藉由單類矩陣分解進行搜尋推薦

Trending Query Recommendation by One-class Matrix Factorization

蘇傳堯(Chuan-Yao Su)

指導教授 ： 林智仁

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2018年)

https://doi.org/10.6342/NTU201801887

摘要

目前，對於有使用隱式用戶反饋的推薦系統，已經考慮了單類矩陣分解，但是，大多數現有工作都集中在方法論上。他們對一些公共或甚至人工生成的數據進行評估，而不是將他們的方法部署到大型推薦系統。因此，沒有討論到許多實際因素。在本文中，我們目標在通過提供在大規模趨勢查詢推薦服務上，應用一類矩陣分解的研究來填補這一空白。同時，我們也證明了此方法在推薦系統上有15％以上的改進。在方法論方面，基於實際數據，我們指出了過去工作中未涉及的一些計算瓶頸，並提供了有效率的訓練方法。

關鍵字

推薦系統 ； 矩陣分解 ； 梯度下降法 ； 機器學習 ； 隱性回饋 ； 正例與未標注學習

並列摘要

Recently, one-class matrix factorization has been considered for recommendation systems that have only implicit user feedbacks. However, most of existing works focus on the methodology. They conduct evaluations on some public or even artificially generated data, rather than deploying their approaches to a large production system. Therefore, many practical considerations are not discussed. In this thesis, we aim to fill the gap by providing an end-to-end study of applying one-class matrix factorization on a large-scale service of trending query recommendation. We discuss some practical challenges and demonstrate a more than 20\% improvement in our online production system. On the methodology side, based on properties of real data, we point out some computational bottlenecks not addressed in past works and provide efficient training procedures.

並列關鍵字

Recommender system ； Matrix factorization ； Coordinate descent ； Machine learning ； Implicit feedback ； PU learning

參考文獻

Reference

1 R. Baeza-Yates, C. Hurtado, and M. Mendoza. Query recommendation using query logs in search engines. In International Conference on Extending Database Technology, pages 588–596, 2004.

2 W.-S. Chin, Y. Zhuang, Y.-C. Juan, and C.-J. Lin. A fast parallel stochastic gradient method for matrix factorization in shared memory systems. ACM Transactions on Intelligent Systems and Technology, 6:2:1–2:24, 2015. URL http://www.csie.ntu. edu.tw/˜cjlin/papers/libmf/libmf_journal.pdf.

3 W.-S. Chin, B.-W. Yuan, M.-Y. Yang, Y. Zhuang, Y.-C. Juan, and C.-J. Lin. LIBMF: A library for parallel matrix factorization in shared-memory systems. Journal of Machine Learning Research, 17(86):1–5, 2016. URL https://www.csie.ntu.edu.tw/ ˜cjlin/papers/libmf/libmf_open_source.pdf.

4 R. Devooght, N. Kourtellis, and A. Mantrach. Dynamic matrix factorization with priors on unknown values. In Proceedings of the 21th ACM SIGKDD international conference on knowledge discovery and data mining (KDD), pages 189–198, 2015.