基於偏態排序最佳化探討圖形學習表示法之分佈於推薦系統

Exploring Distribution of Graph Embedding Based on Skewness Ranking Optimization for Recommender Systems

莊喻能(Chuang Yu-Neng)

指導教授 ： 蔡銘峰

國立政治大學/理學院/資訊科學系/碩士(2020年)

https://doi.org/10.6814/NCCU202000474

摘要

近年來大數據以及機器學習技術的蓬勃發展，推薦系統被廣泛應用於各種資訊系( Information Systems )上。如何有效地利用這些巨量的資料增進推薦系統效能，成為具有挑戰的工作。圖形學習表示法( Graph Embedding )便是一種特徵提取 ( Feature Extraction ) 的技術，此方法目的在於如何有效的將不同節點以及節點間的關係投射到低維度向量空間並賦予特徵向量。因此，如何有效率且精準的描述這些向量空間的概念，也被加入到圖形學習表示法的領域。本論文基於非對稱常態分佈( Skew Normal Distribution)之特性，提出以機率分佈重新檢視表示法向量空間，並針對使用者與喜好物品在非對稱常態分佈上會趨向正向偏態( Positive Skewness )的特性，將偏態之概念加入目標函式中進行優化。特別的是，本論文所提出之偏態項優化式為一通用優化項，能適用於過去各種 State of The Art 推薦演算法上，進而重塑各種推薦演算法所構建之向量空間。從理論面來論述，我們證明了如何在優化各種推薦演算法上之餘，同時優化基於非對稱常態分佈之 Shape 參數，此參數與分佈之偏態值為正相關。此外，針對所提出之演算法能同時最大化接收者操作特徵曲線( Receiver Operating Characteristic Curve ( ROC Curve ) )之論述，我們也提出一數學證述來解釋與分析。在數據實驗上，本文以將此偏態優化項主要實驗於矩陣分解類之推薦算法上，且為了展示方法的一致性，我們也將此偏態優化項實驗在基於圖形學習表示法的推薦演算法上，來做驗證本方法的可行性與正確性。而為了驗證此方法，本文實驗於五種不同的真實世 界巨量資料上，並且針對兩種常見的推薦任務: Top-N 推薦任務以及 Query-based 推薦任務上皆有所比較與操作。最後，在實驗結果的部分，結果呈現出我們所提之演算法與過去各種 State of The Art 之推薦演算法中實際比較後皆取得更優的表現。

關鍵字

推薦系統 ； 協同過濾法 ； 圖形學習表示法 ； 矩陣分解 ； 偏態排序法

並列摘要

In recent years, machine learning technology has drastically improved in adapting big data among various fields, including commercial streaming online service and recommendation systems. Especially in recommendation systems case, the user-based recommendation systems or personalized recommendation is one of the most challenging tasks. In this paper, hence, we propose a novel optimization criterion that leverages features of the skew normal distribution to better model the problem of personalized recommendation. Specifically, the developed criterion borrows the concept and the flexibility of the skew normal distribution and also based on three hyper-parameters to not only provide the degree of freedom in optimization and also highly attached to the optimization criterion. Moreover, we both provide the relation of optimization of the proposed criterion and the shape parameter in the skew normal distribution from theoretical point of view and provide the analogies and provide the theoretical proof on asymptotic analysis of the area under the ROC curve to our proposed method. Experimental results conducted on five large-scale real-world datasets reveal that our proposed optimization criterion significantly achieve the best performance of the state of the art and yields consistently on all tested datasets.

並列關鍵字

Recommender systems ； Collaborative filitering ； Graph embedding ； Matrix factorization ； Skewness optimization ranking

參考文獻

[1] C.-M. Chen, M.-F. Tsai, Y.-C. Lin, and Y.-H. Yang. Query-based music recommendations via preference embedding. In Proceedings of the 10th ACM Conference on Recommender Systems, RecSys ’16, page 79–82. Association for Computing Machinery.

[2] C.-M. Chen, C.-J. Wang, M.-F. Tsai, and Y.-H. Yang. Collaborative similarity em- bedding for recommender systems. In Proceedings of the 28th International Conference on World Wide Web, WWW ’19, page 2637–2643. Association for Computing Machinery.

[3] A. Grover and J. Leskovec. Node2vec: Scalable feature learning for networks. In Proceedings of the 22th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD ’16, pages 855–864. Association for Computing Machinery, 2016.

[4] M. A. Hasnat, J. Bohne, S. G. J. Milgram, and L. Chen. von mises-fisher mixture model-based deep learning: Application to face verification. In arXiv preprint arXiv:1706.04264, 2017.

[5] R. He, W.-C. Kang, and J. McAuley. Translation-based recommendation. In Proceedings of the 11th ACM Conference on Recommender Systems, RecSys ’17, page 161–169. Association for Computing Machinery.