - measurements and implications. Proceedings of SPIE - The International Society for Optical Engineering, 6818, Jan 2008.
22. P. Gill, M. Arlitt, Z. Li, and A. Mahanti. Youtube traffic characterization: A view from the edge. In Proceedings of the 7th ACM SIGCOMM Conference on Internet Measurement, IMC ’07, pages 15–28, New York, NY, USA, 2007. ACM.
23. A. McCallum, K. Nigam, and L. H. Ungar. Efficient clustering of high-dimensional data sets with application to reference matching. In Proceedings of the Sixth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD’00, pages 169–178, New York, NY, USA, 2000. ACM.