為了避免由骨幹網路到小基地台傳輸限制而造成的瓶頸，Mobile Edge Computing (MEC)，成為了一個有效的辦法。MEC可以進行運算並快取影片在小基地台，如此一來可以降低影片的傳輸量以及重疏的等候時間。此外，若我們能夠運用好的預測模型預測未來的熱門影片並預先快取在小基地台中，我們可以有效地提升MEC的效能。因此我們的目標是透過推薦系統演算法來建出一個可以預測未來熱門影片的模型。
隨著機器學習的蓬勃發展，Auto-encoder – 一種根據非監督式學習方法的人工神經網路模型，也被廣泛應用於推薦系統演算法中。根據協同過濾的概念，協同過濾式的Auto-encoder可以根據使用者過去的觀影紀錄以及對影片評分的情形來推薦適合他們的影片。過去許多篇論文都顯示了他們提出的協同過濾式 Auto-encoder 演算法都比早期協同過濾演算法的表現更突出。
本篇論文提出以分群演算法來優化 Auto-encoder 模型的推薦效果。我們根據使用者的相似度利用k-means分群後，再根據每個群體的觀影特性，調整並訓練預先訓練好的 Auto-encoder 模型。在我們的實驗中，我們提出的模型可以比原先的 Auto-encoder 在Netflix小資料的部分進步17% (Average Precision)，而在Netflix大資料的部分可以進步5% (Average Precision)。更進一步的，我們使用我們提出的模型來預測未來的熱門影片。由於原本的模型是被設計來提供各個使用者屬於他們適合的影片，因此我們也提出了將模型預測出來的結果轉換成預測未來熱門影片的方法。在我們的實驗中，我們所提出的模型在預測未來影片中，在小資料的部分能夠有69%的準確率(Recall)，而在大資料的部分能夠有78%的準確率(Recall)。

To avoid bottleneck in the limited capacity of backhaul link, Mobile Edge Computing (MEC) is a promising solution, which computing and caching in the mobile edge in such a way that the buffered video clips could be delivered with less network latency and traffic load. Furthermore, the performance gain can be further upgraded if we can predict and cache the Top-K popular videos clips in advance with the help of a better forecasting model. Our goal in this thesis is to build a Top-K forecasting solution from renowned recommender systems.
An auto-encoder is a sort of artificial neural networks (ANNs) recognized to learn the concept of data (Latent space representation) in an unsupervised manner, and it has been widely used in recommender systems. For instance, several collaborative auto-encoder models have shown that their performance gains can outperform that of the collaborative filtering (CF)-based models.
In this thesis, we first proposed an advanced auto-encoder based recommendation algorithm with the help of k-means clustering that can upgrade the performance of the original auto-encoder model. The experimental results demonstrate that 17% improvement (average precision) in Netflix smaller datasets and around 5% improvement (average precision) for Netflix’s larger datasets. Next, we extent to solve the predication of the Top-K popular video clips in advance (next period, e.g. per week). As mentioned, the Auto-encoder model was originally built for endorsing higher favorability rating videos to users, we proposed an ensemble method based on these recommending contents, to aggregate them thus forecast the potential Top-K popular video clips for the next period. Our experimental results show that the proposed method can achieve the hit ratio around 69% for small datasets and 78% for gigantic datasets (Netflix 1 Year).

[1] Forecast, Cisco VNI, "Cisco Visual Networking Index: Global Mobile Data Traffic Forcast Update, 2-15-2020 White Paper". Cisco Public Information, February 2016.
[2] I. Hwang, B. Song, and S. Soliman, "A Holistic View on Hyper-Dense Heterogeneous and Small Cell Networks," IEEEE Communications Magazine, Vol. 51, No. 6, Jun. 2013, pp. 20-27.
[3] Yun Chao Hu, Milan Patel, Dario Sabella, Nurit Sprecher and Valerie Young. "Mobile Edge Computing: A key technology towards 5G", ETSI White Paper No. 11, September 2015.
[4] F. Ricci, L. Rokach, and B. Shapira. Introduction to Recommender Systems Handbook, Springer, 2011.
[5] C. Wang and D. M. Blei. Collaborative Topic Modeling for Recommending Scientific Articles, In Proceedings of the 17st ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, San Diego, CA, Augest 21-24, 2011, pp. 448-456.
[6] R. Salakhutdinov and A. Mnih, "Probabilistic Matrix Factorization," In Proceedings of the Twenty-First Annual Conference on Neural Information Processing Systems, Vancouver, Canada, December 3-6, 2007. pp. 1257-1264.
[7] Hao Wang, Xingjian Shi, Dit-Yan Yeung, "Collaborative Recurrent Autoencoder: Recommend while Learning to Fill in the Blanks," In Proceedings of the Twenty-Ninth Annual Conference on Neural Information Processing Systems, Barcelona, Spain, December 5-10, 2016.
[8] Badrul Sarwar, Geoge Karypis, Joseph Konstan, and John Riedl. Item-based Collaborative Filtering Recommendation Algorithms, In Proceedings of 10th ACM International Conference on World Wide Web, Hong Kong, Hong Kong, May 01-05, 2001. pp. 285-295.
[9] S. Sedhain, A. K. Menon, S. Sanner, and L. Xie. Autorec: Autoencoders meet collaborative filtering, In Proceedings of the 24th ACM International Conference on World Wide Web, Florence, Italy, May 18 - 22, 2015. pp. 111-112.
[10] Y. Wu, C. DuBois, A. X. Zheng, M. Ester. Collaborative Denoising Auto-Encoders for Top-N Recommender Systems, In Proceedings of the Ninth ACM International Conference on Web Search and Data Mining, San Francisco, California, USA, February 22 - 25, 2016. pp. 153-162.
[11] Yosuke Suzuki, Tomonobu Ozaki. "Stacked Denoising Autoencoder-based Deep Collaborative Filtering Using the Change of Similarity," IEEE International Conference on Advanced Information Networking and Applications Workshops, Taipei, Taiwan, March 27 - 29, 2017. pp. 498-502.
[12] B. Hidasi, A. Karatzoglou, L. Baltrunas and D. Tikk. "Session-Based Recommendations with Recurrent Neural Networks," In Proceedings of Annual Internaltional Conference on Learning Representations, San Juan, Puerto Rico, November 12 - 19, 2015.
[13] C. Y. Wu, A. Ahmed, A. Beutel, A. J. Smola, and H. Jing. Recurrent Recommender Networks, In Proceedings of the Tenth ACM International Conference on Web Search and Data Mining, Cambridge, United Kindom, February 06 - 10, 2017. pp. 495-503.
[14] Z. Chen, J. Lee, T. Q. S. Quek and M. Kountouris, "Cooperative Caching and Transmission Design in Cluster-Centric Small Cell Networks," IEEE Transaction on Wireless Communications, Vol. 16, No. 5, May 2017. pp. 3401-3415.
[15] Y. T. Chen, C. C. Yen, Y. T. Lin and J. S. Wang. "Cooperative Caching Plan of Popular Videos for Mobile Users by Grouping Preferences," In Proceedings of IEEE International Conference on Pervasive Intelligence and Computing, Athens, Greece, August 12-15, 2018.
[16] P. Y. Tsai. Prediction-Based Caching of Popular Videos in Cluster-Centric Small Cell Networks, Master Thesis, National Tsing Hua University, Hsinchu, Taiwan, ROC, July 2017.
[17] Y. Koren. Factorization meets the neighborhood: a multifaceted collaborative filtering model, In Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Las Vegas, USA, August 24-27, 2008. pp. 426-434.
[18] X. Ning and G. Karypis. "SLIM: Sparse linear methods for top-N recommender systems," In Proceedings of the 11th IEEE International Conference on Data Mining, Vancouver,BC, Canada, December 11-14, 2011.
[19] D. Arthur and S. Vassilvitskii. "k-means++: The Advantages of Careful Seeding," In SODA '07: Proceedings of the 18th Annual ACM-SIAM Symposium on Discrete Algorithms, 2007.
[20] N. Cesa-Bianchi and G. Lugosi, Prediction, learning, and games, Cambridge University Press, 2006.
[21] N. Ben Hassine, D. Marinca, P. Minet and D. Barth, "Expert-based on-line learning and prediction in Content Delivery Networks," In 2016 International Wireless Communications and Mobile Computing Conference (IWCMC), 2016, pp. 182-18227.